[gated on afi-governance #53] CAL-BENCHKIT-RETIRE: afi-econ: re-point --scores merit path to a synthetic merit-scores file

[Gio2050]

Gate

This PR may merge only after afi-governance PR #53 (CAL-GOV, analyst-calibration-record-v0.1) is merged with the CAL-BENCHKIT-RETIRE slot expressly authorized. It is a DRAFT until then. Clause implemented: D-CAL-6(3), the afi-econ act — "re-point afi-econ's --scores merit path to synthetic inputs or retire it (research plane, MATH-GOV §3 non-canonical; proven by afi-econ's own tests)".

PoI / PoInsight are NOT retired (D-CAL-5(4)). Proof-of-Intelligence and Proof-of-Insight remain reserved, preserved protocol reputation primitives (CONST-GOV D-CONST-5). The retired thing is afi-benchkit, an ungoverned implementation on reserved ground. The poi / poinsight keys in the synthetic file are placeholders that stand in for no protocol value.

What changed

• examples/merit_scores.synthetic.json (new) — the documented synthetic merit-scores input, same shape as before: {reputation:{score}, poi:{score}, poinsight:{score}, n, stamp}; stamp.source = "synthetic". Replaces tests/fixtures/scores_min.json.
• src/afi_econ_kit/cli.py — _load_benchkit_scores → _load_merit_scores; --scores help on simulate/replay/gauge now reads: merit scores JSON file (optional; synthetic — see examples/merit_scores.synthetic.json). The protocol source is the CAL-GOV analyst calibration record — not a scalar — and any merit conversion is CHAIN-GOV reserved. Stdout line: Using merit scores (synthetic): ….
• scenarios.py / gauge.py / schemas.py / params/gauge_v0.yaml — wording; provenance stamp key benchkit_stamp → merit_stamp (pass-through unchanged).
• scripts/end_to_end_audit.py — no longer locates a sibling afi-benchkit or shells out to afi-bench; reads the synthetic file; report key benchkit_influence_ok → merit_influence_ok.
• tests/test_scores_ingest_with_file.py, examples/pipeline_demo.sh, examples/README.md — re-pointed; README's format block now shows the real file shape and documents the provenance boundary.
• Removed tracked generated artifacts: out/, out_audit/, out_base/, out_with/ (the latter three pointed at /Users/secretservice/afi-benchkit/out/scores.json) and the stale src/afi_econ_kit.egg-info/ (a copy of a pre-harvest README naming afi-benchkit). All were already .gitignored; *.egg-info/ added. Nothing new committed in their place.

Slot gate (§8 CAL-BENCHKIT-RETIRE)

• Zero score/hash/golden/governed-schema movement — nothing under afi-config/schemas or the source-disclosure-profile surface is touched.
• Proven by afi-econ's own tests: pytest tests → 88 passed, 13 failed, 1 skipped (baseline on origin/main: 86 passed, 15 failed). The 13 remaining failures are pre-existing and unrelated (missing sibling afi_emissions module, missing data/timeseries_rebuilt.csv). Both --scores tests and both end_to_end_audit tests pass.
• No document here describes PoI or PoInsight as retired, deprecated, or narrowed.

Note for the reviewer

The audit's previous synthetic fallback (0.62/0.68/0.65) tripped the audit's own gauge_invariants check (with-scores shares summed to 1.0082 after smoothing) — a pre-existing merit-path defect that this PR does not touch (it would move a research-plane score). With the documented 0.58/0.55/0.60 file the check passes.

🤖 Generated with Claude Code

[kilo-code-bot]

Kilo Code Review could not run — your account is out of credits.

Add credits or switch to a free model to enable reviews on this change.