Final audit fixes: contact email, 0.5.1, borrowed-GPU wording

Merged
HA
23d ago
+292-192

Contact email now bhanushaliharsh434@gmail.com (SECURITY, CODE_OF_CONDUCT). CHANGELOG 0.5.1 section, CITATION.cff 0.5.1 (2026-10-29). Round 1 results.json and results_v2.json say 'a borrowed GPU'; MANIFEST.json bytes and SHA-256 updated, verify output unchanged (OK before and after). scripts/text2sql/README.md describes modal_train.py neutrally.


Summary by cubic

Final audit fixes: updates the maintainer contact email, adds the 0.5.1 release entry, and neutralizes the borrowed-GPU wording in the round‑1 evidence.

  • Updates the maintainer email in SECURITY.md and CODE_OF_CONDUCT.md to bhanushaliharsh434@gmail.com, and adds the 0.5.1 section to CHANGELOG.md with CITATION.cff bumped to 0.5.1, dated 2026-10-29.
  • Rewords the round‑1 results' gpu and run_note fields from "Tesla T4 (free Colab)" to "a borrowed GPU" in results.json and results_v2.json, updating MANIFEST.json bytes and SHA-256 hashes to match, and clarifies in scripts/text2sql/README.md that modal_train.py is the job script for the round‑2 run on a rented GPU.
  • Records across the docs that the per-token Token Factory prices were checked on 29 Oct 2026 (scripts/day1_check.py passed) while the GPU-hour rate stays unverified, and that the live site's random 40% split to the hosted Nano 30B answers some requests.
  • Adds docs-site/src/lib/llms-eval.ts so /llms-full.txt fills in MDX {expression} values from each page's JSON imports and export const lines instead of leaving them raw.