[SkyeAv]

Table sections gain a third source kind, parquet, so .parquet tables can drive builds directly as table inputs shift toward parquet. The kind flows end to end: config model, section pipeline, agent table tooling, docs, and changelog.

Source model

• New kind token. Files.PARQUET = "parquet" in src/tablassert/enums.py.
• Parquet(BaseSource) in src/tablassert/models.py adds no format-specific fields: parquet is self-describing (typed columns, real headers, no sheets or delimiters), so extra="forbid" rejects Excel-style sheet/delimiter keys instead of silently ignoring them. Shared rows/row_slice/reindex controls carry over.
• Union widened. Section.source is now Excel | Text | Parquet (smart union on the kind literal; all three kinds still validate), and _section_source_label accepts the widened union, rendering parquet as the bare path.

Pipeline reader

• lib.parquet in src/tablassert/lib.py scans via pl.scan_parquet (lazy, native, no new dependency) and renames the file's columns to the positional column_1.. convention in schema order. Why positional: Encoding.excel_style_columns only accepts Excel-style A-ZZ letters for method: column encodings for every source kind, so configs address columns positionally and the reader must match what the headerless csv/excel readers already produce. Accepted tradeoff: a parquet file's own header names are not referenceable from configs, same as CSV/TSV.
• Wiring. Table._source_ops dispatches exactly one load op per source.kind; PHASE_OF[parquet] is "load" for progress UX; runcache.OP_COST[lib.parquet] = 2 matches the other load ops for snapshot pricing.

Agent

• Table reads. agent._load_table dispatches .parquet to pl.read_parquet (typed columns as-is; parse failures still collapse to the clear ValueError; unknown suffixes still raise), and _effective_rows_cached counts parquet rows with the same all-null blank-row rule as the Excel branch.
• Bucketing. TABLE_EXTENSIONS gains .parquet, so is_table_file/is_useful_file accept parquet payloads during fetch. The agent keeps real column names here: it renders previews, it does not author positional configs.

Docs

• docs/configuration/table.md gains a #### Parquet Source section (field table + YAML example) and the "Specify kind explicitly" callout enumerates the parquet reader.
• CHANGELOG.md gains an Unreleased Added entry.
• tests/test_docs_source_of_truth.py binds Parquet via ModelTableBinding(Parquet, ..., "Parquet Source") so the docs field table is checked bidirectionally against the live model like every other config model.

Testing

• uv run pytest -> 1674 passed, 52 skipped (skips are pre-existing extra/platform gates; the new tests execute by name in the targeted runs below)
• Targeted across all impacted files: uv run pytest tests/test_enums.py tests/test_cover_models.py tests/test_cli_validation.py tests/test_cover_lib.py tests/test_e2e_smoke.py tests/test_run_cache.py tests/test_cli_progress.py tests/test_cover_agent_core.py tests/test_agent_fetch.py tests/test_docs_examples.py tests/test_docs_source_of_truth.py tests/test_docs_cli_coverage.py -q -> 401 passed
• uv run ruff check . and uv run ruff format --check . -> clean
• uv run pyright -> 0 errors, 0 warnings, 0 informations
• cargo test --manifest-path rust/Cargo.toml -> 127 passed, 11 passed / 1 ignored, 17 passed, 1 passed; 0 failed (no Rust changes)
• CLI exercise: tablassert validate -s table on a kind: parquet config against a real .parquet file -> exit 0
• New e2e smoke test_build_pipeline_over_parquet_source builds a KG from a parquet source through a REAL redb and asserts HGNC:1100/HGNC:6871 in the KGX nodes and edges output

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Advanced

Run ID: d21e1830-c9f1-4fcd-a5b1-48854450680c

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}