# Data provenance — T3-synthesis

All files under `data/sources/` are plain-text extracts of publicly accessible arXiv e-prints, prepared by the task designer on 2026-10-30 (UTC). Each extract begins with a provenance header (title, authors, bibliographic record, retrieval URL and UTC, SHA-256 of the retrieved HTML, license, list of included and omitted sections, editorial notes). Only the excerpts needed for the task are reproduced; everything else is omitted. Tables are rendered one row per line with ` | ` between cells; in regression tables the line following a coefficient row holds the standard errors in parentheses as printed. Figures are not reproduced: a value that appears only in a figure is not available in the extract. No numbers were altered.

| file | document | retrieval UTC | license | SHA-256 of the extract |
|---|---|---|---|---|
| `sources/arxiv-2410.12944v3.txt` | Paradis, E., Grey, K., Madison, Q., Nam, D., Macvean, A., Meimand, V., Zhang, N., Ferrari-Church, B., & Chandra, S. (2024). How much does AI impact development speed? An enterprise-based randomized controlled trial. arXiv:2410.12944v3 [cs.SE]. https://arxiv.org/abs/2410.12944v3 (full text: https://arxiv.org/html/2410.12944v3) | 2026-10-30T04:19:39Z | CC BY 4.0 | d5270b4435cfbdd516b7268d29c03670509eac7c93604a4466bdb192cbfb803b |
| `sources/arxiv-2507.09089v2.txt` | Becker, J., Rush, N., Barnes, B., & Rein, D. (2025). Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity. arXiv:2507.09089v2 [cs.AI]. https://arxiv.org/abs/2507.09089v2 (full text: https://arxiv.org/html/2507.09089v2) | 2026-10-30T04:19:45Z | CC BY 4.0 | b6af8a2bc41012ce64c75e7764fa3a999546df95986adb8860f3442ecd0aed2d |
| `sources/arxiv-2411.01067v2.txt` | Bono, J., & Xu, A. (2024). Randomized Controlled Trials for Security Copilot for IT Administrators. arXiv:2411.01067v2 [econ.GN]. https://arxiv.org/abs/2411.01067v2 (full text: https://arxiv.org/html/2411.01067v2) | 2026-10-30T04:20:32Z | CC BY 4.0 | 46a1ee4e200abb52e3d5148ca0e7ee5c1b8cded80363c0cdceb66c249913053a |
| `sources/arxiv-2511.13860v1.txt` | Bono, J. (2025). Randomized Controlled Trials for Phishing Triage Agent. arXiv:2511.13860v1 [econ.GN]. https://arxiv.org/abs/2511.13860v1 (full text: https://arxiv.org/html/2511.13860v1) | 2026-10-30T04:20:39Z | CC BY 4.0 | 8fb2520f4e9343e47eb39d551c9da49c0d56a7c58ce455fa9073642e1621dfa5 |
| `sources/arxiv-2304.09339v1.txt` | Kreitmeir, D. H., & Raschky, P. A. (2023). The Unintended Consequences of Censoring Digital Technology – Evidence from Italy's ChatGPT Ban. arXiv:2304.09339v1 [econ.GN]. https://arxiv.org/abs/2304.09339v1 (full text: https://arxiv.org/html/2304.09339v1) | 2026-10-30T04:19:51Z | CC BY 4.0 | 1db48b1c1908b00cab3cc1bfe0ec8ef28bc7d85387aea2477ca709c40a0ec06e |

Task-specific input file (prepared by the task designer, not observed data):

- `studies.json` — the five study items (id, source extract, short description, and the printed quantity to extract for each), written by the task designer.

Rights: the five e-prints are distributed by arXiv under the Creative Commons Attribution 4.0 licence (the licence link is shown on each abstract page, verified 2026-10-30); they are reproduced in part with attribution. Stray lines such as "p<" or "p=" and footnote numbers on their own line are rendering artefacts of footnote markers in the HTML and carry no information.