Does AI Help? October 2026 affiliation patch

Affiliation-only presentation patch: Jit Team appears beneath Witold Bołt's credit on the PDF title page and alongside the front-page web byline. Findings, source database and 13 October 2026 cutoff are unchanged. The final editorial checkpoint october-2026-final and all earlier versioned artifacts remain intact.

Does AI Help? Final October 2026 release

Final October baseline, version 2026.10.14, evidence cutoff 13 October 2026. The final editorial self-review is recorded in manuscript/october-final-review.md. Corrections distinguish randomized and quasi-experimental evidence, include mixed customer-support subgroup outcomes and learning evidence, and clarify the scope of the mathematical translation critique. Tags v2026.10.14 and october-2026-final identify the same publication commit. November preparation proceeds separately on codex/november-2026-preparation.

Does AI Help? October 2026, version 2026.10.13

October publication baseline with reviewed topic classifications, concise article references, and shared GitHub star navigation. Evidence cutoff: 13 October 2026.

Does AI Help? October 2026, version 2026.10.12

Project-specific full-version filenames for all advertised downloads: does-ai-work bibliography files; does-ai-help evidence, citation, PDF, archive, methods and verification files. Filtered exports follow the same release naming scheme. Includes a generated filename manifest, backward-compatible generic URLs, and archive boundaries that exclude older versioned downloads. Research findings, source counts and the 13 October cutoff are unchanged.

Does AI Help? October 2026, version 2026.10.11

Adds the Leiden Declaration on Artificial Intelligence and Mathematics with 16 registered authors, original 2 June 2026 date, Zenodo DOI and separate arXiv version history. Verified IMU endorsement and added qualified mathematical-governance context, distinct from empirical productivity findings. Updated evidence exports, web article and PDF. Cutoff remains 13 October 2026; previous release archives are preserved.

Does AI Help? October 2026, version 2026.10.10

Adds KPMG Global AI Pulse Q3 2026 as a primary source with its three named authors, survey methodology, qualified value-versus-cost findings and a page-referenced appraisal. Retains the June Q2 announcement as a separate related wave. Includes updated web article, evidence database, citation exports and PDF. Evidence cutoff remains 13 October 2026; earlier archives are preserved. Third-party PDF is not redistributed.

Does AI Help? October 2026, version 2026.10.9

Presentation-only patch: BibTeX uses 12-character repository commit IDs. Full hashes remain in canonical evidence records, CSL-JSON and RIS; document SHA-256 checksums are preserved. Includes the web citation page and back-to-top navigation. Findings, source counts and the 13 October 2026 cutoff are unchanged. Earlier snapshots remain immutable.

Does AI Help? October 2026, version 2026.10.8

Corrects fifteen missing author credits using primary bylines, institutional credits and recommended citations. Preserves publishers, quoted spokespersons, registered creators and later endorsers as distinct roles. Math and AI displays Terence Tao et al.; exports preserve the registered 25-creator list. Anthropic follows its seven-author recommended citation, while KPMG receives institutional credit. The already included February 2026 METR follow-up is explicitly linked without replacing the 2025 experiment. Includes authorship provenance and refreshed bibliography/PDF. No changes to findings, source counts or the 13 October 2026 cutoff.

Does AI Help? October 2026, version 2026.10.7

Replaces four undated Gravity, Zig and Asahi entries with repository-backed document dates. Includes pinned source revisions, date basis, original commit timestamps and distinctions between source-declared publication, Git revision and policy adoption. Refreshes citation exports and PDF reference typography without changing findings or source counts. Evidence cutoff remains 13 October 2026; earlier snapshots preserved.

Does AI Help? October 2026, version 2026.10.6

Adds verified open-source contribution-policy contrasts, the reported Torvalds debugging episode, the Asahi–Gravity policy split, and a critical appraisal of recent policy research. Corrects QEMU repository-date provenance and supplies the SWE-bench authors recommended citations. Includes 227 records, 211 checked identities, 121 annotated sources and 118 article citations. Evidence cutoff: 13 October 2026. Earlier snapshots preserved. Not peer reviewed; policy evidence is not a causal productivity estimate.