From d79eda9dac9c9794091b6ea3544ab2bf69646db6 Mon Sep 17 00:00:00 2001
From: Alex Rubinsteyn <alex.rubinsteyn@gmail.com>
Date: Tue, 6 Oct 2026 12:54:27 -0400
Subject: [PATCH 1/3] Optimize attribute parsing and bound Arrow assembly
 memory (3.0.1)

---
 benchmarks/README.md                 |    7 +-
 benchmarks/compare_readers.py        |    2 +
 benchmarks/compare_releases.py       |  139 +
 benchmarks/optimization-results.json | 7145 ++++++++++++++++++++++++++
 benchmarks/optimization.md           |  147 +
 gtfparse/attribute_parsing.py        |   87 +-
 gtfparse/read_gtf.py                 |   67 +-
 tasks/todo.md                        |  106 +
 tests/test_attribute_optimization.py |  128 +
 9 files changed, 7771 insertions(+), 57 deletions(-)
 create mode 100644 benchmarks/compare_releases.py
 create mode 100644 benchmarks/optimization-results.json
 create mode 100644 benchmarks/optimization.md
 create mode 100644 tests/test_attribute_optimization.py

diff --git a/benchmarks/README.md b/benchmarks/README.md
index 06379db..87711b7 100644
--- a/benchmarks/README.md
+++ b/benchmarks/README.md
@@ -1,9 +1,12 @@
 # GTF reader backend comparison
 
-Version 3 uses PyArrow for fixed-column parsing and pandas for attribute
+The 3.0.0 migration uses PyArrow for fixed-column parsing and pandas for attribute
 processing and writing. Polars is an optional output adapter. This directory
 contains the measurement harness and the evidence for that decision.
 
+The follow-up [3.0.1 optimization report](optimization.md) compares the optimized
+complete reader with both the 2.9.1 Polars and 3.0.0 Arrow/pandas releases.
+
 ## Reproduction
 
 The recorded runs use Python 3.12.6, NumPy 2.5.3, PyArrow 25.0.1, Polars 1.44.2,
@@ -54,7 +57,7 @@ include raw attributes, filtering, selection, aliases, version casts, biotype
 inference and converters; separate regressions cover streams, nullable fields,
 comments, malformed inputs and progress/cancellation semantics.
 
-## Production results
+## 3.0.0 migration results
 
 The table shows three-run medians for equivalent pandas output. Complete load
 time includes fixed-column reading, attribute expansion, assembly and transforms.
diff --git a/benchmarks/compare_readers.py b/benchmarks/compare_readers.py
index d41778e..04db5e1 100644
--- a/benchmarks/compare_readers.py
+++ b/benchmarks/compare_readers.py
@@ -17,6 +17,7 @@
 from pathlib import Path
 from time import perf_counter, process_time
 
+import numpy as np
 import pandas as pd
 import pyarrow as pa
 import pyarrow.compute as pc
@@ -257,6 +258,7 @@ def main():
         "versions": {
             "gtfparse": __version__,
             "pandas": pd.__version__,
+            "numpy": np.__version__,
             "polars": pl.__version__ if pl is not None else None,
             "pyarrow": pa.__version__,
             "python": platform.python_version(),
diff --git a/benchmarks/compare_releases.py b/benchmarks/compare_releases.py
new file mode 100644
index 0000000..c610a48
--- /dev/null
+++ b/benchmarks/compare_releases.py
@@ -0,0 +1,139 @@
+"""Repeat complete production reads sequentially in fresh processes.
+
+Pass --baseline LABEL ROOT for each older checkout and --python PATH for each
+interpreter. See optimization.md for the measured corpus and reproduction.
+"""
+
+import argparse
+import hashlib
+import json
+import os
+import statistics
+import subprocess
+import sys
+from datetime import datetime, timezone
+from pathlib import Path
+
+
+def summarize(records):
+    groups = {}
+    for record in records:
+        key = (
+            record["versions"]["pandas"],
+            record["path"],
+            record["release"],
+            tuple(sorted(record["versions"].items())),
+        )
+        groups.setdefault(key, []).append(record)
+    summaries = []
+    for (pandas, path, release, _versions), runs in groups.items():
+        summary = {"pandas": pandas, "input": path, "release": release, "runs": len(runs)}
+        summary["versions"] = runs[0]["versions"]
+        for name in ("wall_seconds", "cpu_seconds", "peak_rss_mib"):
+            values = [run[name] for run in runs]
+            summary[name] = {
+                "median": statistics.median(values),
+                "min": min(values),
+                "max": max(values),
+            }
+        summary["stage_medians"] = {
+            name: statistics.median(run["stages"][name] for run in runs)
+            for name in runs[0]["stages"]
+        }
+        summary["checksum"] = runs[0]["checksum"]
+        summaries.append(summary)
+    return summaries
+
+
+def main():
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("inputs", nargs="+", type=Path)
+    parser.add_argument(
+        "--baseline", nargs=2, action="append", default=[], metavar=("LABEL", "ROOT")
+    )
+    parser.add_argument("--python", action="append", dest="interpreters")
+    parser.add_argument("--repeats", type=int, default=3)
+    parser.add_argument("--threads", type=int, default=4)
+    parser.add_argument("--result-type", choices=["pandas", "polars"], default="pandas")
+    parser.add_argument("--features", nargs="+")
+    parser.add_argument("--usecols", nargs="+")
+    parser.add_argument("--output", required=True, type=Path)
+    args = parser.parse_args()
+    if args.repeats < 1 or args.threads < 1:
+        parser.error("repeats and threads must be positive")
+    if len({path.name for path in args.inputs}) != len(args.inputs):
+        parser.error("input basenames must be distinct")
+    script = Path(__file__).with_name("compare_readers.py")
+    releases = [*args.baseline, ("optimized", str(script.parent.parent))]
+    expected = {}
+    records = []
+    environment = dict(os.environ, POLARS_MAX_THREADS=str(args.threads))
+    with args.output.with_suffix(".jsonl").open("w") as raw:
+        for interpreter in args.interpreters or [sys.executable]:
+            for repeat in range(1, args.repeats + 1):
+                # Rotate reader order and reverse corpus order to distribute ordering effects.
+                offset = (repeat - 1) % len(releases)
+                order = releases[offset:] + releases[:offset]
+                paths = args.inputs if repeat % 2 else args.inputs[::-1]
+                for path in paths:
+                    for label, root in order:
+                        command = [
+                            interpreter,
+                            str(script),
+                            str(path),
+                            "--backend",
+                            "production",
+                            "--library-root",
+                            root,
+                            "--threads",
+                            str(args.threads),
+                            "--result-type",
+                            args.result_type,
+                        ]
+                        for name in ("features", "usecols"):
+                            value = getattr(args, name)
+                            if value:
+                                command.extend(["--" + name, *value])
+                        run = subprocess.run(
+                            command, env=environment, capture_output=True, text=True, check=True
+                        )
+                        record = json.loads(run.stdout)
+                        key = path.name
+                        content = record["checksum"], record["rows"], record["columns"]
+                        if key in expected and expected[key] != content:
+                            raise RuntimeError(
+                                f"Reader results differ: {path}, {label}, {interpreter}"
+                            )
+                        expected[key] = content
+                        record.update(path=path.name, release=label, repeat=repeat)
+                        records.append(record)
+                        raw.write(json.dumps(record, sort_keys=True) + "\n")
+                        raw.flush()
+                        print(
+                            record["versions"]["pandas"],
+                            repeat,
+                            path.name,
+                            label,
+                            round(record["wall_seconds"], 3),
+                            round(record["peak_rss_mib"]),
+                            flush=True,
+                        )
+    result = {
+        "description": "Complete production readers; imports/checksums outside wall time; sequential fresh processes",
+        "date": datetime.now(timezone.utc).isoformat(),
+        "threads": args.threads,
+        "inputs": {
+            path.name: {
+                "bytes": path.stat().st_size,
+                "sha256": hashlib.sha256(path.read_bytes()).hexdigest(),
+            }
+            for path in args.inputs
+        },
+        "summaries": summarize(records),
+        "measurements": records,
+    }
+    args.output.write_text(json.dumps(result, indent=2) + "\n")
+
+
+if __name__ == "__main__":
+    main()
diff --git a/benchmarks/optimization-results.json b/benchmarks/optimization-results.json
new file mode 100644
index 0000000..2e04d7c
--- /dev/null
+++ b/benchmarks/optimization-results.json
@@ -0,0 +1,7145 @@
+{
+  "description": "Final 3.0.1 captured-pair parser and 250000-row batched Arrow attribute assembly; complete production pipelines.",
+  "date": "2026-10-06",
+  "baseline_sources": {
+    "2.9.1-polars": "3e0912c",
+    "3.0.0-arrow": "be449cbed70be8f515bb92b3eb2f43e55276917c",
+    "optimized": "perf/attribute-expansion PR against be449cb; version 3.0.1"
+  },
+  "method": {
+    "threads": 4,
+    "final_measurements": 84,
+    "subset_repeats": 3,
+    "other_repeats": 1,
+    "order": "sequential fresh processes; rotating reader order and alternating forward/reversed corpus order",
+    "timing": "complete read_gtf, including expansion, assembly and transforms; imports/checksums excluded",
+    "memory": "process peak RSS including imports, sampled before hashing; shared workstation, no OS cache flushing",
+    "equivalence": "all column values, row count and column order; normalized dtype representations for checksumming; exact dtypes checked separately against 3.0.0"
+  },
+  "inputs": {
+    "ensembl-250k.gtf": {
+      "bytes": 111900013,
+      "sha256": "ac84dc46b89df30eea025c5a36ea3c452f6546be037f77d1eb68fa813226958e",
+      "rows": 250000,
+      "source_url": "https://ftp.ensembl.org/pub/release-114/gtf/homo_sapiens/Homo_sapiens.GRCh38.114.gtf.gz",
+      "selection": "first 250000 annotation rows plus original header"
+    },
+    "ensembl-250k.gtf.gz": {
+      "bytes": 3524624,
+      "sha256": "3000128cb22ea68754c01bd3d72fe347c9c86d05b9cae9fed65d1440c57291cd",
+      "rows": 250000,
+      "source_url": "https://ftp.ensembl.org/pub/release-114/gtf/homo_sapiens/Homo_sapiens.GRCh38.114.gtf.gz",
+      "selection": "first 250000 annotation rows plus original header"
+    },
+    "gencode-250k.gtf": {
+      "bytes": 109586053,
+      "sha256": "04e929726d7c7837c7669b3795883b55537d389274ba6661a8ab56b5df06bc23",
+      "rows": 250000,
+      "source_url": "https://ftp.ebi.ac.uk/pub/databases/gencode/Gencode_human/release_48/gencode.v48.primary_assembly.annotation.gtf.gz",
+      "selection": "first 250000 annotation rows plus original header"
+    },
+    "gencode-250k.gtf.gz": {
+      "bytes": 3520115,
+      "sha256": "6d560c4098d78e6beb8616f08e2bdbbf3071a026edca622e505e294be7f5011d",
+      "rows": 250000,
+      "source_url": "https://ftp.ebi.ac.uk/pub/databases/gencode/Gencode_human/release_48/gencode.v48.primary_assembly.annotation.gtf.gz",
+      "selection": "first 250000 annotation rows plus original header"
+    },
+    "Homo_sapiens.GRCh38.114.gtf.gz": {
+      "bytes": 64228966,
+      "sha256": "75ec5b7a1bbb8ce566622dd9abf974ad37d390130325dae6d197d1a590854c50",
+      "rows": 4116048,
+      "source_url": "https://ftp.ensembl.org/pub/release-114/gtf/homo_sapiens/Homo_sapiens.GRCh38.114.gtf.gz",
+      "selection": "full file"
+    },
+    "gencode.v48.primary_assembly.annotation.gtf.gz": {
+      "bytes": 59094311,
+      "sha256": "ad15e956fb7643172afca6dcaae176cc76400532392978a191c981d0a052e668",
+      "rows": 4119244,
+      "source_url": "https://ftp.ebi.ac.uk/pub/databases/gencode/Gencode_human/release_48/gencode.v48.primary_assembly.annotation.gtf.gz",
+      "selection": "full file"
+    }
+  },
+  "final_sweeps": {
+    "subsets": {
+      "description": "Complete production readers; imports/checksums outside wall time; sequential fresh processes",
+      "date": "2026-10-06T16:44:26.773200+00:00",
+      "threads": 4,
+      "inputs": {
+        "ensembl-250k.gtf": {
+          "bytes": 111900013,
+          "sha256": "ac84dc46b89df30eea025c5a36ea3c452f6546be037f77d1eb68fa813226958e",
+          "rows": 250000,
+          "source_url": "https://ftp.ensembl.org/pub/release-114/gtf/homo_sapiens/Homo_sapiens.GRCh38.114.gtf.gz",
+          "selection": "first 250000 annotation rows plus original header"
+        },
+        "ensembl-250k.gtf.gz": {
+          "bytes": 3524624,
+          "sha256": "3000128cb22ea68754c01bd3d72fe347c9c86d05b9cae9fed65d1440c57291cd",
+          "rows": 250000,
+          "source_url": "https://ftp.ensembl.org/pub/release-114/gtf/homo_sapiens/Homo_sapiens.GRCh38.114.gtf.gz",
+          "selection": "first 250000 annotation rows plus original header"
+        },
+        "gencode-250k.gtf": {
+          "bytes": 109586053,
+          "sha256": "04e929726d7c7837c7669b3795883b55537d389274ba6661a8ab56b5df06bc23",
+          "rows": 250000,
+          "source_url": "https://ftp.ebi.ac.uk/pub/databases/gencode/Gencode_human/release_48/gencode.v48.primary_assembly.annotation.gtf.gz",
+          "selection": "first 250000 annotation rows plus original header"
+        },
+        "gencode-250k.gtf.gz": {
+          "bytes": 3520115,
+          "sha256": "6d560c4098d78e6beb8616f08e2bdbbf3071a026edca622e505e294be7f5011d",
+          "rows": 250000,
+          "source_url": "https://ftp.ebi.ac.uk/pub/databases/gencode/Gencode_human/release_48/gencode.v48.primary_assembly.annotation.gtf.gz",
+          "selection": "first 250000 annotation rows plus original header"
+        }
+      },
+      "summaries": [
+        {
+          "pandas": "2.3.3",
+          "input": "ensembl-250k.gtf",
+          "release": "2.9.1-polars",
+          "runs": 3,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": {
+            "median": 2.5705371249932796,
+            "min": 2.545209417003207,
+            "max": 2.871901207996416
+          },
+          "cpu_seconds": {
+            "median": 2.640136,
+            "min": 2.633258,
+            "max": 2.717941
+          },
+          "peak_rss_mib": {
+            "median": 560.328125,
+            "min": 560.3125,
+            "max": 560.671875
+          },
+          "stage_medians": {
+            "attributes": 1.8935110830061603,
+            "convert": 0.34992329200031236,
+            "other": 0.23882933298591524,
+            "read": 0.08401687500008848
+          },
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8"
+        },
+        {
+          "pandas": "2.3.3",
+          "input": "ensembl-250k.gtf",
+          "release": "3.0.0-arrow",
+          "runs": 3,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": {
+            "median": 2.9074682079954073,
+            "min": 2.8575992499972926,
+            "max": 3.467279165997752
+          },
+          "cpu_seconds": {
+            "median": 2.986738,
+            "min": 2.943254,
+            "max": 3.3548989999999996
+          },
+          "peak_rss_mib": {
+            "median": 616.6875,
+            "min": 615.40625,
+            "max": 627.53125
+          },
+          "stage_medians": {
+            "attributes": 1.87293291700189,
+            "convert": 0.28285558299103286,
+            "other": 0.4966888319904683,
+            "read": 0.2723564589978196
+          },
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8"
+        },
+        {
+          "pandas": "2.3.3",
+          "input": "ensembl-250k.gtf",
+          "release": "optimized",
+          "runs": 3,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": {
+            "median": 2.2852479170105653,
+            "min": 2.253485958004603,
+            "max": 2.4167407499917317
+          },
+          "cpu_seconds": {
+            "median": 2.373172,
+            "min": 2.334993,
+            "max": 2.539217
+          },
+          "peak_rss_mib": {
+            "median": 583.40625,
+            "min": 576.484375,
+            "max": 593.40625
+          },
+          "stage_medians": {
+            "attributes": 1.2345331250107847,
+            "convert": 0.2886152910068631,
+            "other": 0.502303834000486,
+            "read": 0.2740694169915514
+          },
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8"
+        },
+        {
+          "pandas": "2.3.3",
+          "input": "ensembl-250k.gtf.gz",
+          "release": "2.9.1-polars",
+          "runs": 3,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": {
+            "median": 2.8604007920075674,
+            "min": 2.602536874997895,
+            "max": 3.8669654580007773
+          },
+          "cpu_seconds": {
+            "median": 2.8331690000000003,
+            "min": 2.6671,
+            "max": 3.038677
+          },
+          "peak_rss_mib": {
+            "median": 561.046875,
+            "min": 524.484375,
+            "max": 561.09375
+          },
+          "stage_medians": {
+            "attributes": 1.9897701250010869,
+            "convert": 0.34451254199666437,
+            "other": 0.2918442920054076,
+            "read": 0.21517366600164678
+          },
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8"
+        },
+        {
+          "pandas": "2.3.3",
+          "input": "ensembl-250k.gtf.gz",
+          "release": "3.0.0-arrow",
+          "runs": 3,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": {
+            "median": 4.57787587500934,
+            "min": 3.4941771660087397,
+            "max": 5.500599250008236
+          },
+          "cpu_seconds": {
+            "median": 3.36688,
+            "min": 3.302426,
+            "max": 3.37763
+          },
+          "peak_rss_mib": {
+            "median": 619.71875,
+            "min": 615.78125,
+            "max": 625.609375
+          },
+          "stage_medians": {
+            "attributes": 2.818784125003731,
+            "convert": 0.4906274579989258,
+            "other": 0.9829520420025801,
+            "read": 0.2925547919876408
+          },
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8"
+        },
+        {
+          "pandas": "2.3.3",
+          "input": "ensembl-250k.gtf.gz",
+          "release": "optimized",
+          "runs": 3,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": {
+            "median": 2.361613708999357,
+            "min": 2.2743365000060294,
+            "max": 3.9647754170000553
+          },
+          "cpu_seconds": {
+            "median": 2.433655,
+            "min": 2.355343,
+            "max": 2.82855
+          },
+          "peak_rss_mib": {
+            "median": 583.96875,
+            "min": 574.75,
+            "max": 591.171875
+          },
+          "stage_medians": {
+            "attributes": 1.2559461250057211,
+            "convert": 0.2896352910029236,
+            "other": 0.535507876003976,
+            "read": 0.28052441698673647
+          },
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8"
+        },
+        {
+          "pandas": "2.3.3",
+          "input": "gencode-250k.gtf",
+          "release": "2.9.1-polars",
+          "runs": 3,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": {
+            "median": 3.1804067909979494,
+            "min": 2.3660601659939857,
+            "max": 5.55182199999399
+          },
+          "cpu_seconds": {
+            "median": 2.5377099999999997,
+            "min": 2.266625,
+            "max": 2.560311
+          },
+          "peak_rss_mib": {
+            "median": 567.3125,
+            "min": 566.484375,
+            "max": 567.546875
+          },
+          "stage_medians": {
+            "attributes": 2.646453332999954,
+            "convert": 0.09764800000994,
+            "other": 0.3458024169958662,
+            "read": 0.1014160000049742
+          },
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641"
+        },
+        {
+          "pandas": "2.3.3",
+          "input": "gencode-250k.gtf",
+          "release": "3.0.0-arrow",
+          "runs": 3,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": {
+            "median": 2.585112458007643,
+            "min": 2.5604790829966078,
+            "max": 5.565355083002942
+          },
+          "cpu_seconds": {
+            "median": 2.658798,
+            "min": 2.651787,
+            "max": 3.191441
+          },
+          "peak_rss_mib": {
+            "median": 587.6875,
+            "min": 551.328125,
+            "max": 591.8125
+          },
+          "stage_medians": {
+            "attributes": 1.7814520410029218,
+            "convert": 7.35829962650314e-05,
+            "other": 0.5366334179998375,
+            "read": 0.26695341600861866
+          },
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641"
+        },
+        {
+          "pandas": "2.3.3",
+          "input": "gencode-250k.gtf",
+          "release": "optimized",
+          "runs": 3,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": {
+            "median": 3.716955208001309,
+            "min": 2.337004041997716,
+            "max": 4.0017947089945665
+          },
+          "cpu_seconds": {
+            "median": 2.425999,
+            "min": 2.2283709999999997,
+            "max": 2.4871540000000003
+          },
+          "peak_rss_mib": {
+            "median": 549.84375,
+            "min": 527.203125,
+            "max": 565.484375
+          },
+          "stage_medians": {
+            "attributes": 2.1337991250038613,
+            "convert": 5.26670046383515e-05,
+            "other": 0.8941883750085253,
+            "read": 0.4248130839987425
+          },
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641"
+        },
+        {
+          "pandas": "2.3.3",
+          "input": "gencode-250k.gtf.gz",
+          "release": "2.9.1-polars",
+          "runs": 3,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": {
+            "median": 3.60983508299978,
+            "min": 2.207000333000906,
+            "max": 3.854776332998881
+          },
+          "cpu_seconds": {
+            "median": 2.671091,
+            "min": 2.2800469999999997,
+            "max": 2.790294
+          },
+          "peak_rss_mib": {
+            "median": 567.796875,
+            "min": 567.5,
+            "max": 568.4375
+          },
+          "stage_medians": {
+            "attributes": 2.8298818330076756,
+            "convert": 0.131129208006314,
+            "other": 0.4548585419834126,
+            "read": 0.19396550000237767
+          },
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641"
+        },
+        {
+          "pandas": "2.3.3",
+          "input": "gencode-250k.gtf.gz",
+          "release": "3.0.0-arrow",
+          "runs": 3,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": {
+            "median": 3.8968885840004077,
+            "min": 2.4905160420021275,
+            "max": 3.9853524999925867
+          },
+          "cpu_seconds": {
+            "median": 3.075842,
+            "min": 2.574939,
+            "max": 3.087378
+          },
+          "peak_rss_mib": {
+            "median": 583.796875,
+            "min": 580.96875,
+            "max": 587.78125
+          },
+          "stage_medians": {
+            "attributes": 2.6965094580082223,
+            "convert": 8.433299080934376e-05,
+            "other": 0.7963529169937829,
+            "read": 0.3873044999927515
+          },
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641"
+        },
+        {
+          "pandas": "2.3.3",
+          "input": "gencode-250k.gtf.gz",
+          "release": "optimized",
+          "runs": 3,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": {
+            "median": 3.736287124993396,
+            "min": 1.9886879580008099,
+            "max": 3.8399662080046255
+          },
+          "cpu_seconds": {
+            "median": 2.54815,
+            "min": 2.080072,
+            "max": 2.6218580000000005
+          },
+          "peak_rss_mib": {
+            "median": 557.890625,
+            "min": 495.1875,
+            "max": 566.9375
+          },
+          "stage_medians": {
+            "attributes": 2.2640662500052713,
+            "convert": 4.354199336376041e-05,
+            "other": 0.9998161669936962,
+            "read": 0.39063262499985285
+          },
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641"
+        },
+        {
+          "pandas": "3.0.6",
+          "input": "ensembl-250k.gtf",
+          "release": "2.9.1-polars",
+          "runs": 3,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": {
+            "median": 2.5418692079983884,
+            "min": 2.3415217919973657,
+            "max": 2.7177458749938523
+          },
+          "cpu_seconds": {
+            "median": 2.420429,
+            "min": 2.414792,
+            "max": 2.556296
+          },
+          "peak_rss_mib": {
+            "median": 552.5,
+            "min": 552.21875,
+            "max": 552.828125
+          },
+          "stage_medians": {
+            "attributes": 1.8557262500107754,
+            "convert": 0.22684966700035147,
+            "other": 0.23259929296909831,
+            "read": 0.03862558300897945
+          },
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8"
+        },
+        {
+          "pandas": "3.0.6",
+          "input": "ensembl-250k.gtf",
+          "release": "3.0.0-arrow",
+          "runs": 3,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": {
+            "median": 2.824426291990676,
+            "min": 2.7780951250024373,
+            "max": 2.897920041010366
+          },
+          "cpu_seconds": {
+            "median": 2.896116,
+            "min": 2.85385,
+            "max": 2.979457
+          },
+          "peak_rss_mib": {
+            "median": 599.765625,
+            "min": 590.28125,
+            "max": 601.796875
+          },
+          "stage_medians": {
+            "attributes": 1.8735739160038065,
+            "convert": 0.18741612500161864,
+            "other": 0.5475739179964876,
+            "read": 0.21546933299396187
+          },
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8"
+        },
+        {
+          "pandas": "3.0.6",
+          "input": "ensembl-250k.gtf",
+          "release": "optimized",
+          "runs": 3,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": {
+            "median": 1.9066367920022458,
+            "min": 1.8812497079925379,
+            "max": 2.1686059169878718
+          },
+          "cpu_seconds": {
+            "median": 1.9879889999999998,
+            "min": 1.981777,
+            "max": 2.111438
+          },
+          "peak_rss_mib": {
+            "median": 576.71875,
+            "min": 574.484375,
+            "max": 578.09375
+          },
+          "stage_medians": {
+            "attributes": 1.255898375005927,
+            "convert": 0.18790645799890626,
+            "other": 0.2516916669992497,
+            "read": 0.21316074999049306
+          },
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8"
+        },
+        {
+          "pandas": "3.0.6",
+          "input": "ensembl-250k.gtf.gz",
+          "release": "2.9.1-polars",
+          "runs": 3,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": {
+            "median": 2.3489610830001766,
+            "min": 2.341782832998433,
+            "max": 2.4118617909989553
+          },
+          "cpu_seconds": {
+            "median": 2.415416,
+            "min": 2.4144940000000004,
+            "max": 2.4895519999999998
+          },
+          "peak_rss_mib": {
+            "median": 553.15625,
+            "min": 553.046875,
+            "max": 553.421875
+          },
+          "stage_medians": {
+            "attributes": 1.8477172499988228,
+            "convert": 0.20435254200128838,
+            "other": 0.22192491599707864,
+            "read": 0.0779351250093896
+          },
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8"
+        },
+        {
+          "pandas": "3.0.6",
+          "input": "ensembl-250k.gtf.gz",
+          "release": "3.0.0-arrow",
+          "runs": 3,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": {
+            "median": 2.80403087499144,
+            "min": 2.7633427499968093,
+            "max": 2.8217112080019433
+          },
+          "cpu_seconds": {
+            "median": 2.888417,
+            "min": 2.854932,
+            "max": 2.9196619999999998
+          },
+          "peak_rss_mib": {
+            "median": 599.984375,
+            "min": 595.125,
+            "max": 607.015625
+          },
+          "stage_medians": {
+            "attributes": 1.8276292090013158,
+            "convert": 0.18400516700057779,
+            "other": 0.556629457991221,
+            "read": 0.23586362499918323
+          },
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8"
+        },
+        {
+          "pandas": "3.0.6",
+          "input": "ensembl-250k.gtf.gz",
+          "release": "optimized",
+          "runs": 3,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": {
+            "median": 1.876802041006158,
+            "min": 1.8565174999966985,
+            "max": 1.9292772500048159
+          },
+          "cpu_seconds": {
+            "median": 1.958045,
+            "min": 1.9496789999999997,
+            "max": 2.03234
+          },
+          "peak_rss_mib": {
+            "median": 574.140625,
+            "min": 572.890625,
+            "max": 574.84375
+          },
+          "stage_medians": {
+            "attributes": 1.2322392920032144,
+            "convert": 0.1897575000039069,
+            "other": 0.23446987399074715,
+            "read": 0.22112158400705084
+          },
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8"
+        },
+        {
+          "pandas": "3.0.6",
+          "input": "gencode-250k.gtf",
+          "release": "2.9.1-polars",
+          "runs": 3,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": {
+            "median": 2.047686125006294,
+            "min": 1.996198959008325,
+            "max": 2.0576451669912785
+          },
+          "cpu_seconds": {
+            "median": 2.116965,
+            "min": 2.082073,
+            "max": 2.142511
+          },
+          "peak_rss_mib": {
+            "median": 559.5,
+            "min": 559.421875,
+            "max": 559.671875
+          },
+          "stage_medians": {
+            "attributes": 1.7327500000101281,
+            "convert": 0.017430042003979906,
+            "other": 0.24500191600236576,
+            "read": 0.027468333006254397
+          },
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641"
+        },
+        {
+          "pandas": "3.0.6",
+          "input": "gencode-250k.gtf",
+          "release": "3.0.0-arrow",
+          "runs": 3,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": {
+            "median": 2.682168958010152,
+            "min": 2.5138637089985423,
+            "max": 2.747774916992057
+          },
+          "cpu_seconds": {
+            "median": 2.681481,
+            "min": 2.606457,
+            "max": 2.745542
+          },
+          "peak_rss_mib": {
+            "median": 593.15625,
+            "min": 591.375,
+            "max": 598.234375
+          },
+          "stage_medians": {
+            "attributes": 1.8231396670016693,
+            "convert": 0.00015558300947304815,
+            "other": 0.6528683740034467,
+            "read": 0.21073287499893922
+          },
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641"
+        },
+        {
+          "pandas": "3.0.6",
+          "input": "gencode-250k.gtf",
+          "release": "optimized",
+          "runs": 3,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": {
+            "median": 1.7089971670066006,
+            "min": 1.6545389160019113,
+            "max": 1.8188800839998294
+          },
+          "cpu_seconds": {
+            "median": 1.784423,
+            "min": 1.7458540000000002,
+            "max": 1.804453
+          },
+          "peak_rss_mib": {
+            "median": 576.578125,
+            "min": 576.234375,
+            "max": 578.703125
+          },
+          "stage_medians": {
+            "attributes": 1.216064499996719,
+            "convert": 2.5416011339984834e-05,
+            "other": 0.28129791701212525,
+            "read": 0.21160841699747834
+          },
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641"
+        },
+        {
+          "pandas": "3.0.6",
+          "input": "gencode-250k.gtf.gz",
+          "release": "2.9.1-polars",
+          "runs": 3,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": {
+            "median": 2.099596250001923,
+            "min": 2.0580523329990683,
+            "max": 2.2535872500011465
+          },
+          "cpu_seconds": {
+            "median": 2.1718260000000003,
+            "min": 2.126937,
+            "max": 2.24322
+          },
+          "peak_rss_mib": {
+            "median": 560.15625,
+            "min": 560.0,
+            "max": 560.78125
+          },
+          "stage_medians": {
+            "attributes": 1.7444164579937933,
+            "convert": 0.018364208008279093,
+            "other": 0.26727887500601355,
+            "read": 0.07006466700113378
+          },
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641"
+        },
+        {
+          "pandas": "3.0.6",
+          "input": "gencode-250k.gtf.gz",
+          "release": "3.0.0-arrow",
+          "runs": 3,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": {
+            "median": 2.524030959000811,
+            "min": 2.5057187499915017,
+            "max": 2.5275498749979306
+          },
+          "cpu_seconds": {
+            "median": 2.6059170000000003,
+            "min": 2.589577,
+            "max": 2.6144279999999998
+          },
+          "peak_rss_mib": {
+            "median": 595.03125,
+            "min": 589.8125,
+            "max": 595.265625
+          },
+          "stage_medians": {
+            "attributes": 1.7339677080017282,
+            "convert": 0.0001550410088384524,
+            "other": 0.5685903760022484,
+            "read": 0.2197710839973297
+          },
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641"
+        },
+        {
+          "pandas": "3.0.6",
+          "input": "gencode-250k.gtf.gz",
+          "release": "optimized",
+          "runs": 3,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": {
+            "median": 1.7293930830055615,
+            "min": 1.6913924160035094,
+            "max": 1.730756167002255
+          },
+          "cpu_seconds": {
+            "median": 1.789281,
+            "min": 1.7782019999999998,
+            "max": 1.827663
+          },
+          "peak_rss_mib": {
+            "median": 574.265625,
+            "min": 572.78125,
+            "max": 583.171875
+          },
+          "stage_medians": {
+            "attributes": 1.23295112499909,
+            "convert": 2.5416986318305135e-05,
+            "other": 0.268862833007006,
+            "read": 0.22319108400552068
+          },
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641"
+        }
+      ],
+      "measurements": [
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.717941,
+          "features": null,
+          "input_bytes": 111900013,
+          "path": "ensembl-250k.gtf",
+          "peak_rss_mib": 560.3125,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.8935110830061603,
+            "convert": 0.3650127500004601,
+            "other": 0.23882933298591524,
+            "read": 0.37454804200388025
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.871901207996416,
+          "release": "2.9.1-polars",
+          "repeat": 1
+        },
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.986738,
+          "features": null,
+          "input_bytes": 111900013,
+          "path": "ensembl-250k.gtf",
+          "peak_rss_mib": 615.40625,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.8604427090031095,
+            "convert": 0.2779802080040099,
+            "other": 0.4966888319904683,
+            "read": 0.2723564589978196
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.9074682079954073,
+          "release": "3.0.0-arrow",
+          "repeat": 1
+        },
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.334993,
+          "features": null,
+          "input_bytes": 111900013,
+          "path": "ensembl-250k.gtf",
+          "peak_rss_mib": 583.40625,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.2237206660065567,
+            "convert": 0.2886152910068631,
+            "other": 0.4670805839996319,
+            "read": 0.2740694169915514
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.253485958004603,
+          "release": "optimized",
+          "repeat": 1
+        },
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.8331690000000003,
+          "features": null,
+          "input_bytes": 3524624,
+          "path": "ensembl-250k.gtf.gz",
+          "peak_rss_mib": 524.484375,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.9897701250010869,
+            "convert": 0.34451254199666437,
+            "other": 0.2918442920054076,
+            "read": 0.23427383300440852
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.8604007920075674,
+          "release": "2.9.1-polars",
+          "repeat": 1
+        },
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 3.36688,
+          "features": null,
+          "input_bytes": 3524624,
+          "path": "ensembl-250k.gtf.gz",
+          "peak_rss_mib": 619.71875,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 2.199549124998157,
+            "convert": 0.32413274999998976,
+            "other": 0.6779404990229523,
+            "read": 0.2925547919876408
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 3.4941771660087397,
+          "release": "3.0.0-arrow",
+          "repeat": 1
+        },
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.433655,
+          "features": null,
+          "input_bytes": 3524624,
+          "path": "ensembl-250k.gtf.gz",
+          "peak_rss_mib": 583.96875,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.2559461250057211,
+            "convert": 0.2896352910029236,
+            "other": 0.535507876003976,
+            "read": 0.28052441698673647
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.361613708999357,
+          "release": "optimized",
+          "repeat": 1
+        },
+        {
+          "backend": "production",
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.266625,
+          "features": null,
+          "input_bytes": 109586053,
+          "path": "gencode-250k.gtf",
+          "peak_rss_mib": 567.3125,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.758518542003003,
+            "convert": 0.06522274999588262,
+            "other": 0.2602612069895258,
+            "read": 0.28205766700557433
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.3660601659939857,
+          "release": "2.9.1-polars",
+          "repeat": 1
+        },
+        {
+          "backend": "production",
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.651787,
+          "features": null,
+          "input_bytes": 109586053,
+          "path": "gencode-250k.gtf",
+          "peak_rss_mib": 591.8125,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.7611734999954933,
+            "convert": 6.516699795611203e-05,
+            "other": 0.5358809580066008,
+            "read": 0.26335945799655747
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.5604790829966078,
+          "release": "3.0.0-arrow",
+          "repeat": 1
+        },
+        {
+          "backend": "production",
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.2283709999999997,
+          "features": null,
+          "input_bytes": 109586053,
+          "path": "gencode-250k.gtf",
+          "peak_rss_mib": 549.84375,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.285506374988472,
+            "convert": 5.26670046383515e-05,
+            "other": 0.7597022910049418,
+            "read": 0.2917427089996636
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.337004041997716,
+          "release": "optimized",
+          "repeat": 1
+        },
+        {
+          "backend": "production",
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.671091,
+          "features": null,
+          "input_bytes": 3520115,
+          "path": "gencode-250k.gtf.gz",
+          "peak_rss_mib": 567.796875,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 2.8298818330076756,
+            "convert": 0.131129208006314,
+            "other": 0.4548585419834126,
+            "read": 0.19396550000237767
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 3.60983508299978,
+          "release": "2.9.1-polars",
+          "repeat": 1
+        },
+        {
+          "backend": "production",
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 3.075842,
+          "features": null,
+          "input_bytes": 3520115,
+          "path": "gencode-250k.gtf.gz",
+          "peak_rss_mib": 587.78125,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 2.6965094580082223,
+            "convert": 8.433299080934376e-05,
+            "other": 0.8129902930086246,
+            "read": 0.3873044999927515
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 3.8968885840004077,
+          "release": "3.0.0-arrow",
+          "repeat": 1
+        },
+        {
+          "backend": "production",
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.54815,
+          "features": null,
+          "input_bytes": 3520115,
+          "path": "gencode-250k.gtf.gz",
+          "peak_rss_mib": 566.9375,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 2.345767665989115,
+            "convert": 7.06670107319951e-05,
+            "other": 0.9998161669936962,
+            "read": 0.39063262499985285
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 3.736287124993396,
+          "release": "optimized",
+          "repeat": 1
+        },
+        {
+          "backend": "production",
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 3.087378,
+          "features": null,
+          "input_bytes": 3520115,
+          "path": "gencode-250k.gtf.gz",
+          "peak_rss_mib": 583.796875,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 2.7883104579959763,
+            "convert": 0.00013437500456348062,
+            "other": 0.7963529169937829,
+            "read": 0.400554749998264
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 3.9853524999925867,
+          "release": "3.0.0-arrow",
+          "repeat": 2
+        },
+        {
+          "backend": "production",
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.6218580000000005,
+          "features": null,
+          "input_bytes": 3520115,
+          "path": "gencode-250k.gtf.gz",
+          "peak_rss_mib": 495.1875,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 2.2640662500052713,
+            "convert": 4.354199336376041e-05,
+            "other": 1.0697560000116937,
+            "read": 0.5061004159942968
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 3.8399662080046255,
+          "release": "optimized",
+          "repeat": 2
+        },
+        {
+          "backend": "production",
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.790294,
+          "features": null,
+          "input_bytes": 3520115,
+          "path": "gencode-250k.gtf.gz",
+          "peak_rss_mib": 568.4375,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 2.8794367499940563,
+            "convert": 0.19175450000329874,
+            "other": 0.5437821660016198,
+            "read": 0.2398029169999063
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 3.854776332998881,
+          "release": "2.9.1-polars",
+          "repeat": 2
+        },
+        {
+          "backend": "production",
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 3.191441,
+          "features": null,
+          "input_bytes": 109586053,
+          "path": "gencode-250k.gtf",
+          "peak_rss_mib": 551.328125,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 3.7154556659952505,
+            "convert": 7.458300387952477e-05,
+            "other": 1.3441707089950796,
+            "read": 0.5056541250087321
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 5.565355083002942,
+          "release": "3.0.0-arrow",
+          "repeat": 2
+        },
+        {
+          "backend": "production",
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.425999,
+          "features": null,
+          "input_bytes": 109586053,
+          "path": "gencode-250k.gtf",
+          "peak_rss_mib": 565.484375,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 2.2242981659946963,
+            "convert": 7.416699372697622e-05,
+            "other": 0.8941883750085253,
+            "read": 0.5983945000043605
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 3.716955208001309,
+          "release": "optimized",
+          "repeat": 2
+        },
+        {
+          "backend": "production",
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.5377099999999997,
+          "features": null,
+          "input_bytes": 109586053,
+          "path": "gencode-250k.gtf",
+          "peak_rss_mib": 566.484375,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 4.694422458007466,
+            "convert": 0.10437379199720453,
+            "other": 0.6516097499843454,
+            "read": 0.1014160000049742
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 5.55182199999399,
+          "release": "2.9.1-polars",
+          "repeat": 2
+        },
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 3.37763,
+          "features": null,
+          "input_bytes": 3524624,
+          "path": "ensembl-250k.gtf.gz",
+          "peak_rss_mib": 625.609375,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 3.433137957996223,
+            "convert": 0.5808867920131888,
+            "other": 1.0155575410026358,
+            "read": 0.4710169589961879
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 5.500599250008236,
+          "release": "3.0.0-arrow",
+          "repeat": 2
+        },
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.82855,
+          "features": null,
+          "input_bytes": 3524624,
+          "path": "ensembl-250k.gtf.gz",
+          "peak_rss_mib": 591.171875,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.8764499579992844,
+            "convert": 0.557577957995818,
+            "other": 1.1187510839954484,
+            "read": 0.4119964170095045
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 3.9647754170000553,
+          "release": "optimized",
+          "repeat": 2
+        },
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 3.038677,
+          "features": null,
+          "input_bytes": 3524624,
+          "path": "ensembl-250k.gtf.gz",
+          "peak_rss_mib": 561.046875,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 2.7519126660045004,
+            "convert": 0.5172319999983301,
+            "other": 0.3826471259963,
+            "read": 0.21517366600164678
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 3.8669654580007773,
+          "release": "2.9.1-polars",
+          "repeat": 2
+        },
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 3.3548989999999996,
+          "features": null,
+          "input_bytes": 111900013,
+          "path": "ensembl-250k.gtf",
+          "peak_rss_mib": 627.53125,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 2.187712124999962,
+            "convert": 0.3620601250004256,
+            "other": 0.5586184159910772,
+            "read": 0.35888850000628736
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 3.467279165997752,
+          "release": "3.0.0-arrow",
+          "repeat": 2
+        },
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.539217,
+          "features": null,
+          "input_bytes": 111900013,
+          "path": "ensembl-250k.gtf",
+          "peak_rss_mib": 593.40625,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.255412290993263,
+            "convert": 0.29558341699885204,
+            "other": 0.5362654589989688,
+            "read": 0.32947958300064784
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.4167407499917317,
+          "release": "optimized",
+          "repeat": 2
+        },
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.633258,
+          "features": null,
+          "input_bytes": 111900013,
+          "path": "ensembl-250k.gtf",
+          "peak_rss_mib": 560.328125,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.868775209004525,
+            "convert": 0.3353991249896353,
+            "other": 0.25759920800919645,
+            "read": 0.08343587499985006
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.545209417003207,
+          "release": "2.9.1-polars",
+          "repeat": 2
+        },
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.373172,
+          "features": null,
+          "input_bytes": 111900013,
+          "path": "ensembl-250k.gtf",
+          "peak_rss_mib": 576.484375,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.2345331250107847,
+            "convert": 0.28705333299876656,
+            "other": 0.502303834000486,
+            "read": 0.26135762500052806
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.2852479170105653,
+          "release": "optimized",
+          "repeat": 3
+        },
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.640136,
+          "features": null,
+          "input_bytes": 111900013,
+          "path": "ensembl-250k.gtf",
+          "peak_rss_mib": 560.671875,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.898370958995656,
+            "convert": 0.34992329200031236,
+            "other": 0.23822599899722263,
+            "read": 0.08401687500008848
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.5705371249932796,
+          "release": "2.9.1-polars",
+          "repeat": 3
+        },
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.943254,
+          "features": null,
+          "input_bytes": 111900013,
+          "path": "ensembl-250k.gtf",
+          "peak_rss_mib": 616.6875,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.87293291700189,
+            "convert": 0.28285558299103286,
+            "other": 0.4352961250115186,
+            "read": 0.26651462499285117
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.8575992499972926,
+          "release": "3.0.0-arrow",
+          "repeat": 3
+        },
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.355343,
+          "features": null,
+          "input_bytes": 3524624,
+          "path": "ensembl-250k.gtf.gz",
+          "peak_rss_mib": 574.75,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.2360343339969404,
+            "convert": 0.2817149159964174,
+            "other": 0.4799745410127798,
+            "read": 0.27661270899989177
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.2743365000060294,
+          "release": "optimized",
+          "repeat": 3
+        },
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.6671,
+          "features": null,
+          "input_bytes": 3524624,
+          "path": "ensembl-250k.gtf.gz",
+          "peak_rss_mib": 561.09375,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.8928518329921644,
+            "convert": 0.33677229199383873,
+            "other": 0.24612533301115036,
+            "read": 0.1267874170007417
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.602536874997895,
+          "release": "2.9.1-polars",
+          "repeat": 3
+        },
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 3.302426,
+          "features": null,
+          "input_bytes": 3524624,
+          "path": "ensembl-250k.gtf.gz",
+          "peak_rss_mib": 615.78125,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 2.818784125003731,
+            "convert": 0.4906274579989258,
+            "other": 0.9829520420025801,
+            "read": 0.2855122500041034
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 4.57787587500934,
+          "release": "3.0.0-arrow",
+          "repeat": 3
+        },
+        {
+          "backend": "production",
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.4871540000000003,
+          "features": null,
+          "input_bytes": 109586053,
+          "path": "gencode-250k.gtf",
+          "peak_rss_mib": 527.203125,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 2.1337991250038613,
+            "convert": 4.9000009312294424e-05,
+            "other": 1.4431334999826504,
+            "read": 0.4248130839987425
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 4.0017947089945665,
+          "release": "optimized",
+          "repeat": 3
+        },
+        {
+          "backend": "production",
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.560311,
+          "features": null,
+          "input_bytes": 109586053,
+          "path": "gencode-250k.gtf",
+          "peak_rss_mib": 567.546875,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 2.646453332999954,
+            "convert": 0.09764800000994,
+            "other": 0.3458024169958662,
+            "read": 0.0905030409921892
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 3.1804067909979494,
+          "release": "2.9.1-polars",
+          "repeat": 3
+        },
+        {
+          "backend": "production",
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.658798,
+          "features": null,
+          "input_bytes": 109586053,
+          "path": "gencode-250k.gtf",
+          "peak_rss_mib": 587.6875,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.7814520410029218,
+            "convert": 7.35829962650314e-05,
+            "other": 0.5366334179998375,
+            "read": 0.26695341600861866
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.585112458007643,
+          "release": "3.0.0-arrow",
+          "repeat": 3
+        },
+        {
+          "backend": "production",
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.080072,
+          "features": null,
+          "input_bytes": 3520115,
+          "path": "gencode-250k.gtf.gz",
+          "peak_rss_mib": 557.890625,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.1859509170026286,
+            "convert": 4.124999395571649e-05,
+            "other": 0.5394327910034917,
+            "read": 0.2632630000007339
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 1.9886879580008099,
+          "release": "optimized",
+          "repeat": 3
+        },
+        {
+          "backend": "production",
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.2800469999999997,
+          "features": null,
+          "input_bytes": 3520115,
+          "path": "gencode-250k.gtf.gz",
+          "peak_rss_mib": 567.5,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.7495677080005407,
+            "convert": 0.06322920799721032,
+            "other": 0.27436150101129897,
+            "read": 0.11984191599185579
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.207000333000906,
+          "release": "2.9.1-polars",
+          "repeat": 3
+        },
+        {
+          "backend": "production",
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.574939,
+          "features": null,
+          "input_bytes": 3520115,
+          "path": "gencode-250k.gtf.gz",
+          "peak_rss_mib": 580.96875,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.7354764170013368,
+            "convert": 6.454098911490291e-05,
+            "other": 0.5007614180067321,
+            "read": 0.25421366600494366
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "2.3.3",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.4905160420021275,
+          "release": "3.0.0-arrow",
+          "repeat": 3
+        },
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.420429,
+          "features": null,
+          "input_bytes": 111900013,
+          "path": "ensembl-250k.gtf",
+          "peak_rss_mib": 552.5,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.8500060840015067,
+            "convert": 0.22684966700035147,
+            "other": 0.2211527069885051,
+            "read": 0.2438607500080252
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.5418692079983884,
+          "release": "2.9.1-polars",
+          "repeat": 1
+        },
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.85385,
+          "features": null,
+          "input_bytes": 111900013,
+          "path": "ensembl-250k.gtf",
+          "peak_rss_mib": 601.796875,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.83066029100155,
+            "convert": 0.18445454101311043,
+            "other": 0.5475739179964876,
+            "read": 0.21540637499128934
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.7780951250024373,
+          "release": "3.0.0-arrow",
+          "repeat": 1
+        },
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 1.9879889999999998,
+          "features": null,
+          "input_bytes": 111900013,
+          "path": "ensembl-250k.gtf",
+          "peak_rss_mib": 576.71875,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.255898375005927,
+            "convert": 0.18790645799890626,
+            "other": 0.2516916669992497,
+            "read": 0.21114029199816287
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 1.9066367920022458,
+          "release": "optimized",
+          "repeat": 1
+        },
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.415416,
+          "features": null,
+          "input_bytes": 3524624,
+          "path": "ensembl-250k.gtf.gz",
+          "peak_rss_mib": 553.046875,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.846398291992955,
+            "convert": 0.20270275000075344,
+            "other": 0.22192491599707864,
+            "read": 0.0779351250093896
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.3489610830001766,
+          "release": "2.9.1-polars",
+          "repeat": 1
+        },
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.888417,
+          "features": null,
+          "input_bytes": 3524624,
+          "path": "ensembl-250k.gtf.gz",
+          "peak_rss_mib": 599.984375,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.8276292090013158,
+            "convert": 0.18389533399022184,
+            "other": 0.5644444149947958,
+            "read": 0.2280619170051068
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.80403087499144,
+          "release": "3.0.0-arrow",
+          "repeat": 1
+        },
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 1.958045,
+          "features": null,
+          "input_bytes": 3524624,
+          "path": "ensembl-250k.gtf.gz",
+          "peak_rss_mib": 574.140625,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.2322392920032144,
+            "convert": 0.1897575000039069,
+            "other": 0.23446987399074715,
+            "read": 0.2203353750082897
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 1.876802041006158,
+          "release": "optimized",
+          "repeat": 1
+        },
+        {
+          "backend": "production",
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.116965,
+          "features": null,
+          "input_bytes": 109586053,
+          "path": "gencode-250k.gtf",
+          "peak_rss_mib": 559.671875,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.7515429170016432,
+            "convert": 0.017430042003979906,
+            "other": 0.24500191600236576,
+            "read": 0.03371124999830499
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.047686125006294,
+          "release": "2.9.1-polars",
+          "repeat": 1
+        },
+        {
+          "backend": "production",
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.606457,
+          "features": null,
+          "input_bytes": 109586053,
+          "path": "gencode-250k.gtf",
+          "peak_rss_mib": 598.234375,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.736094084000797,
+            "convert": 0.00014029200247023255,
+            "other": 0.5668964579963358,
+            "read": 0.21073287499893922
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.5138637089985423,
+          "release": "3.0.0-arrow",
+          "repeat": 1
+        },
+        {
+          "backend": "production",
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 1.784423,
+          "features": null,
+          "input_bytes": 109586053,
+          "path": "gencode-250k.gtf",
+          "peak_rss_mib": 578.703125,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.3149310419976246,
+            "convert": 2.5416011339984834e-05,
+            "other": 0.28562641798635013,
+            "read": 0.2182972080045147
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 1.8188800839998294,
+          "release": "optimized",
+          "repeat": 1
+        },
+        {
+          "backend": "production",
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.1718260000000003,
+          "features": null,
+          "input_bytes": 3520115,
+          "path": "gencode-250k.gtf.gz",
+          "peak_rss_mib": 560.0,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.7444164579937933,
+            "convert": 0.017836250000982545,
+            "other": 0.26727887500601355,
+            "read": 0.07006466700113378
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.099596250001923,
+          "release": "2.9.1-polars",
+          "repeat": 1
+        },
+        {
+          "backend": "production",
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.589577,
+          "features": null,
+          "input_bytes": 3520115,
+          "path": "gencode-250k.gtf.gz",
+          "peak_rss_mib": 595.265625,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.7193977919960162,
+            "convert": 0.0001550410088384524,
+            "other": 0.5679449589806609,
+            "read": 0.21822095800598618
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.5057187499915017,
+          "release": "3.0.0-arrow",
+          "repeat": 1
+        },
+        {
+          "backend": "production",
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 1.789281,
+          "features": null,
+          "input_bytes": 3520115,
+          "path": "gencode-250k.gtf.gz",
+          "peak_rss_mib": 583.171875,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.208379874995444,
+            "convert": 2.6458001229912043e-05,
+            "other": 0.26576029100397136,
+            "read": 0.21722579200286418
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 1.6913924160035094,
+          "release": "optimized",
+          "repeat": 1
+        },
+        {
+          "backend": "production",
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.6144279999999998,
+          "features": null,
+          "input_bytes": 3520115,
+          "path": "gencode-250k.gtf.gz",
+          "peak_rss_mib": 589.8125,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.7339677080017282,
+            "convert": 0.00014095898950472474,
+            "other": 0.5685903760022484,
+            "read": 0.2213319160073297
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.524030959000811,
+          "release": "3.0.0-arrow",
+          "repeat": 2
+        },
+        {
+          "backend": "production",
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 1.827663,
+          "features": null,
+          "input_bytes": 3520115,
+          "path": "gencode-250k.gtf.gz",
+          "peak_rss_mib": 574.265625,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.236397083994234,
+            "convert": 2.5416986318305135e-05,
+            "other": 0.27114258201618213,
+            "read": 0.22319108400552068
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 1.730756167002255,
+          "release": "optimized",
+          "repeat": 2
+        },
+        {
+          "backend": "production",
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.24322,
+          "features": null,
+          "input_bytes": 3520115,
+          "path": "gencode-250k.gtf.gz",
+          "peak_rss_mib": 560.78125,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.8704789170005824,
+            "convert": 0.024456458995700814,
+            "other": 0.2771191240171902,
+            "read": 0.08153274998767301
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.2535872500011465,
+          "release": "2.9.1-polars",
+          "repeat": 2
+        },
+        {
+          "backend": "production",
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.745542,
+          "features": null,
+          "input_bytes": 109586053,
+          "path": "gencode-250k.gtf",
+          "peak_rss_mib": 591.375,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.8712519999971846,
+            "convert": 0.00015558300947304815,
+            "other": 0.6532173339946894,
+            "read": 0.22314999999071006
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.747774916992057,
+          "release": "3.0.0-arrow",
+          "repeat": 2
+        },
+        {
+          "backend": "production",
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 1.7458540000000002,
+          "features": null,
+          "input_bytes": 109586053,
+          "path": "gencode-250k.gtf",
+          "peak_rss_mib": 576.578125,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.1965720830048667,
+            "convert": 2.504199801478535e-05,
+            "other": 0.25334479099547025,
+            "read": 0.20459700000355951
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 1.6545389160019113,
+          "release": "optimized",
+          "repeat": 2
+        },
+        {
+          "backend": "production",
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.082073,
+          "features": null,
+          "input_bytes": 109586053,
+          "path": "gencode-250k.gtf",
+          "peak_rss_mib": 559.5,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.71495758301171,
+            "convert": 0.016908165998756886,
+            "other": 0.23707550199469551,
+            "read": 0.027257708003162406
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 1.996198959008325,
+          "release": "2.9.1-polars",
+          "repeat": 2
+        },
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.854932,
+          "features": null,
+          "input_bytes": 3524624,
+          "path": "ensembl-250k.gtf.gz",
+          "peak_rss_mib": 607.015625,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.81500345900713,
+            "convert": 0.18400516700057779,
+            "other": 0.5284704989899183,
+            "read": 0.23586362499918323
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.7633427499968093,
+          "release": "3.0.0-arrow",
+          "repeat": 2
+        },
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 1.9496789999999997,
+          "features": null,
+          "input_bytes": 3524624,
+          "path": "ensembl-250k.gtf.gz",
+          "peak_rss_mib": 572.890625,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.2216533339960733,
+            "convert": 0.18549491598969325,
+            "other": 0.22824766600388102,
+            "read": 0.22112158400705084
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 1.8565174999966985,
+          "release": "optimized",
+          "repeat": 2
+        },
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.4144940000000004,
+          "features": null,
+          "input_bytes": 3524624,
+          "path": "ensembl-250k.gtf.gz",
+          "peak_rss_mib": 553.15625,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.8477172499988228,
+            "convert": 0.20435254200128838,
+            "other": 0.21732766600325704,
+            "read": 0.07238537499506492
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.341782832998433,
+          "release": "2.9.1-polars",
+          "repeat": 2
+        },
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.896116,
+          "features": null,
+          "input_bytes": 111900013,
+          "path": "ensembl-250k.gtf",
+          "peak_rss_mib": 599.765625,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.8735739160038065,
+            "convert": 0.1892745829973137,
+            "other": 0.5461084599955939,
+            "read": 0.21546933299396187
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.824426291990676,
+          "release": "3.0.0-arrow",
+          "repeat": 2
+        },
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 1.981777,
+          "features": null,
+          "input_bytes": 111900013,
+          "path": "ensembl-250k.gtf",
+          "peak_rss_mib": 578.09375,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.241973957992741,
+            "convert": 0.1873781250033062,
+            "other": 0.23873687500599772,
+            "read": 0.21316074999049306
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 1.8812497079925379,
+          "release": "optimized",
+          "repeat": 2
+        },
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.414792,
+          "features": null,
+          "input_bytes": 111900013,
+          "path": "ensembl-250k.gtf",
+          "peak_rss_mib": 552.828125,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.8557262500107754,
+            "convert": 0.2145706660085125,
+            "other": 0.23259929296909831,
+            "read": 0.03862558300897945
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.3415217919973657,
+          "release": "2.9.1-polars",
+          "repeat": 2
+        },
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.111438,
+          "features": null,
+          "input_bytes": 111900013,
+          "path": "ensembl-250k.gtf",
+          "peak_rss_mib": 574.484375,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.4716607090085745,
+            "convert": 0.19871799999964423,
+            "other": 0.28257362397562247,
+            "read": 0.21565358400403056
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.1686059169878718,
+          "release": "optimized",
+          "repeat": 3
+        },
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.556296,
+          "features": null,
+          "input_bytes": 111900013,
+          "path": "ensembl-250k.gtf",
+          "peak_rss_mib": 552.21875,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 2.1370639169908827,
+            "convert": 0.23608004199923016,
+            "other": 0.31135054100013804,
+            "read": 0.033251375003601424
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.7177458749938523,
+          "release": "2.9.1-polars",
+          "repeat": 3
+        },
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.979457,
+          "features": null,
+          "input_bytes": 111900013,
+          "path": "ensembl-250k.gtf",
+          "peak_rss_mib": 590.28125,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.8968122079968452,
+            "convert": 0.18741612500161864,
+            "other": 0.5885198740143096,
+            "read": 0.22517183399759233
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.897920041010366,
+          "release": "3.0.0-arrow",
+          "repeat": 3
+        },
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.03234,
+          "features": null,
+          "input_bytes": 3524624,
+          "path": "ensembl-250k.gtf.gz",
+          "peak_rss_mib": 574.84375,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.2583790000062436,
+            "convert": 0.19016345900308806,
+            "other": 0.2520635830005631,
+            "read": 0.22867120799492113
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 1.9292772500048159,
+          "release": "optimized",
+          "repeat": 3
+        },
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.4895519999999998,
+          "features": null,
+          "input_bytes": 3524624,
+          "path": "ensembl-250k.gtf.gz",
+          "peak_rss_mib": 553.421875,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.8483212909923168,
+            "convert": 0.21014762499544304,
+            "other": 0.2704182920133462,
+            "read": 0.08297458299784921
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.4118617909989553,
+          "release": "2.9.1-polars",
+          "repeat": 3
+        },
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.9196619999999998,
+          "features": null,
+          "input_bytes": 3524624,
+          "path": "ensembl-250k.gtf.gz",
+          "peak_rss_mib": 595.125,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.8424518750107381,
+            "convert": 0.1856644170038635,
+            "other": 0.556629457991221,
+            "read": 0.23696545799612068
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.8217112080019433,
+          "release": "3.0.0-arrow",
+          "repeat": 3
+        },
+        {
+          "backend": "production",
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 1.804453,
+          "features": null,
+          "input_bytes": 109586053,
+          "path": "gencode-250k.gtf",
+          "peak_rss_mib": 576.234375,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.216064499996719,
+            "convert": 2.633300027810037e-05,
+            "other": 0.28129791701212525,
+            "read": 0.21160841699747834
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 1.7089971670066006,
+          "release": "optimized",
+          "repeat": 3
+        },
+        {
+          "backend": "production",
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.142511,
+          "features": null,
+          "input_bytes": 109586053,
+          "path": "gencode-250k.gtf",
+          "peak_rss_mib": 559.421875,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.7327500000101281,
+            "convert": 0.02303095800743904,
+            "other": 0.27439587596745696,
+            "read": 0.027468333006254397
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.0576451669912785,
+          "release": "2.9.1-polars",
+          "repeat": 3
+        },
+        {
+          "backend": "production",
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.681481,
+          "features": null,
+          "input_bytes": 109586053,
+          "path": "gencode-250k.gtf",
+          "peak_rss_mib": 593.15625,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.8231396670016693,
+            "convert": 0.00015658400661777705,
+            "other": 0.6528683740034467,
+            "read": 0.20600433299841825
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.682168958010152,
+          "release": "3.0.0-arrow",
+          "repeat": 3
+        },
+        {
+          "backend": "production",
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 1.7782019999999998,
+          "features": null,
+          "input_bytes": 3520115,
+          "path": "gencode-250k.gtf.gz",
+          "peak_rss_mib": 572.78125,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.23295112499909,
+            "convert": 2.3667002096772194e-05,
+            "other": 0.268862833007006,
+            "read": 0.22755545799736865
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 1.7293930830055615,
+          "release": "optimized",
+          "repeat": 3
+        },
+        {
+          "backend": "production",
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.126937,
+          "features": null,
+          "input_bytes": 3520115,
+          "path": "gencode-250k.gtf.gz",
+          "peak_rss_mib": 560.15625,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.7179349170037312,
+            "convert": 0.018364208008279093,
+            "other": 0.2524259169877041,
+            "read": 0.06932729099935386
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.0580523329990683,
+          "release": "2.9.1-polars",
+          "repeat": 3
+        },
+        {
+          "backend": "production",
+          "checksum": "8d394118a91daefd2d5ee4f50c07ef5dee567e6ad373c052fa6cf3898c807641",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.6059170000000003,
+          "features": null,
+          "input_bytes": 3520115,
+          "path": "gencode-250k.gtf.gz",
+          "peak_rss_mib": 595.03125,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.7348792080010753,
+            "convert": 0.00019987499399576336,
+            "other": 0.5726997080055298,
+            "read": 0.2197710839973297
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.5275498749979306,
+          "release": "3.0.0-arrow",
+          "repeat": 3
+        }
+      ]
+    },
+    "full": {
+      "description": "Complete production readers; imports/checksums outside wall time; sequential fresh processes",
+      "date": "2026-10-06T16:49:49.180246+00:00",
+      "threads": 4,
+      "inputs": {
+        "Homo_sapiens.GRCh38.114.gtf.gz": {
+          "bytes": 64228966,
+          "sha256": "75ec5b7a1bbb8ce566622dd9abf974ad37d390130325dae6d197d1a590854c50",
+          "rows": 4116048,
+          "source_url": "https://ftp.ensembl.org/pub/release-114/gtf/homo_sapiens/Homo_sapiens.GRCh38.114.gtf.gz",
+          "selection": "full file"
+        },
+        "gencode.v48.primary_assembly.annotation.gtf.gz": {
+          "bytes": 59094311,
+          "sha256": "ad15e956fb7643172afca6dcaae176cc76400532392978a191c981d0a052e668",
+          "rows": 4119244,
+          "source_url": "https://ftp.ebi.ac.uk/pub/databases/gencode/Gencode_human/release_48/gencode.v48.primary_assembly.annotation.gtf.gz",
+          "selection": "full file"
+        }
+      },
+      "summaries": [
+        {
+          "pandas": "3.0.6",
+          "input": "Homo_sapiens.GRCh38.114.gtf.gz",
+          "release": "2.9.1-polars",
+          "runs": 1,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": {
+            "median": 47.106003042004886,
+            "min": 47.106003042004886,
+            "max": 47.106003042004886
+          },
+          "cpu_seconds": {
+            "median": 41.689209,
+            "min": 41.689209,
+            "max": 41.689209
+          },
+          "peak_rss_mib": {
+            "median": 6504.0625,
+            "min": 6504.0625,
+            "max": 6504.0625
+          },
+          "stage_medians": {
+            "attributes": 34.80686404100561,
+            "convert": 4.723838749996503,
+            "other": 6.347857333996217,
+            "read": 1.227442917006556
+          },
+          "checksum": "86a89f01e15e261a4d93558f429f843ea5bfe7f7b9ab862b7b7cbc59df8d6988"
+        },
+        {
+          "pandas": "3.0.6",
+          "input": "Homo_sapiens.GRCh38.114.gtf.gz",
+          "release": "3.0.0-arrow",
+          "runs": 1,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": {
+            "median": 67.63390304200584,
+            "min": 67.63390304200584,
+            "max": 67.63390304200584
+          },
+          "cpu_seconds": {
+            "median": 54.188731,
+            "min": 54.188731,
+            "max": 54.188731
+          },
+          "peak_rss_mib": {
+            "median": 7198.8125,
+            "min": 7198.8125,
+            "max": 7198.8125
+          },
+          "stage_medians": {
+            "attributes": 47.70297658399795,
+            "convert": 3.748629665991757,
+            "other": 11.267300334016909,
+            "read": 4.914996457999223
+          },
+          "checksum": "86a89f01e15e261a4d93558f429f843ea5bfe7f7b9ab862b7b7cbc59df8d6988"
+        },
+        {
+          "pandas": "3.0.6",
+          "input": "Homo_sapiens.GRCh38.114.gtf.gz",
+          "release": "optimized",
+          "runs": 1,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": {
+            "median": 36.860391041002,
+            "min": 36.860391041002,
+            "max": 36.860391041002
+          },
+          "cpu_seconds": {
+            "median": 35.70663,
+            "min": 35.70663,
+            "max": 35.70663
+          },
+          "peak_rss_mib": {
+            "median": 3639.3125,
+            "min": 3639.3125,
+            "max": 3639.3125
+          },
+          "stage_medians": {
+            "attributes": 27.825113583006896,
+            "convert": 3.362754040994332,
+            "other": 0.15611587499734014,
+            "read": 5.516407542003435
+          },
+          "checksum": "86a89f01e15e261a4d93558f429f843ea5bfe7f7b9ab862b7b7cbc59df8d6988"
+        },
+        {
+          "pandas": "3.0.6",
+          "input": "gencode.v48.primary_assembly.annotation.gtf.gz",
+          "release": "2.9.1-polars",
+          "runs": 1,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": {
+            "median": 43.73683041699405,
+            "min": 43.73683041699405,
+            "max": 43.73683041699405
+          },
+          "cpu_seconds": {
+            "median": 38.00544,
+            "min": 38.00544,
+            "max": 38.00544
+          },
+          "peak_rss_mib": {
+            "median": 7090.890625,
+            "min": 7090.890625,
+            "max": 7090.890625
+          },
+          "stage_medians": {
+            "attributes": 36.88679691699508,
+            "convert": 0.4109674579958664,
+            "other": 4.4899386250035604,
+            "read": 1.949127416999545
+          },
+          "checksum": "52d9ca5d36dd0795856c27637e70e93cff34511b12226af66a07b4eb114d3d49"
+        },
+        {
+          "pandas": "3.0.6",
+          "input": "gencode.v48.primary_assembly.annotation.gtf.gz",
+          "release": "3.0.0-arrow",
+          "runs": 1,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": {
+            "median": 52.35360045799462,
+            "min": 52.35360045799462,
+            "max": 52.35360045799462
+          },
+          "cpu_seconds": {
+            "median": 49.702405,
+            "min": 49.702405,
+            "max": 49.702405
+          },
+          "peak_rss_mib": {
+            "median": 6559.75,
+            "min": 6559.75,
+            "max": 6559.75
+          },
+          "stage_medians": {
+            "attributes": 35.62854295800207,
+            "convert": 0.00019679100660141557,
+            "other": 12.81204349998734,
+            "read": 3.9128172089986037
+          },
+          "checksum": "52d9ca5d36dd0795856c27637e70e93cff34511b12226af66a07b4eb114d3d49"
+        },
+        {
+          "pandas": "3.0.6",
+          "input": "gencode.v48.primary_assembly.annotation.gtf.gz",
+          "release": "optimized",
+          "runs": 1,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": {
+            "median": 29.878576290997444,
+            "min": 29.878576290997444,
+            "max": 29.878576290997444
+          },
+          "cpu_seconds": {
+            "median": 30.355336,
+            "min": 30.355336,
+            "max": 30.355336
+          },
+          "peak_rss_mib": {
+            "median": 3697.5625,
+            "min": 3697.5625,
+            "max": 3697.5625
+          },
+          "stage_medians": {
+            "attributes": 25.754451957996935,
+            "convert": 2.766599936876446e-05,
+            "other": 0.14164995899773203,
+            "read": 3.9824467080034083
+          },
+          "checksum": "52d9ca5d36dd0795856c27637e70e93cff34511b12226af66a07b4eb114d3d49"
+        }
+      ],
+      "measurements": [
+        {
+          "backend": "production",
+          "checksum": "86a89f01e15e261a4d93558f429f843ea5bfe7f7b9ab862b7b7cbc59df8d6988",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 41.689209,
+          "features": null,
+          "input_bytes": 64228966,
+          "path": "Homo_sapiens.GRCh38.114.gtf.gz",
+          "peak_rss_mib": 6504.0625,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 4116048,
+          "stages": {
+            "attributes": 34.80686404100561,
+            "convert": 4.723838749996503,
+            "other": 6.347857333996217,
+            "read": 1.227442917006556
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 47.106003042004886,
+          "release": "2.9.1-polars",
+          "repeat": 1
+        },
+        {
+          "backend": "production",
+          "checksum": "86a89f01e15e261a4d93558f429f843ea5bfe7f7b9ab862b7b7cbc59df8d6988",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 54.188731,
+          "features": null,
+          "input_bytes": 64228966,
+          "path": "Homo_sapiens.GRCh38.114.gtf.gz",
+          "peak_rss_mib": 7198.8125,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 4116048,
+          "stages": {
+            "attributes": 47.70297658399795,
+            "convert": 3.748629665991757,
+            "other": 11.267300334016909,
+            "read": 4.914996457999223
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 67.63390304200584,
+          "release": "3.0.0-arrow",
+          "repeat": 1
+        },
+        {
+          "backend": "production",
+          "checksum": "86a89f01e15e261a4d93558f429f843ea5bfe7f7b9ab862b7b7cbc59df8d6988",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 35.70663,
+          "features": null,
+          "input_bytes": 64228966,
+          "path": "Homo_sapiens.GRCh38.114.gtf.gz",
+          "peak_rss_mib": 3639.3125,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 4116048,
+          "stages": {
+            "attributes": 27.825113583006896,
+            "convert": 3.362754040994332,
+            "other": 0.15611587499734014,
+            "read": 5.516407542003435
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 36.860391041002,
+          "release": "optimized",
+          "repeat": 1
+        },
+        {
+          "backend": "production",
+          "checksum": "52d9ca5d36dd0795856c27637e70e93cff34511b12226af66a07b4eb114d3d49",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid",
+            "artif_dupl"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 38.00544,
+          "features": null,
+          "input_bytes": 59094311,
+          "path": "gencode.v48.primary_assembly.annotation.gtf.gz",
+          "peak_rss_mib": 7090.890625,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 4119244,
+          "stages": {
+            "attributes": 36.88679691699508,
+            "convert": 0.4109674579958664,
+            "other": 4.4899386250035604,
+            "read": 1.949127416999545
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 43.73683041699405,
+          "release": "2.9.1-polars",
+          "repeat": 1
+        },
+        {
+          "backend": "production",
+          "checksum": "52d9ca5d36dd0795856c27637e70e93cff34511b12226af66a07b4eb114d3d49",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid",
+            "artif_dupl"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 49.702405,
+          "features": null,
+          "input_bytes": 59094311,
+          "path": "gencode.v48.primary_assembly.annotation.gtf.gz",
+          "peak_rss_mib": 6559.75,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 4119244,
+          "stages": {
+            "attributes": 35.62854295800207,
+            "convert": 0.00019679100660141557,
+            "other": 12.81204349998734,
+            "read": 3.9128172089986037
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 52.35360045799462,
+          "release": "3.0.0-arrow",
+          "repeat": 1
+        },
+        {
+          "backend": "production",
+          "checksum": "52d9ca5d36dd0795856c27637e70e93cff34511b12226af66a07b4eb114d3d49",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_type",
+            "gene_name",
+            "level",
+            "tag",
+            "transcript_id",
+            "transcript_type",
+            "transcript_name",
+            "exon_number",
+            "exon_id",
+            "transcript_support_level",
+            "havana_transcript",
+            "hgnc_id",
+            "havana_gene",
+            "ont",
+            "protein_id",
+            "ccdsid",
+            "artif_dupl"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 30.355336,
+          "features": null,
+          "input_bytes": 59094311,
+          "path": "gencode.v48.primary_assembly.annotation.gtf.gz",
+          "peak_rss_mib": 3697.5625,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 4119244,
+          "stages": {
+            "attributes": 25.754451957996935,
+            "convert": 2.766599936876446e-05,
+            "other": 0.14164995899773203,
+            "read": 3.9824467080034083
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 29.878576290997444,
+          "release": "optimized",
+          "repeat": 1
+        }
+      ]
+    },
+    "selective": {
+      "description": "Complete production readers; imports/checksums outside wall time; sequential fresh processes",
+      "date": "2026-10-06T16:50:13.534320+00:00",
+      "threads": 4,
+      "inputs": {
+        "ensembl-250k.gtf.gz": {
+          "bytes": 3524624,
+          "sha256": "3000128cb22ea68754c01bd3d72fe347c9c86d05b9cae9fed65d1440c57291cd",
+          "rows": 250000,
+          "source_url": "https://ftp.ensembl.org/pub/release-114/gtf/homo_sapiens/Homo_sapiens.GRCh38.114.gtf.gz",
+          "selection": "first 250000 annotation rows plus original header"
+        }
+      },
+      "summaries": [
+        {
+          "pandas": "3.0.6",
+          "input": "ensembl-250k.gtf.gz",
+          "release": "2.9.1-polars",
+          "runs": 1,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": {
+            "median": 1.3659025410015602,
+            "min": 1.3659025410015602,
+            "max": 1.3659025410015602
+          },
+          "cpu_seconds": {
+            "median": 0.9866109999999999,
+            "min": 0.9866109999999999,
+            "max": 0.9866109999999999
+          },
+          "peak_rss_mib": {
+            "median": 319.515625,
+            "min": 319.515625,
+            "max": 319.515625
+          },
+          "stage_medians": {
+            "attributes": 1.0164038330112817,
+            "convert": 0.08103683299850672,
+            "other": 0.010387374990386888,
+            "read": 0.2580745000013849
+          },
+          "checksum": "82d74ad0f84f29bf2bf7aaa190846055aa244cb570b6799827b61970894e035b"
+        },
+        {
+          "pandas": "3.0.6",
+          "input": "ensembl-250k.gtf.gz",
+          "release": "3.0.0-arrow",
+          "runs": 1,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": {
+            "median": 1.4581702080031391,
+            "min": 1.4581702080031391,
+            "max": 1.4581702080031391
+          },
+          "cpu_seconds": {
+            "median": 1.165252,
+            "min": 1.165252,
+            "max": 1.165252
+          },
+          "peak_rss_mib": {
+            "median": 403.0625,
+            "min": 403.0625,
+            "max": 403.0625
+          },
+          "stage_medians": {
+            "attributes": 1.015940708006383,
+            "convert": 0.0028895419964101166,
+            "other": 0.052480124999419786,
+            "read": 0.3868598330009263
+          },
+          "checksum": "82d74ad0f84f29bf2bf7aaa190846055aa244cb570b6799827b61970894e035b"
+        },
+        {
+          "pandas": "3.0.6",
+          "input": "ensembl-250k.gtf.gz",
+          "release": "optimized",
+          "runs": 1,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": {
+            "median": 1.0357164169981843,
+            "min": 1.0357164169981843,
+            "max": 1.0357164169981843
+          },
+          "cpu_seconds": {
+            "median": 0.9732489999999999,
+            "min": 0.9732489999999999,
+            "max": 0.9732489999999999
+          },
+          "peak_rss_mib": {
+            "median": 402.984375,
+            "min": 402.984375,
+            "max": 402.984375
+          },
+          "stage_medians": {
+            "attributes": 0.6760162090067752,
+            "convert": 0.0008421660022577271,
+            "other": 0.01728591699793469,
+            "read": 0.3415721249912167
+          },
+          "checksum": "82d74ad0f84f29bf2bf7aaa190846055aa244cb570b6799827b61970894e035b"
+        }
+      ],
+      "measurements": [
+        {
+          "backend": "production",
+          "checksum": "82d74ad0f84f29bf2bf7aaa190846055aa244cb570b6799827b61970894e035b",
+          "columns": [
+            "seqname",
+            "start",
+            "end",
+            "gene_id",
+            "transcript_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 0.9866109999999999,
+          "features": [
+            "exon"
+          ],
+          "input_bytes": 3524624,
+          "path": "ensembl-250k.gtf.gz",
+          "peak_rss_mib": 319.515625,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 116284,
+          "stages": {
+            "attributes": 1.0164038330112817,
+            "convert": 0.08103683299850672,
+            "other": 0.010387374990386888,
+            "read": 0.2580745000013849
+          },
+          "threads": 4,
+          "usecols": [
+            "seqname",
+            "start",
+            "end",
+            "gene_id",
+            "transcript_id"
+          ],
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 1.3659025410015602,
+          "release": "2.9.1-polars",
+          "repeat": 1
+        },
+        {
+          "backend": "production",
+          "checksum": "82d74ad0f84f29bf2bf7aaa190846055aa244cb570b6799827b61970894e035b",
+          "columns": [
+            "seqname",
+            "start",
+            "end",
+            "gene_id",
+            "transcript_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 1.165252,
+          "features": [
+            "exon"
+          ],
+          "input_bytes": 3524624,
+          "path": "ensembl-250k.gtf.gz",
+          "peak_rss_mib": 403.0625,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 116284,
+          "stages": {
+            "attributes": 1.015940708006383,
+            "convert": 0.0028895419964101166,
+            "other": 0.052480124999419786,
+            "read": 0.3868598330009263
+          },
+          "threads": 4,
+          "usecols": [
+            "seqname",
+            "start",
+            "end",
+            "gene_id",
+            "transcript_id"
+          ],
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 1.4581702080031391,
+          "release": "3.0.0-arrow",
+          "repeat": 1
+        },
+        {
+          "backend": "production",
+          "checksum": "82d74ad0f84f29bf2bf7aaa190846055aa244cb570b6799827b61970894e035b",
+          "columns": [
+            "seqname",
+            "start",
+            "end",
+            "gene_id",
+            "transcript_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 0.9732489999999999,
+          "features": [
+            "exon"
+          ],
+          "input_bytes": 3524624,
+          "path": "ensembl-250k.gtf.gz",
+          "peak_rss_mib": 402.984375,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "pandas",
+          "rows": 116284,
+          "stages": {
+            "attributes": 0.6760162090067752,
+            "convert": 0.0008421660022577271,
+            "other": 0.01728591699793469,
+            "read": 0.3415721249912167
+          },
+          "threads": 4,
+          "usecols": [
+            "seqname",
+            "start",
+            "end",
+            "gene_id",
+            "transcript_id"
+          ],
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 1.0357164169981843,
+          "release": "optimized",
+          "repeat": 1
+        }
+      ]
+    },
+    "optional": {
+      "description": "Complete production readers; imports/checksums outside wall time; sequential fresh processes",
+      "date": "2026-10-06T16:50:25.858821+00:00",
+      "threads": 4,
+      "inputs": {
+        "ensembl-250k.gtf.gz": {
+          "bytes": 3524624,
+          "sha256": "3000128cb22ea68754c01bd3d72fe347c9c86d05b9cae9fed65d1440c57291cd",
+          "rows": 250000,
+          "source_url": "https://ftp.ensembl.org/pub/release-114/gtf/homo_sapiens/Homo_sapiens.GRCh38.114.gtf.gz",
+          "selection": "first 250000 annotation rows plus original header"
+        }
+      },
+      "summaries": [
+        {
+          "pandas": "3.0.6",
+          "input": "ensembl-250k.gtf.gz",
+          "release": "2.9.1-polars",
+          "runs": 1,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": {
+            "median": 3.2213837079907535,
+            "min": 3.2213837079907535,
+            "max": 3.2213837079907535
+          },
+          "cpu_seconds": {
+            "median": 2.815719,
+            "min": 2.815719,
+            "max": 2.815719
+          },
+          "peak_rss_mib": {
+            "median": 511.4375,
+            "min": 511.4375,
+            "max": 511.4375
+          },
+          "stage_medians": {
+            "attributes": 2.372206415995606,
+            "convert": 0.4352557079982944,
+            "other": 0.3047657509887358,
+            "read": 0.10915583300811704
+          },
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8"
+        },
+        {
+          "pandas": "3.0.6",
+          "input": "ensembl-250k.gtf.gz",
+          "release": "3.0.0-arrow",
+          "runs": 1,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": {
+            "median": 3.4607763749954756,
+            "min": 3.4607763749954756,
+            "max": 3.4607763749954756
+          },
+          "cpu_seconds": {
+            "median": 3.2452810000000003,
+            "min": 3.2452810000000003,
+            "max": 3.2452810000000003
+          },
+          "peak_rss_mib": {
+            "median": 599.03125,
+            "min": 599.03125,
+            "max": 599.03125
+          },
+          "stage_medians": {
+            "attributes": 2.323183833999792,
+            "convert": 0.2275274999992689,
+            "other": 0.6323161249893019,
+            "read": 0.2777489160071127
+          },
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8"
+        },
+        {
+          "pandas": "3.0.6",
+          "input": "ensembl-250k.gtf.gz",
+          "release": "optimized",
+          "runs": 1,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": {
+            "median": 2.401918500007014,
+            "min": 2.401918500007014,
+            "max": 2.401918500007014
+          },
+          "cpu_seconds": {
+            "median": 2.172716,
+            "min": 2.172716,
+            "max": 2.172716
+          },
+          "peak_rss_mib": {
+            "median": 581.59375,
+            "min": 581.59375,
+            "max": 581.59375
+          },
+          "stage_medians": {
+            "attributes": 1.5562499160005245,
+            "convert": 0.2698556249961257,
+            "other": 0.3231227930082241,
+            "read": 0.25269016600213945
+          },
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8"
+        }
+      ],
+      "measurements": [
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.815719,
+          "features": null,
+          "input_bytes": 3524624,
+          "path": "ensembl-250k.gtf.gz",
+          "peak_rss_mib": 511.4375,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "polars",
+          "rows": 250000,
+          "stages": {
+            "attributes": 2.372206415995606,
+            "convert": 0.4352557079982944,
+            "other": 0.3047657509887358,
+            "read": 0.10915583300811704
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "2.9.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 3.2213837079907535,
+          "release": "2.9.1-polars",
+          "repeat": 1
+        },
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 3.2452810000000003,
+          "features": null,
+          "input_bytes": 3524624,
+          "path": "ensembl-250k.gtf.gz",
+          "peak_rss_mib": 599.03125,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "polars",
+          "rows": 250000,
+          "stages": {
+            "attributes": 2.323183833999792,
+            "convert": 0.2275274999992689,
+            "other": 0.6323161249893019,
+            "read": 0.2777489160071127
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.0",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 3.4607763749954756,
+          "release": "3.0.0-arrow",
+          "repeat": 1
+        },
+        {
+          "backend": "production",
+          "checksum": "be944cfa699afc930317973a3ee1ca1dc6d309c10071bc32b63a9b54abbdcca8",
+          "columns": [
+            "seqname",
+            "source",
+            "feature",
+            "start",
+            "end",
+            "score",
+            "strand",
+            "frame",
+            "gene_id",
+            "gene_version",
+            "gene_name",
+            "gene_source",
+            "gene_biotype",
+            "transcript_id",
+            "transcript_version",
+            "transcript_name",
+            "transcript_source",
+            "transcript_biotype",
+            "tag",
+            "transcript_support_level",
+            "exon_number",
+            "exon_id",
+            "exon_version",
+            "protein_id",
+            "protein_version",
+            "ccds_id"
+          ],
+          "cpu_count": 10,
+          "cpu_seconds": 2.172716,
+          "features": null,
+          "input_bytes": 3524624,
+          "path": "ensembl-250k.gtf.gz",
+          "peak_rss_mib": 581.59375,
+          "platform": "macOS-26.5-arm64-arm-64bit",
+          "result_type": "polars",
+          "rows": 250000,
+          "stages": {
+            "attributes": 1.5562499160005245,
+            "convert": 0.2698556249961257,
+            "other": 0.3231227930082241,
+            "read": 0.25269016600213945
+          },
+          "threads": 4,
+          "usecols": null,
+          "versions": {
+            "gtfparse": "3.0.1",
+            "numpy": "2.5.3",
+            "pandas": "3.0.6",
+            "polars": "1.44.2",
+            "pyarrow": "25.0.1",
+            "python": "3.12.6"
+          },
+          "wall_seconds": 2.401918500007014,
+          "release": "optimized",
+          "repeat": 1
+        }
+      ]
+    }
+  },
+  "development_experiments": {
+    "unbatched_arrow_construction": {
+      "status": "rejected due to full-file temporary Python memory; excluded from final tables",
+      "full_sweep": {
+        "description": "Complete production readers; imports/checksums outside wall time; sequential fresh processes",
+        "date": "2026-10-06T16:22:10.192558+00:00",
+        "threads": 4,
+        "inputs": {
+          "Homo_sapiens.GRCh38.114.gtf.gz": {
+            "bytes": 64228966,
+            "sha256": "75ec5b7a1bbb8ce566622dd9abf974ad37d390130325dae6d197d1a590854c50"
+          },
+          "gencode.v48.primary_assembly.annotation.gtf.gz": {
+            "bytes": 59094311,
+            "sha256": "ad15e956fb7643172afca6dcaae176cc76400532392978a191c981d0a052e668"
+          }
+        },
+        "summaries": [
+          {
+            "pandas": "3.0.6",
+            "input": "Homo_sapiens.GRCh38.114.gtf.gz",
+            "release": "2.9.1-polars",
+            "runs": 1,
+            "versions": {
+              "gtfparse": "2.9.1",
+              "numpy": "2.5.3",
+              "pandas": "3.0.6",
+              "polars": "1.44.2",
+              "pyarrow": "25.0.1",
+              "python": "3.12.6"
+            },
+            "wall_seconds": {
+              "median": 38.26553320900712,
+              "min": 38.26553320900712,
+              "max": 38.26553320900712
+            },
+            "cpu_seconds": {
+              "median": 39.141783,
+              "min": 39.141783,
+              "max": 39.141783
+            },
+            "peak_rss_mib": {
+              "median": 5904.65625,
+              "min": 5904.65625,
+              "max": 5904.65625
+            },
+            "stage_medians": {
+              "attributes": 29.537428333002026,
+              "convert": 3.2695007080037612,
+              "other": 4.044133210001746,
+              "read": 1.4144709579995833
+            },
+            "checksum": "86a89f01e15e261a4d93558f429f843ea5bfe7f7b9ab862b7b7cbc59df8d6988"
+          },
+          {
+            "pandas": "3.0.6",
+            "input": "Homo_sapiens.GRCh38.114.gtf.gz",
+            "release": "3.0.0-arrow",
+            "runs": 1,
+            "versions": {
+              "gtfparse": "3.0.0",
+              "numpy": "2.5.3",
+              "pandas": "3.0.6",
+              "polars": "1.44.2",
+              "pyarrow": "25.0.1",
+              "python": "3.12.6"
+            },
+            "wall_seconds": {
+              "median": 50.37725683399185,
+              "min": 50.37725683399185,
+              "max": 50.37725683399185
+            },
+            "cpu_seconds": {
+              "median": 49.17406,
+              "min": 49.17406,
+              "max": 49.17406
+            },
+            "peak_rss_mib": {
+              "median": 4761.96875,
+              "min": 4761.96875,
+              "max": 4761.96875
+            },
+            "stage_medians": {
+              "attributes": 30.595317583007272,
+              "convert": 3.0611445410031592,
+              "other": 13.332102001979365,
+              "read": 3.388692708002054
+            },
+            "checksum": "86a89f01e15e261a4d93558f429f843ea5bfe7f7b9ab862b7b7cbc59df8d6988"
+          },
+          {
+            "pandas": "3.0.6",
+            "input": "Homo_sapiens.GRCh38.114.gtf.gz",
+            "release": "optimized",
+            "runs": 1,
+            "versions": {
+              "gtfparse": "3.0.1",
+              "numpy": "2.5.3",
+              "pandas": "3.0.6",
+              "polars": "1.44.2",
+              "pyarrow": "25.0.1",
+              "python": "3.12.6"
+            },
+            "wall_seconds": {
+              "median": 36.632469124990166,
+              "min": 36.632469124990166,
+              "max": 36.632469124990166
+            },
+            "cpu_seconds": {
+              "median": 35.212921,
+              "min": 35.212921,
+              "max": 35.212921
+            },
+            "peak_rss_mib": {
+              "median": 4345.953125,
+              "min": 4345.953125,
+              "max": 4345.953125
+            },
+            "stage_medians": {
+              "attributes": 22.29295879100391,
+              "convert": 3.045151915997849,
+              "other": 7.17012666798837,
+              "read": 4.124231750000035
+            },
+            "checksum": "86a89f01e15e261a4d93558f429f843ea5bfe7f7b9ab862b7b7cbc59df8d6988"
+          },
+          {
+            "pandas": "3.0.6",
+            "input": "gencode.v48.primary_assembly.annotation.gtf.gz",
+            "release": "2.9.1-polars",
+            "runs": 1,
+            "versions": {
+              "gtfparse": "2.9.1",
+              "numpy": "2.5.3",
+              "pandas": "3.0.6",
+              "polars": "1.44.2",
+              "pyarrow": "25.0.1",
+              "python": "3.12.6"
+            },
+            "wall_seconds": {
+              "median": 38.069211957990774,
+              "min": 38.069211957990774,
+              "max": 38.069211957990774
+            },
+            "cpu_seconds": {
+              "median": 38.543923,
+              "min": 38.543923,
+              "max": 38.543923
+            },
+            "peak_rss_mib": {
+              "median": 4241.171875,
+              "min": 4241.171875,
+              "max": 4241.171875
+            },
+            "stage_medians": {
+              "attributes": 29.99933870800305,
+              "convert": 0.44022654100263026,
+              "other": 6.081221167973126,
+              "read": 1.5484255410119658
+            },
+            "checksum": "52d9ca5d36dd0795856c27637e70e93cff34511b12226af66a07b4eb114d3d49"
+          },
+          {
+            "pandas": "3.0.6",
+            "input": "gencode.v48.primary_assembly.annotation.gtf.gz",
+            "release": "3.0.0-arrow",
+            "runs": 1,
+            "versions": {
+              "gtfparse": "3.0.0",
+              "numpy": "2.5.3",
+              "pandas": "3.0.6",
+              "polars": "1.44.2",
+              "pyarrow": "25.0.1",
+              "python": "3.12.6"
+            },
+            "wall_seconds": {
+              "median": 52.19000641599996,
+              "min": 52.19000641599996,
+              "max": 52.19000641599996
+            },
+            "cpu_seconds": {
+              "median": 49.825912,
+              "min": 49.825912,
+              "max": 49.825912
+            },
+            "peak_rss_mib": {
+              "median": 3982.234375,
+              "min": 3982.234375,
+              "max": 3982.234375
+            },
+            "stage_medians": {
+              "attributes": 31.662579041003482,
+              "convert": 0.0013554580073105171,
+              "other": 17.05685912599438,
+              "read": 3.4692127909947885
+            },
+            "checksum": "52d9ca5d36dd0795856c27637e70e93cff34511b12226af66a07b4eb114d3d49"
+          },
+          {
+            "pandas": "3.0.6",
+            "input": "gencode.v48.primary_assembly.annotation.gtf.gz",
+            "release": "optimized",
+            "runs": 1,
+            "versions": {
+              "gtfparse": "3.0.1",
+              "numpy": "2.5.3",
+              "pandas": "3.0.6",
+              "polars": "1.44.2",
+              "pyarrow": "25.0.1",
+              "python": "3.12.6"
+            },
+            "wall_seconds": {
+              "median": 36.182878458013874,
+              "min": 36.182878458013874,
+              "max": 36.182878458013874
+            },
+            "cpu_seconds": {
+              "median": 34.28559,
+              "min": 34.28559,
+              "max": 34.28559
+            },
+            "peak_rss_mib": {
+              "median": 5319.375,
+              "min": 5319.375,
+              "max": 5319.375
+            },
+            "stage_medians": {
+              "attributes": 24.223106332996394,
+              "convert": 6.958399899303913e-05,
+              "other": 8.451739916024962,
+              "read": 3.5079626249935245
+            },
+            "checksum": "52d9ca5d36dd0795856c27637e70e93cff34511b12226af66a07b4eb114d3d49"
+          }
+        ],
+        "measurements": [
+          {
+            "backend": "production",
+            "checksum": "86a89f01e15e261a4d93558f429f843ea5bfe7f7b9ab862b7b7cbc59df8d6988",
+            "columns": [
+              "seqname",
+              "source",
+              "feature",
+              "start",
+              "end",
+              "score",
+              "strand",
+              "frame",
+              "gene_id",
+              "gene_version",
+              "gene_name",
+              "gene_source",
+              "gene_biotype",
+              "transcript_id",
+              "transcript_version",
+              "transcript_name",
+              "transcript_source",
+              "transcript_biotype",
+              "tag",
+              "transcript_support_level",
+              "exon_number",
+              "exon_id",
+              "exon_version",
+              "protein_id",
+              "protein_version",
+              "ccds_id"
+            ],
+            "cpu_count": 10,
+            "cpu_seconds": 39.141783,
+            "features": null,
+            "input_bytes": 64228966,
+            "path": "Homo_sapiens.GRCh38.114.gtf.gz",
+            "peak_rss_mib": 5904.65625,
+            "platform": "macOS-26.5-arm64-arm-64bit",
+            "result_type": "pandas",
+            "rows": 4116048,
+            "stages": {
+              "attributes": 29.537428333002026,
+              "convert": 3.2695007080037612,
+              "other": 4.044133210001746,
+              "read": 1.4144709579995833
+            },
+            "threads": 4,
+            "usecols": null,
+            "versions": {
+              "gtfparse": "2.9.1",
+              "numpy": "2.5.3",
+              "pandas": "3.0.6",
+              "polars": "1.44.2",
+              "pyarrow": "25.0.1",
+              "python": "3.12.6"
+            },
+            "wall_seconds": 38.26553320900712,
+            "release": "2.9.1-polars",
+            "repeat": 1
+          },
+          {
+            "backend": "production",
+            "checksum": "86a89f01e15e261a4d93558f429f843ea5bfe7f7b9ab862b7b7cbc59df8d6988",
+            "columns": [
+              "seqname",
+              "source",
+              "feature",
+              "start",
+              "end",
+              "score",
+              "strand",
+              "frame",
+              "gene_id",
+              "gene_version",
+              "gene_name",
+              "gene_source",
+              "gene_biotype",
+              "transcript_id",
+              "transcript_version",
+              "transcript_name",
+              "transcript_source",
+              "transcript_biotype",
+              "tag",
+              "transcript_support_level",
+              "exon_number",
+              "exon_id",
+              "exon_version",
+              "protein_id",
+              "protein_version",
+              "ccds_id"
+            ],
+            "cpu_count": 10,
+            "cpu_seconds": 49.17406,
+            "features": null,
+            "input_bytes": 64228966,
+            "path": "Homo_sapiens.GRCh38.114.gtf.gz",
+            "peak_rss_mib": 4761.96875,
+            "platform": "macOS-26.5-arm64-arm-64bit",
+            "result_type": "pandas",
+            "rows": 4116048,
+            "stages": {
+              "attributes": 30.595317583007272,
+              "convert": 3.0611445410031592,
+              "other": 13.332102001979365,
+              "read": 3.388692708002054
+            },
+            "threads": 4,
+            "usecols": null,
+            "versions": {
+              "gtfparse": "3.0.0",
+              "numpy": "2.5.3",
+              "pandas": "3.0.6",
+              "polars": "1.44.2",
+              "pyarrow": "25.0.1",
+              "python": "3.12.6"
+            },
+            "wall_seconds": 50.37725683399185,
+            "release": "3.0.0-arrow",
+            "repeat": 1
+          },
+          {
+            "backend": "production",
+            "checksum": "86a89f01e15e261a4d93558f429f843ea5bfe7f7b9ab862b7b7cbc59df8d6988",
+            "columns": [
+              "seqname",
+              "source",
+              "feature",
+              "start",
+              "end",
+              "score",
+              "strand",
+              "frame",
+              "gene_id",
+              "gene_version",
+              "gene_name",
+              "gene_source",
+              "gene_biotype",
+              "transcript_id",
+              "transcript_version",
+              "transcript_name",
+              "transcript_source",
+              "transcript_biotype",
+              "tag",
+              "transcript_support_level",
+              "exon_number",
+              "exon_id",
+              "exon_version",
+              "protein_id",
+              "protein_version",
+              "ccds_id"
+            ],
+            "cpu_count": 10,
+            "cpu_seconds": 35.212921,
+            "features": null,
+            "input_bytes": 64228966,
+            "path": "Homo_sapiens.GRCh38.114.gtf.gz",
+            "peak_rss_mib": 4345.953125,
+            "platform": "macOS-26.5-arm64-arm-64bit",
+            "result_type": "pandas",
+            "rows": 4116048,
+            "stages": {
+              "attributes": 22.29295879100391,
+              "convert": 3.045151915997849,
+              "other": 7.17012666798837,
+              "read": 4.124231750000035
+            },
+            "threads": 4,
+            "usecols": null,
+            "versions": {
+              "gtfparse": "3.0.1",
+              "numpy": "2.5.3",
+              "pandas": "3.0.6",
+              "polars": "1.44.2",
+              "pyarrow": "25.0.1",
+              "python": "3.12.6"
+            },
+            "wall_seconds": 36.632469124990166,
+            "release": "optimized",
+            "repeat": 1
+          },
+          {
+            "backend": "production",
+            "checksum": "52d9ca5d36dd0795856c27637e70e93cff34511b12226af66a07b4eb114d3d49",
+            "columns": [
+              "seqname",
+              "source",
+              "feature",
+              "start",
+              "end",
+              "score",
+              "strand",
+              "frame",
+              "gene_id",
+              "gene_type",
+              "gene_name",
+              "level",
+              "tag",
+              "transcript_id",
+              "transcript_type",
+              "transcript_name",
+              "exon_number",
+              "exon_id",
+              "transcript_support_level",
+              "havana_transcript",
+              "hgnc_id",
+              "havana_gene",
+              "ont",
+              "protein_id",
+              "ccdsid",
+              "artif_dupl"
+            ],
+            "cpu_count": 10,
+            "cpu_seconds": 38.543923,
+            "features": null,
+            "input_bytes": 59094311,
+            "path": "gencode.v48.primary_assembly.annotation.gtf.gz",
+            "peak_rss_mib": 4241.171875,
+            "platform": "macOS-26.5-arm64-arm-64bit",
+            "result_type": "pandas",
+            "rows": 4119244,
+            "stages": {
+              "attributes": 29.99933870800305,
+              "convert": 0.44022654100263026,
+              "other": 6.081221167973126,
+              "read": 1.5484255410119658
+            },
+            "threads": 4,
+            "usecols": null,
+            "versions": {
+              "gtfparse": "2.9.1",
+              "numpy": "2.5.3",
+              "pandas": "3.0.6",
+              "polars": "1.44.2",
+              "pyarrow": "25.0.1",
+              "python": "3.12.6"
+            },
+            "wall_seconds": 38.069211957990774,
+            "release": "2.9.1-polars",
+            "repeat": 1
+          },
+          {
+            "backend": "production",
+            "checksum": "52d9ca5d36dd0795856c27637e70e93cff34511b12226af66a07b4eb114d3d49",
+            "columns": [
+              "seqname",
+              "source",
+              "feature",
+              "start",
+              "end",
+              "score",
+              "strand",
+              "frame",
+              "gene_id",
+              "gene_type",
+              "gene_name",
+              "level",
+              "tag",
+              "transcript_id",
+              "transcript_type",
+              "transcript_name",
+              "exon_number",
+              "exon_id",
+              "transcript_support_level",
+              "havana_transcript",
+              "hgnc_id",
+              "havana_gene",
+              "ont",
+              "protein_id",
+              "ccdsid",
+              "artif_dupl"
+            ],
+            "cpu_count": 10,
+            "cpu_seconds": 49.825912,
+            "features": null,
+            "input_bytes": 59094311,
+            "path": "gencode.v48.primary_assembly.annotation.gtf.gz",
+            "peak_rss_mib": 3982.234375,
+            "platform": "macOS-26.5-arm64-arm-64bit",
+            "result_type": "pandas",
+            "rows": 4119244,
+            "stages": {
+              "attributes": 31.662579041003482,
+              "convert": 0.0013554580073105171,
+              "other": 17.05685912599438,
+              "read": 3.4692127909947885
+            },
+            "threads": 4,
+            "usecols": null,
+            "versions": {
+              "gtfparse": "3.0.0",
+              "numpy": "2.5.3",
+              "pandas": "3.0.6",
+              "polars": "1.44.2",
+              "pyarrow": "25.0.1",
+              "python": "3.12.6"
+            },
+            "wall_seconds": 52.19000641599996,
+            "release": "3.0.0-arrow",
+            "repeat": 1
+          },
+          {
+            "backend": "production",
+            "checksum": "52d9ca5d36dd0795856c27637e70e93cff34511b12226af66a07b4eb114d3d49",
+            "columns": [
+              "seqname",
+              "source",
+              "feature",
+              "start",
+              "end",
+              "score",
+              "strand",
+              "frame",
+              "gene_id",
+              "gene_type",
+              "gene_name",
+              "level",
+              "tag",
+              "transcript_id",
+              "transcript_type",
+              "transcript_name",
+              "exon_number",
+              "exon_id",
+              "transcript_support_level",
+              "havana_transcript",
+              "hgnc_id",
+              "havana_gene",
+              "ont",
+              "protein_id",
+              "ccdsid",
+              "artif_dupl"
+            ],
+            "cpu_count": 10,
+            "cpu_seconds": 34.28559,
+            "features": null,
+            "input_bytes": 59094311,
+            "path": "gencode.v48.primary_assembly.annotation.gtf.gz",
+            "peak_rss_mib": 5319.375,
+            "platform": "macOS-26.5-arm64-arm-64bit",
+            "result_type": "pandas",
+            "rows": 4119244,
+            "stages": {
+              "attributes": 24.223106332996394,
+              "convert": 6.958399899303913e-05,
+              "other": 8.451739916024962,
+              "read": 3.5079626249935245
+            },
+            "threads": 4,
+            "usecols": null,
+            "versions": {
+              "gtfparse": "3.0.1",
+              "numpy": "2.5.3",
+              "pandas": "3.0.6",
+              "polars": "1.44.2",
+              "pyarrow": "25.0.1",
+              "python": "3.12.6"
+            },
+            "wall_seconds": 36.182878458013874,
+            "release": "optimized",
+            "repeat": 1
+          }
+        ]
+      },
+      "gencode_repeat": {
+        "backend": "production",
+        "checksum": "52d9ca5d36dd0795856c27637e70e93cff34511b12226af66a07b4eb114d3d49",
+        "columns": [
+          "seqname",
+          "source",
+          "feature",
+          "start",
+          "end",
+          "score",
+          "strand",
+          "frame",
+          "gene_id",
+          "gene_type",
+          "gene_name",
+          "level",
+          "tag",
+          "transcript_id",
+          "transcript_type",
+          "transcript_name",
+          "exon_number",
+          "exon_id",
+          "transcript_support_level",
+          "havana_transcript",
+          "hgnc_id",
+          "havana_gene",
+          "ont",
+          "protein_id",
+          "ccdsid",
+          "artif_dupl"
+        ],
+        "cpu_count": 10,
+        "cpu_seconds": 31.624218999999997,
+        "features": null,
+        "input_bytes": 59094311,
+        "path": "/private/tmp/gtfparse-polars-migration/gencode.v48.primary_assembly.annotation.gtf.gz",
+        "peak_rss_mib": 5973.859375,
+        "platform": "macOS-26.5-arm64-arm-64bit",
+        "result_type": "pandas",
+        "rows": 4119244,
+        "stages": {
+          "attributes": 21.284394458009047,
+          "convert": 6.145800580270588e-05,
+          "other": 6.3738490419782465,
+          "read": 3.6806942080002045
+        },
+        "threads": 4,
+        "usecols": null,
+        "versions": {
+          "gtfparse": "3.0.1",
+          "numpy": "2.5.3",
+          "pandas": "3.0.6",
+          "polars": "1.44.2",
+          "pyarrow": "25.0.1",
+          "python": "3.12.6"
+        },
+        "wall_seconds": 31.3389991659933
+      }
+    },
+    "100000_row_batches": {
+      "status": "initial memory diagnostic; final implementation uses 250000 rows; excluded from final tables",
+      "gencode_measurement": {
+        "backend": "production",
+        "checksum": "52d9ca5d36dd0795856c27637e70e93cff34511b12226af66a07b4eb114d3d49",
+        "columns": [
+          "seqname",
+          "source",
+          "feature",
+          "start",
+          "end",
+          "score",
+          "strand",
+          "frame",
+          "gene_id",
+          "gene_type",
+          "gene_name",
+          "level",
+          "tag",
+          "transcript_id",
+          "transcript_type",
+          "transcript_name",
+          "exon_number",
+          "exon_id",
+          "transcript_support_level",
+          "havana_transcript",
+          "hgnc_id",
+          "havana_gene",
+          "ont",
+          "protein_id",
+          "ccdsid",
+          "artif_dupl"
+        ],
+        "cpu_count": 10,
+        "cpu_seconds": 32.886145,
+        "features": null,
+        "input_bytes": 59094311,
+        "path": "/private/tmp/gtfparse-polars-migration/gencode.v48.primary_assembly.annotation.gtf.gz",
+        "peak_rss_mib": 2322.4375,
+        "platform": "macOS-26.5-arm64-arm-64bit",
+        "result_type": "pandas",
+        "rows": 4119244,
+        "stages": {
+          "attributes": 34.13356716699491,
+          "convert": 3.616600588429719e-05,
+          "other": 0.03949662500235718,
+          "read": 5.900347957998747
+        },
+        "threads": 4,
+        "usecols": null,
+        "versions": {
+          "gtfparse": "3.0.1",
+          "numpy": "2.5.3",
+          "pandas": "3.0.6",
+          "polars": "1.44.2",
+          "pyarrow": "25.0.1",
+          "python": "3.12.6"
+        },
+        "wall_seconds": 40.073447916001896
+      }
+    }
+  }
+}
diff --git a/benchmarks/optimization.md b/benchmarks/optimization.md
new file mode 100644
index 0000000..f4a4370
--- /dev/null
+++ b/benchmarks/optimization.md
@@ -0,0 +1,147 @@
+# Complete-load optimization in 3.0.1
+
+Version 3.0.1 retains the pandas default and optional Polars output introduced
+in 3.0.0. It improves the shared Python attribute parser and the construction
+of expanded pandas columns while keeping the supported reader behavior.
+
+## Changes
+
+- Capture attribute names and double-quoted values directly in the compiled
+  regex. Other quoted/unquoted values and pre-split sequences keep their
+  permissive cleanup and custom quote handling. Avoid splitting and slicing
+  every ordinary double-quoted pair and looking up each name twice.
+- Convert raw attributes with missing entries mapped to None. For configured
+  Arrow strings, expand and construct columns in batches of 250,000 rows, then
+  retain compact Arrow chunks. Release each batch and temporary expanded list
+  after conversion. Fill missing prefixes/gaps with empty strings and preserve
+  first-seen column order, fixed-field collisions and global progress counts.
+  This bounds temporary Python values for large files, addressing #91. Other
+  string representations keep their compatible full-list construction.
+- When pandas' configured inferred string dtype uses Arrow storage, construct
+  the Arrow string array directly and wrap it in that same pandas dtype. This
+  avoids an intermediate object-array inference pass. Other configured string
+  representations retain their ordinary pandas construction behavior.
+
+Column order, collisions with fixed fields, repeated and empty attributes,
+custom quote/missing-value parameters, callbacks/cancellation, aliases, casts,
+raw mode, filtering, gzip and caller-owned streams keep their previous behavior.
+No process-wide pandas settings change. Numeric/categorical fixed-field reading
+and the nine-field validation remain the same.
+
+## Reproduction and method
+
+Use the same benchmark interpreter/dependency versions for all release roots.
+The recorded subsets use Python 3.12.6, pandas 2.3.3 or 3.0.6, NumPy 2.5.3,
+Arrow 25.0.1 and optional Polars 1.44.2 with four threads. Extract the old package
+sources into isolated directories, without changing their dependency versions:
+
+```sh
+mkdir -p /tmp/gtfparse-2.9.1 /tmp/gtfparse-3.0.0
+git archive 3e0912c gtfparse | tar -x -C /tmp/gtfparse-2.9.1
+git archive be449cb gtfparse | tar -x -C /tmp/gtfparse-3.0.0
+python benchmarks/compare_releases.py \
+  ensembl-250k.gtf ensembl-250k.gtf.gz gencode-250k.gtf gencode-250k.gtf.gz \
+  --baseline 2.9.1-polars /tmp/gtfparse-2.9.1 \
+  --baseline 3.0.0-arrow /tmp/gtfparse-3.0.0 \
+  --python /path/to/pandas-2.3-env/bin/python \
+  --python /path/to/pandas-3.0-env/bin/python \
+  --repeats 3 --threads 4 --output comparisons.json
+```
+
+Install matching dependencies, including optional Polars for the historical
+reader, in both environments. Each child starts a fresh process and runs the
+complete production `read_gtf`; processes execute sequentially. Reader order
+rotates and corpus order reverses between repetitions. Imports are outside the
+reported load time. Peak RSS includes imported libraries and is captured before
+content hashing. No OS-cache flushing or machine isolation is claimed.
+
+The first 250,000 annotation rows and original header of Ensembl release 114
+human GRCh38 and GENCODE v48 human primary assembly are the same inputs used in
+[the migration benchmark](README.md#corpus-and-equivalence). Original URLs,
+selection rules and hashes are retained there and in results.json. Complete
+files provide separate scaling checks. Every run checks all column contents
+and order against the other releases; different representations are normalized
+only for hashing. Exact dtypes are checked separately against 3.0.0.
+
+Use `--features exon --usecols seqname start end gene_id transcript_id` for
+selective-load comparisons and `--result-type polars` for optional output.
+The JSON records contain versions, content hashes, dimensions, stage times,
+median/min/max and pre-validation peak RSS. A JSON-lines sidecar is flushed
+on each successful read, so failed experiments retain their completed records.
+Compare complete times; the old and current read-stage callbacks cover different
+conversion work and do not establish a pure CSV-engine speed ratio.
+
+## Results
+
+The table contains three-run medians of complete loads with equivalent pandas
+output. All 72 runs match content hashes and column order across releases and
+pandas versions. Times and memory are fresh measurements from this sweep;
+compare readers within this table rather than mixing earlier timings from a
+shared workstation.
+
+| pandas | Input (250k rows) | 2.9.1 Polars s | 3.0.0 Arrow s | 3.0.1 s | Polars MiB | 3.0.0 MiB | 3.0.1 MiB |
+| --- | --- | ---: | ---: | ---: | ---: | ---: | ---: |
+| 2.3.3 | ensembl-250k.gtf | 2.57 | 2.91 | 2.29 | 560 | 617 | 583 |
+| 2.3.3 | ensembl-250k.gtf.gz | 2.86 | 4.58 | 2.36 | 561 | 620 | 584 |
+| 2.3.3 | gencode-250k.gtf | 3.18 | 2.59 | 3.72 | 567 | 588 | 550 |
+| 2.3.3 | gencode-250k.gtf.gz | 3.61 | 3.90 | 3.74 | 568 | 584 | 558 |
+| 3.0.6 | ensembl-250k.gtf | 2.54 | 2.82 | 1.91 | 552 | 600 | 577 |
+| 3.0.6 | ensembl-250k.gtf.gz | 2.35 | 2.80 | 1.88 | 553 | 600 | 574 |
+| 3.0.6 | gencode-250k.gtf | 2.05 | 2.68 | 1.71 | 560 | 593 | 577 |
+| 3.0.6 | gencode-250k.gtf.gz | 2.10 | 2.52 | 1.73 | 560 | 595 | 574 |
+
+On pandas 3.0.6, the optimized reader takes 17–25% less time than the Polars
+baseline and 31–36% less time than 3.0.0 across all four cases. On pandas 2.3.3,
+the Ensembl medians improve, while GENCODE wall times are mixed and highly
+variable (individual complete reads span about 2–5.6 seconds). All retained
+CPU-time medians improve, but that does not establish a wall-time speedup in
+those noisy cases. No runs are discarded. Subset peak RSS is 3–6% lower than
+3.0.0 and ranges from about 3% lower to 4% higher than Polars.
+
+The repeated corpus is a prefix of each annotation, so feature/attribute
+distributions can differ from complete files. This demonstrates results for
+gtfparse's complete pipeline, rather than a universal pandas-versus-Polars
+engine comparison. Batched assembly bounds temporary Python attribute values
+when pandas uses Arrow strings; it does not make the entire reader streaming.
+Python/object string storage remains supported without this memory reduction.
+
+The full-file checks use pandas 3.0.6 and the same versions/thread count, with
+one fresh process per reader. All six complete loads match content and column
+order. These single observations demonstrate scaling and correctness, and do
+not establish repeated full-file throughput.
+
+| Full gzip input | Rows | Polars s | 3.0.0 s | 3.0.1 s | Polars MiB | 3.0.0 MiB | 3.0.1 MiB |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| Ensembl GRCh38.114 | 4,116,048 | 47.11 | 67.63 | 36.86 | 6504 | 7199 | 3639 |
+| GENCODE v48 primary | 4,119,244 | 43.74 | 52.35 | 29.88 | 7091 | 6560 | 3698 |
+
+Here the optimized reader takes 22–32% less time and 44–48% less peak RSS than
+Polars. Compared with 3.0.0, it takes 43–45% less time and 44–49% less peak RSS.
+The shared workstation was running other workloads; RSS and wall times vary
+between sweeps. Compare the matched observations above, not runs from different
+measurement windows. The benefit applies to configured Arrow string storage.
+
+Selective reads and explicit Polars output match all contents in separate
+one-pass comparisons; these checks do not establish repeated throughput for
+those configurations. [optimization-results.json](optimization-results.json)
+retains every final record, version, input hash, stage median and range.
+The rejected unbatched full-file candidate is retained separately: its full
+GENCODE peak RSS reached 5.2 GiB and 5.8 GiB on a repeat. That small-sample
+optimization was replaced with bounded temporary attribute assembly before
+this release. Its timings are excluded from the final result tables.
+
+## Correctness and compatibility
+
+The differential tests freeze the pre-optimization tokenization and compare
+quoted/unquoted, custom quote delimiters, Unicode/whitespace, malformed fragments,
+repeated/empty values, pre-split rows, missing-value sentinels and column
+selection. Additional tests check configured Python/Arrow string storage, batch
+boundaries, late-discovered columns, missing batches, fixed-field attribute
+collisions, first-seen order and global progress counts. Existing tests cover streams, null fields, exact
+integer coordinates, comments/BOM, aliases, converters, progress/cancellation,
+optional Polars, dictionary output and writer round trips.
+
+The verification sweep compares 71 full API/dtype combinations on each pandas
+version against 3.0.0, including all five real fixtures and configured string
+storage. CI exercises Python 3.9+, minimum/current pandas and Arrow, the base
+installation without Polars, optional Polars versions and packaged source tests.
diff --git a/gtfparse/attribute_parsing.py b/gtfparse/attribute_parsing.py
index 495156c..e60dbfc 100644
--- a/gtfparse/attribute_parsing.py
+++ b/gtfparse/attribute_parsing.py
@@ -13,7 +13,6 @@
 import logging
 import re
 from collections import OrderedDict
-from sys import intern
 
 logger = logging.getLogger(__name__)
 
@@ -21,7 +20,9 @@
 # re module so the low-level split column agrees with
 # attribute expansion. Unquoted values and single quotes are accepted as well.
 ATTRIBUTE_PATTERN = r"""[^\s;]+[ \t]+(?:"[^"]*"|'[^']*'|[^;]+)"""
-_ATTRIBUTE_PAIRS = re.compile(ATTRIBUTE_PATTERN)
+# Capture double-quoted values without their delimiters. Other values retain
+# their original quoting for custom quote_char and pre-split input support.
+_ATTRIBUTE_PAIRS = re.compile(r"""([^\s;]+)[ \t]+(?:"([^"]*)"|('[^']*'|[^;]+))""")
 
 
 def expand_attribute_strings(
@@ -61,61 +62,51 @@ def expand_attribute_strings(
     n = len(attribute_strings)
 
     extra_columns = {}
-    column_order = []
-
-    # While parsing millions of repeated strings (e.g. "gene_id" and "TP53"),
-    # we can save a lot of memory by making sure there's only one string
-    # object per unique column name. Cache interned names locally as well.
-    column_interned_strings = {}
 
     if progress_callback is not None:
         progress_callback("attributes", 0, n)
 
     for i, kv_strings in enumerate(attribute_strings):
         if isinstance(kv_strings, str):
-            kv_strings = _ATTRIBUTE_PAIRS.findall(kv_strings)
+            pairs = _ATTRIBUTE_PAIRS.findall(kv_strings)
         elif kv_strings is None:
-            kv_strings = ()
-        for kv in kv_strings:
-            # Split once: values such as transcript_support_level may include
-            # spaces, and whitespace between a key and its value may vary.
-            parts = kv.strip().split(None, 1)
-
-            if len(parts) != 2:
-                continue
-
-            column_name, value = parts
-
-            try:
-                column_name = column_interned_strings[column_name]
-            except KeyError:
-                column_name = intern(column_name)
-                column_interned_strings[column_name] = column_name
-
+            pairs = ()
+        else:
+            # Already separated pairs retain their permissive split behavior.
+            pairs = (
+                (parts[0], "", parts[1])
+                for kv in kv_strings
+                if len(parts := kv.strip().split(None, 1)) == 2
+            )
+        for column_name, double_quoted, other in pairs:
             if usecols is not None and column_name not in usecols:
                 continue
 
-            if len(value) >= 2 and value[0] in ('"', quote_char) and value[-1] == value[0]:
-                value = value[1:-1]
-
-            try:
-                column = extra_columns[column_name]
-                # if an attribute is used repeatedly then
-                # keep track of all its values in a list
-                old_value = column[i]
-                if old_value is missing_value:
-                    column[i] = value
-                else:
-                    column[i] = "%s,%s" % (old_value, value)
-            except KeyError:
+            if other:
+                value = other.strip()
+                if not value:
+                    continue
+                if len(value) >= 2 and value[0] in ('"', quote_char) and value[-1] == value[0]:
+                    value = value[1:-1]
+            else:
+                value = double_quoted
+
+            column = extra_columns.get(column_name)
+            if column is None:
                 column = [missing_value] * n
-                column[i] = value
                 extra_columns[column_name] = column
-                column_order.append(column_name)
-
-        completed = i + 1
-        if progress_callback is not None and (completed % 10_000 == 0 or completed == n):
-            progress_callback("attributes", completed, n)
-
-    logger.info("Extracted GTF attributes: %s", column_order)
-    return OrderedDict((column_name, extra_columns[column_name]) for column_name in column_order)
+                column[i] = value
+            elif column[i] is missing_value:
+                column[i] = value
+            else:
+                # Preserve repeated attributes and the missing-value identity
+                # convention, including repeated empty quoted values.
+                column[i] = "%s,%s" % (column[i], value)
+
+        if progress_callback is not None:
+            completed = i + 1
+            if completed % 10_000 == 0 or completed == n:
+                progress_callback("attributes", completed, n)
+
+    logger.info("Extracted GTF attributes: %s", list(extra_columns))
+    return OrderedDict(extra_columns)
diff --git a/gtfparse/read_gtf.py b/gtfparse/read_gtf.py
index 42b1913..9dc4f0d 100644
--- a/gtfparse/read_gtf.py
+++ b/gtfparse/read_gtf.py
@@ -14,6 +14,7 @@
 from os.path import exists
 
 import pandas as pd
+import pyarrow as pa
 
 from ._polars import require_polars
 from ._read_csv import REQUIRED_COLUMNS as REQUIRED_COLUMNS
@@ -147,6 +148,47 @@ def parse_gtf_pandas(*args, **kwargs):
     return result
 
 
+_ATTRIBUTE_BATCH_SIZE = 250_000
+
+
+def _expand_attributes_as_arrow(attributes, usecols, progress_callback):
+    """Keep temporary Python values bounded while retaining Arrow column chunks."""
+    total = len(attributes)
+    columns = {}
+    string_type = pa.large_string()
+    empty = pa.scalar("", type=string_type)
+    if progress_callback is not None:
+        progress_callback("attributes", 0, total)
+
+    def report(stage, completed, _total):
+        absolute = start + completed
+        if completed and (absolute % 10_000 == 0 or absolute == total):
+            progress_callback(stage, absolute, total)
+
+    for start in range(0, total, _ATTRIBUTE_BATCH_SIZE):
+        stop = min(start + _ATTRIBUTE_BATCH_SIZE, total)
+        values = attributes.iloc[start:stop].to_numpy(dtype=object, na_value=None)
+        expanded = expand_attribute_strings(
+            values,
+            usecols=usecols,
+            progress_callback=report if progress_callback is not None else None,
+        )
+        del values
+        missing = columns.keys() - expanded.keys()
+        if missing:
+            blank = pa.repeat(empty, stop - start)
+            for name in missing:
+                columns[name].append(blank)
+        prefix = None
+        for name in list(expanded):
+            if name not in columns:
+                if start and prefix is None:
+                    prefix = pa.repeat(empty, start)
+                columns[name] = [prefix] if start else []
+            columns[name].append(pa.array(expanded.pop(name), type=string_type))
+    return {name: pa.chunked_array(chunks, type=string_type) for name, chunks in columns.items()}
+
+
 def parse_gtf_and_expand_attributes(
     filepath_or_buffer, restrict_attribute_columns=None, features=None, *, progress_callback=None
 ):
@@ -181,15 +223,26 @@ def parse_gtf_and_expand_attributes(
     elif restrict_attribute_columns:
         restrict_attribute_columns = set(restrict_attribute_columns)
     attributes = df.pop("attribute")
-    attributes = attributes.astype(object).where(attributes.notna(), None)
-    expanded = expand_attribute_strings(
-        attributes,
-        usecols=restrict_attribute_columns,
-        progress_callback=progress_callback,
-    )
+    string_dtype = pd.Series([""]).dtype
+    arrow_strings = isinstance(string_dtype, pd.StringDtype) and string_dtype.storage == "pyarrow"
+    if arrow_strings:
+        expanded = _expand_attributes_as_arrow(
+            attributes, restrict_attribute_columns, progress_callback
+        )
+    else:
+        expanded = expand_attribute_strings(
+            attributes.to_numpy(dtype=object, na_value=None),
+            usecols=restrict_attribute_columns,
+            progress_callback=progress_callback,
+        )
+    # Release raw strings before assembling the final pandas columns.
+    del attributes
     # Assigning a colliding attribute replaces its fixed column, matching the
     # previous reader, while preserving the fixed columns' original positions.
-    for name, values in expanded.items():
+    for name in list(expanded):
+        values = expanded.pop(name)
+        if arrow_strings:
+            values = pd.array(values, dtype=string_dtype)
         df[name] = values
     return df
 
diff --git a/tasks/todo.md b/tasks/todo.md
index 33a153d..f5622e8 100644
--- a/tasks/todo.md
+++ b/tasks/todo.md
@@ -379,3 +379,109 @@ into a fresh base environment loads from site-packages, installs no Polars,
 and round-trips all five real fixtures. Archive runner mismatch fixed as #88;
 shipped lint/tests pass with python -m pytest. Release status and published
 artifact verification will be recorded on the PR to keep clean master clean.
+
+# Follow-up: optimize general GTF loading beyond the Polars baseline
+
+## Specification
+
+The user requested a PR to improve the pandas/Arrow implementation and attempt
+repeatable complete-load performance better than the old Polars implementation,
+without narrowing supported inputs or API behavior. Version this as 3.0.1.
+
+Preserve all expanded/raw GTF semantics, fixed numeric/categorical and inferred
+string dtypes, arbitrary/repeated/quoted attributes, column collisions/order,
+missing-value conventions, custom quote/missing-value parameters, pre-split
+attribute sequences, usecols/aliases/converters/version casts/biotype inference,
+optional Polars/dict output, comments/BOM/exact field validation, gzip/short or
+nonseekable caller-owned streams, callbacks and cancellation. Keep Python 3.9+,
+minimum dependencies and the base installation without Polars working. Do not
+change global pandas options or add a compiled/mandatory optional dependency.
+
+Profile complete reads and attribute expansion on actual Ensembl/GENCODE data.
+Evaluate captured attribute pairs, fewer Python allocations/dictionary lookups,
+and avoiding unnecessary raw-string conversion/copies; evaluate Arrow conversion
+settings only if they help complete-load time or peak memory without semantic
+changes. Keep the smallest robust implementation. Add equivalence tests for
+optimized parsing against the original regex/split behavior, including malformed
+and mixed quoting, empty/repeated fields, arbitrary whitespace and pre-split input.
+Document and file any newly discovered existing bugs rather than silently fixing
+behavior during an optimization.
+
+Preserve 3.0.0 and 2.9.1 baselines. Compare all three complete production pipelines
+in fresh sequential processes with four threads, interleaved/reversed ordering,
+three repeats, pandas 2.3.3 and 3.0.6, Ensembl/GENCODE 250k-row plain/gzip corpora.
+Check full-content hashes outside timing. Also run a full GENCODE scaling check;
+record actual corpus/version/ranges/RSS and distinguish repeated throughput from
+single full-file observations. Include optional output and selective-load checks.
+A failure to beat Polars on every case is an honest result, not permission to
+weaken input validation or generality. Publish a worthwhile measured improvement
+with its limits; do not claim pure CSV speed from differently scoped read stages.
+
+## Plan
+
+- [x] Create feature branch, read lessons/API/benchmarks, preserve 3.0.0 baseline.
+- [x] Profile dominant costs and measure equivalent candidate implementations.
+- [x] Implement the smallest measured improvements, preserving public behavior.
+- [x] Add differential parser/API regression coverage and review edge cases.
+- [x] Run lint/tests in normal, no-Polars, minimum and optional environments.
+- [x] Record repeated complete-reader comparisons and full-file scaling evidence.
+- [x] Update version/benchmark documentation and review final diff.
+- [ ] Open PR, pass checks, merge, deploy from clean master and verify PyPI.
+- [ ] Review remaining relevant issues and record the next candidate on the PR.
+
+## Review
+
+The optimized parser captures double-quoted values directly and removes the
+split/intern lookup work on ordinary raw attributes, while preserving the legacy
+fallback grammar and pre-split support. For configured Arrow strings, 250,000-row
+batches retain compact Arrow chunks and release temporary Python values. Missing
+prefixes/gaps, late columns, first-seen order, collisions and global progress are
+preserved. Python/object storage keeps compatible full-list construction.
+
+The frozen legacy parser oracle covers custom quotes/sentinels, arbitrary
+malformed fragments, repetitions, pre-split values and selection. 71 full API/dtype
+comparisons against 3.0.0 pass on each pandas version (142 total). Configured
+Python/Arrow storage, batch boundaries and fixed-field collisions pass. Final
+script, compatibility-environment and packaging results follow below.
+
+The final repeated subset sweep completes all 72 fresh-process reads with
+matching hashes. On pandas 3.0.6, median load time is 17–25% lower than 2.9.1
+Polars and 31–36% lower than 3.0.0. pandas 2.3.3 Ensembl medians improve, while
+GENCODE wall times are mixed and highly variable; all runs and ranges remain in
+the report. Subset peak RSS is 3–6% lower than 3.0.0 and about -3% to +4% relative
+to Polars. Full-file scaling and final selective/optional checks follow below.
+
+Memory replan: the unbatched candidate's full GENCODE RSS was higher than the
+3.0.0 observation, and a repeat remained high. Both existing 3.0.0 and that
+candidate materialized Python values for the entire file despite compact final
+Arrow strings. Filed #91 and replaced unbounded temporary attribute construction
+with batches, retaining the public expansion contract and configured string
+storage. Rejected full-file observations are retained separately, rather than
+mixed into final medians. The first 100,000-row-batch GENCODE check reduced peak
+RSS to 2.3 GiB and matched content; final comparisons use 250,000-row batches.
+
+Final release status will be recorded on the PR so clean master remains clean
+for ./deploy.sh.
+
+Final validation: ./lint.sh passes. Normal ./test.sh passes 213 tests with
+90% coverage. Base pandas 3.0.6/Arrow 25 and pandas 3.0.6/Arrow 18 pass 199 tests
+plus 10 optional skips (96% coverage). Minimum Python 3.9/pandas 2.2.2/Arrow 18
+passes 199 plus 10 skips (88%; configured object strings omit the Arrow branch).
+Minimum optional Polars 0.20.31 passes 212 plus one modern-only skip (97%, existing
+native categorical-remapping warning). Tests treat DeprecationWarning as errors.
+
+All 84 final benchmark runs match expected content/order. Full Ensembl (4,116,048
+rows) takes 36.86 s/3639 MiB versus Polars 47.11 s/6504 MiB and 3.0.0 67.63 s/7199
+MiB. Full GENCODE (4,119,244 rows) takes 29.88 s/3698 MiB versus Polars 43.74 s/7091
+MiB and 3.0.0 52.35 s/6560 MiB. These are single observations per reader on the
+shared workstation, distinct from the repeated subset medians. Final selective
+exon/five-column and optional Polars-output checks also match all three releases.
+84 final records and separately labeled development evidence are retained in
+benchmarks/optimization-results.json. Version is bumped to 3.0.1.
+
+Packaging review: wheel and source distribution build successfully, have identical
+3.0.1 dependency metadata, and pass strict Twine checks. The wheel excludes tests
+and benchmarks; the source archive includes the new regression tests, runner and
+retained evidence. Its shipped ./lint.sh and ./test.sh pass (199 plus 10 optional
+skips, 96% coverage). Production code, test oracle and benchmark method reviewed;
+release/checklist completion will be recorded on the PR after merge and PyPI.
diff --git a/tests/test_attribute_optimization.py b/tests/test_attribute_optimization.py
new file mode 100644
index 0000000..234c53e
--- /dev/null
+++ b/tests/test_attribute_optimization.py
@@ -0,0 +1,128 @@
+"""Differential coverage for the 3.0.0 attribute parser's permissive grammar."""
+
+import random
+import re
+from collections import OrderedDict
+from io import StringIO
+
+import pytest
+
+from gtfparse import expand_attribute_strings, read_gtf
+
+# Freeze the pre-optimization tokenization independently of the captured regex.
+LEGACY_PATTERN = re.compile(r"""[^\s;]+[ \t]+(?:"[^"]*"|'[^']*'|[^;]+)""")
+
+
+def legacy_expand(rows, quote_char, missing_value, usecols):
+    columns = OrderedDict()
+    for i, row in enumerate(rows):
+        pairs = LEGACY_PATTERN.findall(row) if isinstance(row, str) else row or ()
+        for pair in pairs:
+            parts = pair.strip().split(None, 1)
+            if len(parts) != 2:
+                continue
+            name, value = parts
+            if usecols is not None and name not in usecols:
+                continue
+            if len(value) >= 2 and value[0] in ('"', quote_char) and value[-1] == value[0]:
+                value = value[1:-1]
+            if name not in columns:
+                columns[name] = [missing_value] * len(rows)
+            previous = columns[name][i]
+            columns[name][i] = value if previous is missing_value else "%s,%s" % (previous, value)
+    return columns
+
+
+@pytest.mark.parametrize("quote_char", ["'", '"', "~", ""])
+@pytest.mark.parametrize("missing_value", ["", None, "missing"])
+@pytest.mark.parametrize("usecols", [None, set(), {"a", "tag", "gene_id"}])
+def test_optimized_expansion_matches_legacy_grammar(quote_char, missing_value, usecols):
+    rows = [
+        None,
+        "",
+        'gene_id ""; gene_id "G1"; tag "a; b"; tag "";',
+        "a '  padded  '; a '~custom~'; a ~custom~;",
+        'a "unterminated; b value; a "close"trailing;',
+        "a    ; b\t\t; a \u00a0 trimmed \r\n; malformed;",
+        "a \"é\n\t; quote'\"; a 'single \" quote';",
+        ["broken", 'a ""', 'a "next"', "gene_id\t'G1'", " a    padded  "],
+        [],
+    ]
+    rng = random.Random(85)
+    alphabet = "ab_ .;\t\n\r\"'~é\u00a0"
+    rows += ["".join(rng.choices(alphabet, k=rng.randrange(100))) for _ in range(500)]
+    expected = legacy_expand(rows, quote_char, missing_value, usecols)
+    actual = expand_attribute_strings(
+        rows, quote_char=quote_char, missing_value=missing_value, usecols=usecols
+    )
+    assert isinstance(actual, OrderedDict)
+    assert list(actual) == list(expected)
+    assert actual == expected
+
+
+def test_expanded_attribute_collisions_preserve_position_and_missing_values():
+    text = (
+        '1\ttest\tgene\t1\t10\t.\t+\t.\tgene_id "G1"; start "7"; note "a";\n'
+        '1\ttest\tgene\t11\t20\t.\t+\t.\tgene_id "G2"; note "b";\n'
+    )
+    frame = read_gtf(StringIO(text))
+    assert list(frame) == [
+        "seqname",
+        "source",
+        "feature",
+        "start",
+        "end",
+        "score",
+        "strand",
+        "frame",
+        "gene_id",
+        "note",
+    ]
+    assert frame["start"].tolist() == ["7", ""]
+    assert frame["note"].tolist() == ["a", "b"]
+
+
+@pytest.mark.parametrize("storage", ["python", "pyarrow"])
+def test_expanded_columns_follow_pandas_string_configuration(storage):
+    import pandas as pd
+
+    text = (
+        '1\ttest\tgene\t1\t10\t.\t+\t.\tgene_id "G1"; note "café; value";\n'
+        '1\ttest\tgene\t11\t20\t.\t+\t.\tgene_id "G2";\n'
+    )
+    with pd.option_context("mode.string_storage", storage):
+        frame = read_gtf(StringIO(text))
+        pd.testing.assert_series_equal(frame["gene_id"], pd.Series(["G1", "G2"], name="gene_id"))
+        pd.testing.assert_series_equal(frame["note"], pd.Series(["café; value", ""], name="note"))
+        assert pd.get_option("mode.string_storage") == storage
+
+
+@pytest.mark.parametrize("storage", ["python", "pyarrow"])
+def test_batched_expansion_preserves_late_columns_gaps_collisions_and_progress(
+    storage, monkeypatch
+):
+    import importlib
+
+    import pandas as pd
+
+    reader = importlib.import_module("gtfparse.read_gtf")
+    monkeypatch.setattr(reader, "_ATTRIBUTE_BATCH_SIZE", 10_000)
+    attrs = ['gene_id "G1"; tag "first"; tag "second";'] * 10_000
+    attrs += ['gene_id "G2"; note "middle; value";'] * 10_000
+    attrs += ['gene_id "G3"; start "42"; late "last";'] * 3
+    text = "".join("1\ttest\tgene\t1\t100\t.\t+\t.\t%s\n" % a for a in attrs)
+    events = []
+    with pd.option_context("mode.string_storage", storage):
+        frame = read_gtf(StringIO(text), progress_callback=lambda *event: events.append(event))
+    assert list(frame)[-4:] == ["gene_id", "tag", "note", "late"]
+    assert list(frame).index("start") == 3
+    assert frame["tag"].tolist() == ["first,second"] * 10_000 + [""] * 10_003
+    assert frame["note"].tolist() == [""] * 10_000 + ["middle; value"] * 10_000 + [""] * 3
+    assert frame["late"].tolist() == [""] * 20_000 + ["last"] * 3
+    assert frame["start"].tolist() == [""] * 20_000 + ["42"] * 3
+    assert [e for e in events if e[0] == "attributes"] == [
+        ("attributes", 0, 20_003),
+        ("attributes", 10_000, 20_003),
+        ("attributes", 20_000, 20_003),
+        ("attributes", 20_003, 20_003),
+    ]

From 22ca2df6d98743f5d92a0b6b3fc8a956105dd41a Mon Sep 17 00:00:00 2001
From: Alex Rubinsteyn <alex.rubinsteyn@gmail.com>
Date: Tue, 6 Oct 2026 13:38:41 -0400
Subject: [PATCH 2/3] Reserve 3.0.2 for the independent optimization release

---
 benchmarks/README.md                 |  2 +-
 benchmarks/optimization-results.json |  8 +++++---
 benchmarks/optimization.md           | 13 +++++++++----
 gtfparse/__init__.py                 |  2 +-
 tasks/lessons.md                     |  2 ++
 tasks/todo.md                        | 10 +++++++---
 6 files changed, 25 insertions(+), 12 deletions(-)

diff --git a/benchmarks/README.md b/benchmarks/README.md
index 87711b7..5194e2a 100644
--- a/benchmarks/README.md
+++ b/benchmarks/README.md
@@ -4,7 +4,7 @@ The 3.0.0 migration uses PyArrow for fixed-column parsing and pandas for attribu
 processing and writing. Polars is an optional output adapter. This directory
 contains the measurement harness and the evidence for that decision.
 
-The follow-up [3.0.1 optimization report](optimization.md) compares the optimized
+The follow-up [3.0.2 optimization report](optimization.md) compares the optimized
 complete reader with both the 2.9.1 Polars and 3.0.0 Arrow/pandas releases.
 
 ## Reproduction
diff --git a/benchmarks/optimization-results.json b/benchmarks/optimization-results.json
index 2e04d7c..3a83831 100644
--- a/benchmarks/optimization-results.json
+++ b/benchmarks/optimization-results.json
@@ -1,10 +1,10 @@
 {
-  "description": "Final 3.0.1 captured-pair parser and 250000-row batched Arrow attribute assembly; complete production pipelines.",
+  "description": "Captured-pair parser and 250000-row batched Arrow assembly measured as candidate 3.0.1; same runtime optimization ships independently as 3.0.2 to avoid the parallel documentation PR version.",
   "date": "2026-10-06",
   "baseline_sources": {
     "2.9.1-polars": "3e0912c",
     "3.0.0-arrow": "be449cbed70be8f515bb92b3eb2f43e55276917c",
-    "optimized": "perf/attribute-expansion PR against be449cb; version 3.0.1"
+    "optimized": "bd0cd44b145dc089968a77b4b6ca3bd614716ee2 (measured candidate 3.0.1); release 3.0.2 changes version numbering only"
   },
   "method": {
     "threads": 4,
@@ -7141,5 +7141,7 @@
         "wall_seconds": 40.073447916001896
       }
     }
-  }
+  },
+  "release_version": "3.0.2",
+  "measured_optimization_version": "3.0.1"
 }
diff --git a/benchmarks/optimization.md b/benchmarks/optimization.md
index f4a4370..62c1951 100644
--- a/benchmarks/optimization.md
+++ b/benchmarks/optimization.md
@@ -1,9 +1,14 @@
-# Complete-load optimization in 3.0.1
+# Complete-load optimization in 3.0.2
 
-Version 3.0.1 retains the pandas default and optional Polars output introduced
+Version 3.0.2 retains the pandas default and optional Polars output introduced
 in 3.0.0. It improves the shared Python attribute parser and the construction
 of expanded pandas columns while keeping the supported reader behavior.
 
+The benchmark candidate was labeled 3.0.1 when measured. The concurrent
+documentation PR targets that version, so this independent optimization
+ships as 3.0.2. Raw records retain their measured version labels;
+result tables refer to the release containing that implementation.
+
 ## Changes
 
 - Capture attribute names and double-quoted values directly in the compiled
@@ -79,7 +84,7 @@ pandas versions. Times and memory are fresh measurements from this sweep;
 compare readers within this table rather than mixing earlier timings from a
 shared workstation.
 
-| pandas | Input (250k rows) | 2.9.1 Polars s | 3.0.0 Arrow s | 3.0.1 s | Polars MiB | 3.0.0 MiB | 3.0.1 MiB |
+| pandas | Input (250k rows) | 2.9.1 Polars s | 3.0.0 Arrow s | 3.0.2 s | Polars MiB | 3.0.0 MiB | 3.0.2 MiB |
 | --- | --- | ---: | ---: | ---: | ---: | ---: | ---: |
 | 2.3.3 | ensembl-250k.gtf | 2.57 | 2.91 | 2.29 | 560 | 617 | 583 |
 | 2.3.3 | ensembl-250k.gtf.gz | 2.86 | 4.58 | 2.36 | 561 | 620 | 584 |
@@ -110,7 +115,7 @@ one fresh process per reader. All six complete loads match content and column
 order. These single observations demonstrate scaling and correctness, and do
 not establish repeated full-file throughput.
 
-| Full gzip input | Rows | Polars s | 3.0.0 s | 3.0.1 s | Polars MiB | 3.0.0 MiB | 3.0.1 MiB |
+| Full gzip input | Rows | Polars s | 3.0.0 s | 3.0.2 s | Polars MiB | 3.0.0 MiB | 3.0.2 MiB |
 | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
 | Ensembl GRCh38.114 | 4,116,048 | 47.11 | 67.63 | 36.86 | 6504 | 7199 | 3639 |
 | GENCODE v48 primary | 4,119,244 | 43.74 | 52.35 | 29.88 | 7091 | 6560 | 3698 |
diff --git a/gtfparse/__init__.py b/gtfparse/__init__.py
index 20c94a5..d5c1712 100644
--- a/gtfparse/__init__.py
+++ b/gtfparse/__init__.py
@@ -24,7 +24,7 @@
 )
 from .write_gtf import write_gtf
 
-__version__ = "3.0.0"
+__version__ = "3.0.2"
 
 __all__ = [
     "GENCODE_BIOTYPE_ALIASES",
diff --git a/tasks/lessons.md b/tasks/lessons.md
index edde6e1..39602b9 100644
--- a/tasks/lessons.md
+++ b/tasks/lessons.md
@@ -7,3 +7,5 @@
 - When asked to close an unreproduced historical report, record the evidence
   and close it as unreproduced rather than leaving it open indefinitely or
   implying that an unverified fix shipped.
+- When asked to finish a PR, prioritize its remaining checks, merge and
+  publication before further experiments or follow-up work.
diff --git a/tasks/todo.md b/tasks/todo.md
index f5622e8..9b230ff 100644
--- a/tasks/todo.md
+++ b/tasks/todo.md
@@ -386,7 +386,7 @@ artifact verification will be recorded on the PR to keep clean master clean.
 
 The user requested a PR to improve the pandas/Arrow implementation and attempt
 repeatable complete-load performance better than the old Polars implementation,
-without narrowing supported inputs or API behavior. Version this as 3.0.1.
+without narrowing supported inputs or API behavior. Version this as 3.0.2.
 
 Preserve all expanded/raw GTF semantics, fixed numeric/categorical and inferred
 string dtypes, arbitrary/repeated/quoted attributes, column collisions/order,
@@ -477,11 +477,15 @@ MiB and 3.0.0 52.35 s/6560 MiB. These are single observations per reader on the
 shared workstation, distinct from the repeated subset medians. Final selective
 exon/five-column and optional Polars-output checks also match all three releases.
 84 final records and separately labeled development evidence are retained in
-benchmarks/optimization-results.json. Version is bumped to 3.0.1.
+benchmarks/optimization-results.json. Version is bumped to 3.0.2.
 
 Packaging review: wheel and source distribution build successfully, have identical
-3.0.1 dependency metadata, and pass strict Twine checks. The wheel excludes tests
+3.0.2 dependency metadata, and pass strict Twine checks. The wheel excludes tests
 and benchmarks; the source archive includes the new regression tests, runner and
 retained evidence. Its shipped ./lint.sh and ./test.sh pass (199 plus 10 optional
 skips, 96% coverage). Production code, test oracle and benchmark method reviewed;
 release/checklist completion will be recorded on the PR after merge and PyPI.
+
+Release sequencing: documentation PR #93 targets 3.0.1. Keep this PR
+independent of that unmerged work and ship 3.0.2. Benchmark records retain measured candidate version 3.0.1; runtime
+parsing and assembly behavior is unchanged by version numbering.

From 7fb46e7ed7cc7b0eab9d72cd09514c5a8224abd7 Mon Sep 17 00:00:00 2001
From: Alex Rubinsteyn <alex.rubinsteyn@gmail.com>
Date: Tue, 6 Oct 2026 13:47:23 -0400
Subject: [PATCH 3/3] Exercise opt-in Arrow string batches on pandas 2

---
 benchmarks/optimization.md           | 4 +++-
 tasks/todo.md                        | 9 ++++++++-
 tests/test_attribute_optimization.py | 4 +++-
 3 files changed, 14 insertions(+), 3 deletions(-)

diff --git a/benchmarks/optimization.md b/benchmarks/optimization.md
index 62c1951..c7f250d 100644
--- a/benchmarks/optimization.md
+++ b/benchmarks/optimization.md
@@ -142,7 +142,9 @@ quoted/unquoted, custom quote delimiters, Unicode/whitespace, malformed fragment
 repeated/empty values, pre-split rows, missing-value sentinels and column
 selection. Additional tests check configured Python/Arrow string storage, batch
 boundaries, late-discovered columns, missing batches, fixed-field attribute
-collisions, first-seen order and global progress counts. Existing tests cover streams, null fields, exact
+collisions, first-seen order and global progress counts. The batch regression
+opts pandas 2 into inferred strings so both configured storage backends are
+exercised there; default object-string behavior has separate coverage. Existing tests cover streams, null fields, exact
 integer coordinates, comments/BOM, aliases, converters, progress/cancellation,
 optional Polars, dictionary output and writer round trips.
 
diff --git a/tasks/todo.md b/tasks/todo.md
index 9b230ff..2ab82ed 100644
--- a/tasks/todo.md
+++ b/tasks/todo.md
@@ -464,7 +464,7 @@ Final release status will be recorded on the PR so clean master remains clean
 for ./deploy.sh.
 
 Final validation: ./lint.sh passes. Normal ./test.sh passes 213 tests with
-90% coverage. Base pandas 3.0.6/Arrow 25 and pandas 3.0.6/Arrow 18 pass 199 tests
+97% coverage. Base pandas 3.0.6/Arrow 25 and pandas 3.0.6/Arrow 18 pass 199 tests
 plus 10 optional skips (96% coverage). Minimum Python 3.9/pandas 2.2.2/Arrow 18
 passes 199 plus 10 skips (88%; configured object strings omit the Arrow branch).
 Minimum optional Polars 0.20.31 passes 212 plus one modern-only skip (97%, existing
@@ -489,3 +489,10 @@ release/checklist completion will be recorded on the PR after merge and PyPI.
 Release sequencing: documentation PR #93 targets 3.0.1. Keep this PR
 independent of that unmerged work and ship 3.0.2. Benchmark records retain measured candidate version 3.0.1; runtime
 parsing and assembly behavior is unchanged by version numbering.
+
+Coverage review: the pandas 2.3 flag initially reported a drop because its
+default object strings omitted the Arrow batch branch. Opt the batch regression
+into inferred strings, exercising configured Python/Arrow backends on pandas 2
+as well as pandas 3. Normal tests now pass 213 with 97% coverage; recent pandas
+with minimum Arrow 18 passes 199 plus 10 optional skips (96%). Runtime code and
+all retained benchmarks are unchanged.
diff --git a/tests/test_attribute_optimization.py b/tests/test_attribute_optimization.py
index 234c53e..10b3312 100644
--- a/tests/test_attribute_optimization.py
+++ b/tests/test_attribute_optimization.py
@@ -112,7 +112,9 @@ def test_batched_expansion_preserves_late_columns_gaps_collisions_and_progress(
     attrs += ['gene_id "G3"; start "42"; late "last";'] * 3
     text = "".join("1\ttest\tgene\t1\t100\t.\t+\t.\t%s\n" % a for a in attrs)
     events = []
-    with pd.option_context("mode.string_storage", storage):
+    # pandas 2 requires opting into inferred strings to exercise both storage
+    # backends; pandas 3 infers strings by default.
+    with pd.option_context("mode.string_storage", storage, "future.infer_string", True):
         frame = read_gtf(StringIO(text), progress_callback=lambda *event: events.append(event))
     assert list(frame)[-4:] == ["gene_id", "tag", "note", "late"]
     assert list(frame).index("start") == 3