Improve result plots for the paper

AGENTS.md

@@ -151,7 +151,6 @@ ckpt intermittent schematicO3 [BENCHMARKS...] --trace 1,2 [--cap board] [--devic
 # Analysis
 ckpt analyze strip-mining LOG_FILE [-o CSV]
 ckpt analyze milp-coarse  ...    # coarse-allocation MILP analysis
-ckpt analyze plot       CSV_DIR [--metric M] [--algorithms A...]
 
 # Device interaction
 ckpt device read-serial [--timeout N] [--end-marker M]
@@ -206,8 +205,7 @@ scripts/ckpt/
 │   └── schematic.py     # SCHEMATIC semantic verification
 └── analysis/
     ├── strip_mining.py  # Parse verbose logs for strip-mining K values
-    ├── milp_coarse.py   # Coarse-allocation MILP analysis
-    └── plot.py          # Plot benchmark CSV results (requires `uv sync --extra plot`)
+    └── milp_coarse.py   # Coarse-allocation MILP analysis
 ```
 
 ### Standalone Scripts
@@ -216,9 +214,11 @@ scripts/ckpt/
 # Visualize results from a result directory
 Rscript scripts/plot_results.R [--result-dir DIR] [--output-dir DIR] [--absolute] [--all-metrics]
                                [--benchmarks B,...] [--metrics M,...] [--log-scale] [--config FILE]
+# Box plot of intermittent-power results (results/intermittent/summary.csv)
+Rscript scripts/plot_intermittent.R [--result-dir DIR] [--output-dir DIR]
 ```
 
-`scripts/plot_results.R` produces per-capacitor bar charts, by default `execution_time` and `runtime_region_boundary_calls` normalized against the baseline marked `normalize_ref` in the config (falling back to the first algorithm). `--all-metrics` adds `region_boundaries`, `profiling_time`, and `compilation_time`; `--absolute` turns normalization off. Region boundary data comes from the device-debug CSVs, timing data from the non-debug ones. Which CSVs are read, and their labels/colors/patterns, come from a JSON config — no filename is hardcoded in the R script. `scripts/plot_config.json` is the default; `scripts/plot_config_o0.json` additionally plots `uninstrumentedO0.csv` as a second execution-time baseline (`--config scripts/plot_config_o0.json`). Each config lists `algorithms` (per-capacitor series; CSV names default to `<algo>.csv` / `<algo>_debug.csv`, overridable with `csv` / `debug_csv`) and `baselines` (capacitor-independent series; `csv` required, optional `metrics` restricts which metrics they appear in).
+`scripts/plot_results.R` produces per-capacitor bar charts, by default `execution_time` and `runtime_region_boundary_calls` normalized against the baseline marked `normalize_ref` in the config (falling back to the first algorithm). `--all-metrics` adds `region_boundaries`, `profiling_time`, and `compilation_time`; `--absolute` turns normalization off. Region boundary data comes from the device-debug CSVs, timing data from the non-debug ones. Which CSVs are read, and their labels/colors/patterns, come from a JSON config — no filename is hardcoded in the R script. `scripts/plot_config.json` is the default; `scripts/plot_config_o0.json` additionally plots `uninstrumentedO0.csv` as a second execution-time baseline (`--config scripts/plot_config_o0.json`). Each config lists `algorithms` (per-capacitor series; CSV names default to `<algo>.csv` / `<algo>_debug.csv`, overridable with `csv` / `debug_csv`) and `baselines` (capacitor-independent series; `csv` required, optional `metrics` restricts which metrics they appear in). `scripts/plot_intermittent.R` takes its series labels, colors, and fill patterns from `scripts/plot_config.json` too; shared helpers live in `scripts/plot_common.R`.
 
 ## Architecture
 

scripts/ckpt/cli.py

@@ -14,7 +14,6 @@
 
 import click
 
-from .analysis.plot import ALGORITHMS, METRICS
 from .bench.all import ALL_ALGORITHMS, DEFAULT_ALGORITHMS
 from .compile.common import HALT_MODES
 from .errors import (
@@ -1815,51 +1814,6 @@ def analyze_milp_coarse_cmd(
         write_milp_coarse_summary_csv(summary_rows, Path(output))
 
 
-@analyze.command("plot")
-@click.argument("csv_dir", type=click.Path(exists=True))
-@click.option(
-    "--metric",
-    type=click.Choice(list(METRICS.keys())),
-    default="prologue",
-    help="Metric to plot.",
-)
-@click.option(
-    "--algorithms",
-    multiple=True,
-    type=click.Choice(list(ALGORITHMS.keys())),
-    help="Algorithms to include (default: all).",
-)
-@click.option("--benchmarks", multiple=True, help="Benchmark programs to include.")
-@click.option("--capacitors", multiple=True, help="Capacitor sizes to include.")
-@click.option(
-    "--normalize",
-    type=click.Choice(list(ALGORITHMS.keys())),
-    help="Normalize values relative to this algorithm.",
-)
-@click.option("-o", "--output", type=click.Path(), help="Output file (e.g., plot.png).")
-def analyze_plot_cmd(
-    csv_dir: str,
-    metric: str,
-    algorithms: tuple[str, ...],
-    benchmarks: tuple[str, ...],
-    capacitors: tuple[str, ...],
-    normalize: str | None,
-    output: str | None,
-) -> None:
-    """Plot benchmark comparison chart from CSV results."""
-    from .analysis.plot import plot_benchmarks
-
-    plot_benchmarks(
-        csv_dir=Path(csv_dir),
-        metric=metric,
-        algorithms=_list_or_none(algorithms),
-        benchmarks=_list_or_none(benchmarks),
-        capacitors=_list_or_none(capacitors),
-        normalize=normalize,
-        output_file=_path_or_none(output),
-    )
-
-
 # =========================================================================
 # device group
 # =========================================================================