Flip ratified amendment 2 live: 20-seed mean c2st rule (prospective only) - #69
Merged
Conversation
Applies the amendment ratified by the merge of PR #67 (4e06e24) after its full ceremony (adversarial referee round 4904161939 AMEND, fixes ae0c166, verification 4905067301 RATIFY AS-IS): - pairs-view c2st_auc leaves the per-seed geometry conjunction; the live rule is the across-seed mean over pre-registered seeds 0-19 at the same ratified 0.53 line, guarded by a version-matched per-seed cap of 0.554 (= matched floor mean + 8 sd from runs/c10_diagnostics_v1.json) - standing amendment_rules adopted: no_self_rescue and the scikit-learn classifier version pin - PROSPECTIVE ONLY: no committed run's verdict changes; candidate 10 remains FAIL; amendment_history entry 2 records the ceremony and the delegated-authority ratification - stored-thresholds tests across the 12 run artifacts made supersession-aware (per_seed_rule_superseded, mirroring the amendment-1 reported_not_gated pattern); 6 new tests bind the ratified rules, the OC table, and the no-verdict-change guarantee at their live locations Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
This was referenced Jul 7, 2026
MaxGhenis
added a commit
that referenced
this pull request
Jul 8, 2026
…spective only) (#97) Moves gate-2's ratified amendment 1 from the amendment_proposed object into the locked protocol. Per-cell scoring goes from ONE frozen simulation draw to the MEAN over K=20 pre-registered draws (numpy default_rng(5200 + k), k=0..19) of the cell RATE, scored |ln(rbar_candidate,s / rate_a,s)| at BYTE-IDENTICAL tolerances and the same 46-cell, 4-of-5-seed conjunction. Only the candidate statistic changes; the error budget does not (all 46 tolerances, the power cap, governance, scope, and external anchors are byte-identical to origin/master). Ceremony (mirrors the gate-1 amendment-2 flip #69 and the gate-2 lock flip #81): adversarial round AMEND (PR 96 comment 4915412987) -> fixes 5b70840 (comment 4916048161) -> verification RATIFY AS-IS (comment 4916419901) -> ratified by merge of PR 96 (merge commit fec27eb) under the maintainer's standing campaign directive of 2026-07-07, full ceremony. The five locked-text edits (flip_on_ratification.flip_edits, applied exactly): - protocol.candidate: one frozen replicate -> K=20 draws at default_rng(5200 + k); records the committed 4200+s registration history and the new 5200+k convention explicitly (registration_divergence resolved). - statistic: r_candidate,s -> rbar_candidate,s (the mean over K=20 draws of the cell rate). - protocol.pass_rule: r_candidate,s -> rbar_candidate,s (mean of the cell RATE across draws, then |ln(rbar / rate_a)|). - protocol.faithful_candidate_oc: basis note added -- the draw-noise-free half-normal basis now MATCHES the estimator; p_seed_pass 0.9404 / p_gate_pass_4_of_5 0.9685 stand unchanged (unachievable under the single-draw estimator, finding 1). - protocol.pass_rule: the stray "DRAFT" label removed. Also carries the ratified fresh_run_artifact_schema (the candidate-10 run contract: per_draw_per_cell_rates [20,46,5]; undefined-draw INVALIDATES the run; report-only per-draw dispersion) into the live protocol, adds gate_2's first amendment_history entry (id 2026-07-08-mean-over-draws-estimator, flipped_live: this pull request), and removes amendment_proposed (preserved in git history at the ratification merge fec27eb). PROSPECTIVE ONLY (no_self_rescue): no committed run's verdict changes; candidates 1-9 stand FAIL. The single-draw estimator contradicted the locked block's own ratified OC basis; K=20 mean-over-draws restores ~0.939 / 0.967 ~= the ratified 0.9404 / 0.9685. Tests rewired locked-hot (the gate-1 amendment-2 lesson): the load-bearing guarantees -- the K=20 / 5200+k convention, the artifact schema, the OC basis, the history entry with all four ceremony pointers, and prospective-no-verdict-change vs the nine committed artifacts -- rebind to their LIVE post-flip locations in a new "Gate-2 amendment 1 RATIFIED bindings" section. Proposal-object tests stay dormant-safe (guarded on proposal_number == 1, so a future proposal != 1 never reactivates them); the two tests that asserted the OLD locked text are updated to the new live text. Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
MaxGhenis
added a commit
that referenced
this pull request
Jul 10, 2026
…ro threshold movement) (#112) Moves gate-2's ratified amendment 2 (the tranche split) from the amendment_proposed object into the locked block. STRUCTURAL, zero threshold movement: the locked gate_2.thresholds scored surface is byte-identical to origin/master -- all 46 gated cells, every tolerance and derivation, the option-(a) protocol, the ln(1.5) power cap, and the 4-of-5 seed conjunction unchanged (verified at full-subtree granularity). Only naming, scope prose, the certification map, the unlocked sibling stubs, one governance rule, and the history changed. Ceremony (mirrors flips #69, #81, #97): adversarial round AMEND BEFORE RATIFYING (PR 111 comment 4930448912) -> fixes 81ea41e (comment 4930615774) -> verification RATIFY AS-IS (comment 4930753295) -> ratified by merge of PR 111 (merge commit 8a4a240) under the maintainer's standing campaign directive of 2026-07-07, full ceremony. The six locked-text edits (flip_on_ratification.flip_edits, applied exactly): - description: first clause re-scoped so tranche 2a claims only mh85_23 + cah85_23 (+ ind2023er deaths); household-composition/MX23REL -> gate_2b. - description_formula_clause: the "computes exactly" statutory-formula oracle clause (#78/#80) RETAINED verbatim (fix C1). - holdout_basis: [mh85_23, cah85_23, MX23REL] -> [mh85_23, cah85_23]; MX23REL -> gate_2b.holdout_basis. - scope_note: provision_class_coverage prose gains explicit tranche ids (2a/2b/2c); the certification_scope map is promoted into the locked scope. - scope_note_draft_label: "This DRAFT tranche covers..." relabeled to "tranche 2a_marital_fertility covers...". - scope_note_holdout_xref: the caregiver "named in holdout_basis but not scored in this draft" cross-reference re-pointed to the unlocked gate_2b. The five flip_additions: tranche_id 2a_marital_fertility inside thresholds; gate_2b (relationship_household) and gate_2c (marriage_earnings_joint) UNLOCKED sibling tranches (locked: false, no floor, no scored run, each with its own future lock ceremony naming the verification round); the certification_scope map promoted into the locked scope; and the standing governance rule description_claims_exactly_the_scored_surface promoted into gate_2.governance.amendment_rules (attributed to review #106 finding 4 / PR #111 finding 8), binding 2b/2c/gate_3. gate_2.name ("family and benefit outputs") retained per its C3 disposition (the tranche container). Also applies the two verification-round flip-time nits (comment 4930753295): the 2b bullet-3 legend quote spacing normalized to "poverty/household" to match the derivation_basis quote; and COLA qualified in the A2 own-record example list (its levels ride the Phase-B claiming/mortality components -- #74 marks CA, M, DI; #78 covers CA, M/DI uncertified -- so it is not a pure gate-1 + oracle surface, though still outside the 2a/2b/2c tranches). Adds gate_2's second amendment_history entry (id 2026-07-09-tranche-split; all four ceremony pointers; content: ZERO THRESHOLD MOVEMENT, verdicts v1-v15 FAIL and v16 PASS stand as tranche-2a verdicts, citing #106 finding 4 and the candidate-16 pass #109), and removes amendment_proposed (preserved in git history at the ratification merge 8a4a240). PROSPECTIVE / STRUCTURAL ONLY (no_self_rescue): a naming + certification-scope amendment has no verdict to rescue and re-scores nothing; candidate 16 stays PASS (4/5), candidates 1-15 stay FAIL, as tranche-2a verdicts. Tests rewired locked-hot: the re-scoped description/holdout_basis, tranche_id, the promoted certification_scope map (per-class requires_tranche + coverage tokens bound EXACTLY, both nits asserted), the gate_2b/gate_2c stubs, the promoted standing rule, and the second history entry with all four pointers rebind at their LIVE post-flip locations in a new "Gate-2 amendment 2 RATIFIED bindings" section. The proposal-object tests stay dormant-safe (guarded on proposal_number == 2, so they skip once amendment_proposed is gone), and the D1 merge-safe guard naturally skips post-flip. Full suite: 1319 passed, 68 skipped, 0 failed (-n 8); black -l 79 + ruff clean. Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The flip PR for amendment 2, mirroring #59's pattern for amendment 1. The ceremony completed on #67: adversarial referee round (AMEND BEFORE RATIFYING, six findings), fixes (
ae0c166— prospective-only 20-seed mean rule, no-self-rescue, version-matched cap), verification round (RATIFY AS-IS, every number recomputed and bindings perturbation-tested), ratification by merge (4e06e244) under the maintainer's standing delegation.What goes live:
c2st_mean_rule: pairs-view c2st gates on the mean over pre-registered seeds 0–19 at the unchanged 0.53 line (same floor, same k=4.2 derivation — estimator change only), with the operating-characteristics table showing it is stricter above the line than the per-seed rule (pass prob at true mean 0.533: 0.01 vs 0.03).c2st_per_seed_cap: 0.554 = round(matched-floor mean + 8·sd, 3) from the sklearn-1.8.0 floor inruns/c10_diagnostics_v1.json.amendment_rules(standing):no_self_rescue— no candidate's committed verdict changes under a rule proposed after its own run — and the classifier version pin.Tests: 403 passed, 34 skipped (the 11 proposal-object tests go dormant, as designed); stored-thresholds tests across all 12 artifacts extended with the
per_seed_rule_supersededpop (amendment-1 pattern); 6 new tests bind the live rules, recompute the OC table, and assert the no-verdict-change guarantee directly against the artifacts.🤖 Generated with Claude Code