Repository navigation
Conversation
…hare simulation Quarto manuscript (PolicyBench-style) with fact catalog mapping every quantitative claim to a committed artifact, frozen evidence snapshot (certification report, simulation results, decomposition classifications), and references. Draft for adversarial referee review. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…vidence-structure and data-authenticity caveats Independent auditor rebuilt the engine byte-identically and reproduced FULL_RUN.json byte-identically; certification stands on reproduction. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Ten reports (4 code auditors, 6 referees; reproducibility pending): certification independently reproduced byte-identically; JS draw path proven bitwise-equal to Python; pipeline sound-with-fixes; app model mode pulled pending statistical fixes; manuscript = major revisions across ten themes (EDITORIAL.md). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- Parity reframed: Minimodel chain on edited records, 6,081 of 6,194 in-scope with exclusions tabulated; 'reviewer arithmetic' framing removed; cross-model character acknowledged - Direction errors fixed: coverage is one-sided UNDER-coverage; SMD cross-section attribution corrected to actual cells - Domain additions: PUF truncation of ineligible-case error, QC integrity history (FCA settlements), negative-case/underpayment asymmetries, no-dispute election, OBBBA 10103 heat-and-eat narrowing, mechanical delay-clause test, admin-match cut, bundle caveat on SMD - Numbers from committed corrected artifact (official gate, CASE==1): boundary probabilities, audit deltas, 609M/1,310M bounds - Neutrality rewrites; tables (verification, decomposition, validation, boundary); shrinkage formula; matched frozen pairs; both correlations; case-vs-dollar units; solo-author voice; disclosure statement; new citations (CC'21 French tax compiler, rearrangement, Kane-Staiger, Herd-Moynihan, CRS, FNS 310) with DOIs throughout - Snapshot: per-case layer-1/3 outputs, reconstruction script (path made overridable), mc generator, corrected simulation artifact - FACTS.md rows re-pointed to committed artifacts; stale values fixed Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…(PR #10) Makes the manuscript's 'now corrected' and 'disabled' statements true of the branch of record, per round-2 referees. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…, SSI-CAP taxonomy, figure, reproduction docs - Decomposition rewritten around the 110/195 threshold split (round-2 methodology): 78.4% of above-threshold official errors explained vs 91.4% of sub-threshold deviations; abstract leads with the official- error figure; computation-side upper bound among official errors is 21.6%; official-gate file rate 7.42% and sub-threshold dollar share (.5M of .6M) stated - FCA sentence corrected to DOJ-verified figures: >M, eight states, 2017-2021, cited to the releases and CRS R45147 (round-2 red-team; both prior transcriptions were wrong) - Exclusion taxonomy corrected: all 113 are SSI-CAP standardized-benefit units (new error introduced by the prior revision, now fixed) - Abstract: modifier placement, errata grammar, model-vs-simulator arc, file-recorded-intermediates scope clause - Truncation-direction claim argued and softened with the countervailing stratification component named; delay threshold stated exactly (above 13.33%); LA/MI zero-event contrasts labeled; $44M; 83.64%; 0.52 frozen correlation; byte-for-byte scope; oracle defined; ROC AUC expanded; conclusion self-assessment trimmed - Colorado measured-rate figure added (committed generator, seed 11) - README 'Reproducing the paper' section; .python-version pin; snapshot mc_tool.py gate corrected to match its artifact; provenance stanza for results_by_state_corrected.json; layer-3 national CSV committed; FACTS section H (round-2 verified system facts) and row corrections; DOJ/CRS bib entries; giannella commit pinned - Merged main so the corrected loader and rebuilt data.json are ancestors of the manuscript commit Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
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
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.
Quarto manuscript (paper/index.qmd): the QC-as-oracle verification result with its precise scope, the three-layer error decomposition with the threshold split, the corrected model stack with validation failures reported, and the cost-share simulation findings — plus the complete adversarial apparatus: FACTS.md claim-to-artifact catalog, frozen evidence snapshots (per-case decomposition outputs, certification report, corrected simulation artifact with provenance), and all fourteen review reports across two rounds, archived unedited.
Rendered HTML+PDF published at /paper/ on the site; footer links it. Includes the main merge (corrected loader ancestry), README reproduction section, .python-version pin, and the committed figure generator.
🤖 Generated with Claude Code