Predictive Audit Prompting Pilot: Audit Chain Log F22-F26 and Supplementary Materials

This dataset contains the supplementary materials for the Phase 1 pilotstudy of predictive audit prompting documented in the companion preprint"Structural Regularities in AI-Assisted Audit Chains: A Pilot Study inPredictive Finding Localization" (Arfeen 2026, Zenodo DOI10.5281/zenodo.20093407, https://zenodo.org/records/20093407). Contents: 1. audit_chain_log_F22_F26.csv (5 rows, 18 columns): the structuredaudit-chain log containing five findings (F22 through F26) drawn fromthe Gate 11 Sigstore-CASM verifier pipeline of the furqan-lint staticanalyzer (versions v0.11.2 through v0.11.6, May 2026). Each rowdocuments a closed audit finding with: finding ID, counter (A global orB phase-scoped), global round number, phase round number, ISO-8601closure timestamp, repository slug, semver tag at closure, pipelinestep where the failure surfaced, canonical error code, substrate layer,scope class, severity (CRITICAL/HIGH/MEDIUM/LOW/ADVISORY), fix size inlines of code, fix size class (one_word, few_lines, table, refactor),fix location, root-cause mechanism statement, closure shape, andpipe-separated list of attributed structural regularities. 2. experiment_protocol.md: the pre-registered scoring rubric for thePhase 1 predict-then-reveal experiment. Specifies the five scoringdimensions (P1 failing step, P2 error code, P3 substrate layer, P4fix-size class, P5 root cause), the HIT criterion per dimension, theTIGHT threshold (>= 3/5 HIT), the predict-then-reveal procedure, thepredictor disclosure (Anthropic Claude Opus 4 with shared audit-chaincontext as same-instance contamination), and pre-registration timing. 3. test1_F25_predictions.md and test1_F25_scoring.md: the committedpredictions for finding F25 generated from the F22-F24 sub-chain, andthe scoring document showing 5/5 HIT with per-dimension attribution tostructural regularities R2, R3, R5, R6 plus partial LLM priorcontribution on P5. 4. test2_F26_predictions.md and test2_F26_scoring.md: the committedpredictions for finding F26 generated from the F22-F25 sub-chain, andthe scoring document showing 5/5 HIT with per-dimension attribution tostructural regularities R2, R7, R9, R11 plus R4 fix-size bimodality. 5. corpus_accounting.md: the reconciliation between Counter A (global,framework-mandated, monotonic, rounds 1-37) and Counter B (phase-scoped,parochial, rounds 28-33 within the Gate 11 corrective sequence).Documents the 43-round audit corpus structure across three repositories(furqan-lint, Bayyinah-Integrity-Scanner, Furqan-language) underframework versions 1-4. Justifies the paper's sample-size and corpus-breadth claims. 6. README.md: navigation guide for the bundle, mapping each file to itsrole and providing instructions for reproducing the predict-then-revealexperiment. 7. LICENSE: CC BY-NC 4.0 license terms. The eleven structural regularities R1-R11 extracted from the audit chainare documented in the companion paper sections 4.1 through 4.3 and inAppendix B of the paper. They are testable propositional rules whosepredictive value is the subject of the pre-registered Phase 2 cross-vendor ablation study (registered on the Open Science Framework; DOI tobe added to this record once the OSF pre-registration is published fromembargo). The data is published to support independent verification of theregularities, replication of the predict-then-reveal experiment, andextension to cross-instance and cross-project ablation in subsequentresearch phases. Methodology basis: Bayyinah Audit Framework v3.0 (Arfeen 2026, ZenodoDOI 10.5281/zenodo.20017335, CC BY-NC 4.0). Application Brief: Arfeen2026, Zenodo DOI 10.5281/zenodo.20031395.