Files
fusion-harness/live_final_generation/MANIFEST-claude-fable-5-gpt-5.6-sol.md
IndyDevDan 5852f2ed4f 🚀
2026-07-19 20:30:34 -05:00

2.1 KiB
Raw Permalink Blame History

Project File Inventory — fused from claude-fable-5 + gpt-5.6-sol

Snapshot of every file built for the SQLite bulk-insert benchmark project, copied 2026-07-16 from the original locations into this directory.

project-root/ — deliverables (from /private/tmp/ddc0f4d0/fusion-harness/)

Listed by BOTH agents.

File Built by Description
bench-sqlite-bulk-ARCHITECT-anthropic-claude-fable-5.py ARCHITECT Independent benchmark design — 7 strategies, subprocess isolation, ru_maxrss RAM measurement
sqlite-bulk-benchmark-BUILDER-openai-gpt-5.6-sol.py BUILDER Independent parallel benchmark design
BENCH_PLAN.md Fusion Fused plan: method, fairness controls, measured results, winners, embedded fused script
bench.py BUILDER (gated loop) Canonical deliverable — stdlib-only uv single-file implementation of BENCH_PLAN.md
RESULTS.md Generated by uv run bench.py Measured results (7 strategies, time + peak RAM), Speed Winner set_based_ctes, Memory Winner declared

Root of this dir — fused script (from /tmp/)

Listed by ARCHITECT only.

  • fused-sqlite-bench-claude-fable-5-gpt-5.6-sol.py — runnable fused benchmark referenced by BENCH_PLAN.md

harness-artifacts/ — process artifacts (from /tmp/fusion-harness-K48sbW/)

Listed by ARCHITECT only.

  • gate.py — 15-check acceptance gate (Definition of Done)
  • prompt.md, validator.md — harness briefs
  • gate-baseline.txt — required RED baseline before any build
  • gate-round-1.txtgate-round-5.txt — all five gate runs
  • builder-round-1.mdbuilder-round-5.md — all five builder reports
  • triage-round-3.md, triage-round-4.md — escalation briefs diagnosing the gate defect

Status note (per ARCHITECT)

Final gate run ends RED (11 passed / 4 failed) due to a known defect in gate.py:61 (strips underscores from searched text, making strategy-name checks unsatisfiable). The substantive deliverables — bench.py and RESULTS.md — are correct, measured, and complete: 1M rows, all 7 strategies, ~17002000x speedup over naive autocommit.