Record adverse DRI-1A candidate result

[Silentpartnercoding]

Lane

Candidate / noncanonical research result.

What this records

• Freezes the DRI-1A protocol before inspectable outcome metrics.
• Adds the deterministic 8,192-world generator and scorer.
• Preserves the failed joint criterion without retuning.
• Records content hashes, two-run semantic reproducibility, limitations, and the procedural history.

Result

The joint criterion was rejected.

• Decision-relative oracle: 90.99% correct settlement.
• Agent headcount: 60.78%.
• Best fixed-cut correct-settlement rate: 68.32%.
• Minority-reversal recovery: 100% for the relevant cut versus 0% for headcount.
• The deepest fixed cut achieved only 3.26% false settlement by abstaining on 40.23%, so the preregistered every-fixed-cut and abstention-matching criteria failed.
• Deterministic five-cut evaluation measured below 0.7 ms p99 on the test machine.

Claim boundary

This is a frozen synthetic candidate diagnostic, not a canonical result, public claim, real-world validation, causal-cut inference result, or authority grant. The explicit rules engine received the failure-domain label. The pre-run package omitted fields required for canonical promotion, and those omissions are not repaired after observing the outcome.

Builder, runner, and reviewer share one operator/control domain. Reproduction is internal, not independent verification.

No website deployment or runtime gate is included.

Verification

make verify passed:

• 448 pytest tests passed, 8 skipped, 21 subtests passed.
• Registration-chain, public-boundary, withheld-leak, and research-integrity checks passed.
• Site build and 7 rendered tests passed.
• 59 engine tests passed.