- commit
- 0898f96118c29558a30f16d807c48e8ed9a5ccc9
- parent
- d591b3a2e6eac8a5fe1b3862275234b6474f6abd
- author
- dami <gazey55@gmail.com>
| M | PLAN.md | 32 | ++++++++++++++++++++++++++++++ |
@@ -799,6 +799,38 @@ retroactive verdict: Codex a0 dead at rev 346 (38% wrong), a2 dead at
799799 256 (33%), the honest agent healthy at 0.1%. Awaiting the acceptance800800 run.801801 802+### Acceptance runs under health (2026-08-27) and what they taught803+804+- Codex, re-run with the old probing script: dead at one tile (42%).805+ Codex again, adapted: enumerates solutions and commits their806+ intersection, one ambiguous commit at a time — the technique ladder,807+ arrived at because guessing costs.808+- Fable 5: inferred balance and no-three from solved tiles, then809+ inferred the uniqueness rule too (correctly) and talked itself out of810+ it, committed whole candidate solutions on m=3 tiles at 1-in-3 odds811+ calling it deduction, fired one guess per stuck tile per round across812+ dozens of tiles, and died at 4% (80 of 2,000) at revision 1247 with 15813+ tiles. Root cause on our side: **no tool on the host carried the814+ rules** — the rules text hung off `start_mosaic`, which the host hides815+ — so the agent never read that lines must be distinct, that m is the816+ tile's own solution count, or how health works. Fixed: the rules817+ travel with `join` (the one tool every agent calls), every tool points818+ at the `mosaic://rules` resource, `observe` defines every tile field819+ where it appears, and the rules open with a glossary.820+- Worth keeping from its post-mortem: ambiguity under m >= 2 is normal821+ and is resolved by expansion (a solved neighbour's shared cells822+ collapse the count), so the right policy is reveal more, wait for823+ seams, and guess one cell only when the territory is saturated — the824+ rules now say so; and tile-local marginals over enumerated solutions825+ are not evidence, because the alternatives are planted (a flip826+ rectangle) and the truth is one of m near-symmetric options — say827+ this in the rules too, so no agent trusts a 0.9 "probability".828+- To play with (from the user): the uniqueness rule is subtle enough829+ that an agent cannot confirm it empirically ("no duplicate lines830+ observed" is weak evidence); per-tile rule masks announced in observe831+ (idea 5 above) make it explicit and let tiles differ; a field with832+ uniqueness off on some tiles is a gate, on with it is the standard.833+802834 ### Status of the v2 items under this goal803835 804836 1 real packed run: done (v2 packings and v3 scatter fields played, logs