Preregister DRI-1B human identifiability gate

[Silentpartnercoding]

Lane

Candidate preregistration only. No confirmatory evidence exists.

Corrective purpose

DRI-1A preserved a frozen adverse candidate result, but its pre-run package was not complete enough for canonical promotion. This PR does not retrofit DRI-1A. It starts a new lifecycle correctly and tests whether a decision-relative proximal root is reproducibly identifiable by blinded humans before any AI selector is built.

Commit order

1. 63c5543 freezes the protocol, authoring guide, case schema, and response schema.
2. 2609467 creates the candidate lifecycle record pinned to the earlier protocol commit.

There is no result, manifest, verdict, case packet, reviewer response, model call, or runtime authority effect.

Frozen design

• 60 development and 216 untouched confirmatory cases.
• Six balanced target labels, including joint_or_insufficient.
• 72 material-minority reversals, 72 false-rescue traps, and 72 cases without a headcount minority.
• At least six external case authors, two pre-review adjudicators, and three separate blinded test reviewers.
• Reviewer consensus compared against every fixed cut, always-abstain, random, and oracle controls.
• Registered confidence intervals, bootstrap seed, Holm correction, joint success gate, stop conditions, invalidation rules, environment commitment, and evidence-package requirements.
• A minority is rewarded only when decision-relative independent support warrants the oracle disposition. False minority rescue and excessive abstention are explicit failure endpoints.

Boundaries

The lifecycle record declares same-control-domain and authorityEffect none. Authorship diversity is not claimed as independent verification. This PR does not authorize recruitment, payment, data collection, model testing, WEX field collection, action gating, deployment, or public-result claims.

An AI selector receives a separate identifier and preregistration only if the human-identifiability gate passes.

Verification

make verify passed outside the restricted socket sandbox:

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

The initial sandboxed verification attempt failed only because loopback sockets and the npm cache were blocked; the same suite passed with those test permissions enabled.