メニュー
Verify parity between notebook-like pipeline execution and scripted experiment workflows so research findings are reproducible.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
| name | parity-check |
| description | Verify parity between notebook-like pipeline execution and scripted experiment workflows so research findings are reproducible. |
Use this skill when confirming notebook results and experiment-runner results are consistent for core metrics.
python/tests/test_experiment_scaffold.pyuv run --python .venv/bin/python pytest python/tests/test_experiment_scaffold.py -q
If metrics drift:
pipeline/costs/gates).config_digest and git SHA.Audit experiment artifact folders for completeness, schema consistency, and promotion traceability across single and grid runs.
Build candidate features, run feature screening and diagnostics, and prepare accepted feature sets for downstream flywheel experiments.
Run OpenQuant notebook-to-experiment flywheel iterations (single run and grid run), capture artifacts, and summarize ranked outcomes for promotion decisions.
Execute OpenQuant notebook code cells non-interactively, persist outputs to artifact paths, and fail fast on runtime errors.
Evaluate promotion readiness from flywheel outputs using statistical and economic gates, then produce a concise go/no-go decision note.
Build and run an AFML-grounded, chapter-by-chapter documentation loop for OpenQuant docs. Use this when you need consistent module docs, hierarchical chapter coverage, and iterative progress tracking with MCP-backed prompts.