Skip to content

scripts_index.md

Latest commit

 

History

History
76 lines (61 loc) · 2.97 KB

scripts_index.md

File metadata and controls

76 lines (61 loc) · 2.97 KB

Scripts Index

This index separates the canonical operator scripts from the broader research toolkit in scripts/.

Canonical operator surface

Use these first.

  • scripts/build_private_submission.py
    • builds the final private submission JSON from an eval checkpoint (primary build tool, private phase)
  • scripts/build_v9_recovery.py / scripts/build_v10_recovery.py / scripts/build_v14_v91_hybrid.py
    • version-specific recovery/hybrid builders from checkpoints
  • scripts/build_v16_hybrid.py / scripts/build_v16_super_hybrid.py
    • V16 hybrid builders (V16 answers + V15 pages)
  • scripts/build_v17_hybrid.py / scripts/build_v17_super_hybrid.py
    • V17 hybrid builders (V17 answers + V15 pages, with DOI/boolean/number corrections)
  • scripts/tzuf_v9_full900.py / scripts/tzuf_v11_full900.py / scripts/tzuf_v12_full900.py / scripts/tzuf_v15_full900.py
    • full 900-question eval runners (use with appropriate profile)
  • scripts/start_dashboard.sh
    • start the live dashboard (frontend + parsers)
  • scripts/patch_best_formatting.py
    • fixes formatting issues (double periods, truncation artifacts) in BEST.json
  • scripts/enrich_submission_pages.py
    • enriches submission page references
  • scripts/container_contract_smoke.py
    • deterministic packaging/container smoke
  • scripts/private_doctor_preflight.py
    • bounded private-day doctor / archive safety checks
  • scripts/rehearse_private_run.py
    • deterministic private-run rehearsal flow
  • scripts/build_run_manifest.py
    • deterministic run-manifest fingerprinting
  • scripts/pre_submit_sanity_check.py
    • pre-submission sanity validation (format, pages, coverage, regressions)
  • scripts/build_submit_v2.sh
    • one-command Submit #2 builder — DEPRECATED: V18 failed, Submit #2 cancelled

Submission gating tools

  • scripts/v13_vs_v91_gate.py
    • gate comparison: V13 vs V9.1 on key metrics
  • scripts/v16_vs_v15_gate.py
    • gate comparison: V16 vs V15 on key metrics
  • scripts/question_cluster_analysis.py
    • cluster private questions by type/family for targeted analysis

Grounding / evaluation tools

Use when working on reviewed grounding changes.

  • scripts/capture_query_artifacts.py
  • scripts/score_against_golden.py
  • scripts/run_grounding_sidecar_ablation.py
  • scripts/export_grounding_ml_dataset.py
  • scripts/train_grounding_router.py
  • scripts/train_page_scorer.py
  • scripts/eval_grounding_models.py
  • scripts/mine_within_doc_rerank_opportunities.py

Research and advisory scripts

The rest of scripts/ is primarily research-facing:

  • audits
  • scanners
  • candidate builders
  • ranking/search helpers
  • one-off diagnostics

Treat those as advisory unless a ticket explicitly names them.

Not a cleanup target right now

  • No broad script-tree reorganization is planned in the current competition window.
  • If a script is not listed above, do not assume it is production-critical.
  • Add new scripts to this index only when they become part of the canonical operator or evaluation path.