Drop7 Research
← Docs
docs/research/status/evidence.md

Research status evidence

FindingEvidenceStatus
TypeScript rules engine122 local testsReproduced in this checkout
Native engine and n-tuple checksGradient and self-testsReproduced in this checkout
Native/TypeScript trajectory agreementDeterministic parity sweepReproduced; see reproducibility notes
Fair D4 vs D3: 400,675.25/116.375 vs 235,071.25/71 over 8 gamesDetailed ledgerRecorded, small confirmation cohort
Fair D4: 308,295.578 points and 90.031 moves over 64 gamesDetailed ledger; the 64 seeds, dispersion, censoring, and flow statistics required by methodology.md were not retainedProvisional reference mean pending a re-run under the benchmark contract
Fair D4 reproduced: 321,992 points and 94.06 moves over 64 gamesFresh exploratory seeds 0xa51d0000+, unmodified frozen sourceDevelopment tier; single cohort, consistent with the ledger figure
Depth x chance-resolution factorial, depths 2-5 x 5/7 strata64 paired games per cell on 0xa51d1000+, bit-exact accelerated engineDevelopment tier; the depth-5 cells are partial (32 and 16 games)
Score is 94.29% row-rise bonus; r = 0.9995 with game length64-game decomposition with a per-game score identity checkDevelopment tier; reframes the objective as survival
One D4 game scored 1,246,684Task-record onlyAnecdote; not an average or qualification
Million-point candidate existsFrozen validation protocolNo
AFBR-40 afterstate ideaTask-record onlyProposal; no source, checkpoint, or result

The cleanup reproduced engine behavior and buildability, not the long and expensive training/evaluation runs in the historical ledger.

For a walkthrough with board animations, start at how the game works and the concepts primer; every term is defined in the glossary.