Skip to content

Audit checklist: 6 more sections (determinism, edge cases, offline, logging, deps, license)#34

Merged
lucapinello merged 1 commit intochorus-applicationsfrom
docs/audit-checklist-additions
Apr 21, 2026
Merged

Audit checklist: 6 more sections (determinism, edge cases, offline, logging, deps, license)#34
lucapinello merged 1 commit intochorus-applicationsfrom
docs/audit-checklist-additions

Conversation

@lucapinello
Copy link
Copy Markdown
Contributor

Summary

Follows up v18 (PR #33, merged) by adding six sections to audits/AUDIT_CHECKLIST.md that a real release audit needs but the first pass skipped:

  • §13 Scientific determinism — same input twice → same output; ±0.006 cross-machine band documented.
  • §14 Genomics edge cases — variant near telomere, soft-masked FASTA, multi-allelic, indels (error before running the model), chrM / chrY.
  • §15 Offline / air-gapped — predict + gene-expression with no network; one-line grep for CDN refs in shipped HTMLs.
  • §16 Logging hygiene — grep pattern for leaked HF tokens / AWS keys in any committed output.
  • §17 Dependency supply chainenvironment.yml pins, pip-audit, per-oracle editable install consistency.
  • §18 License / attributionLICENSE, third-party model credits, bundled IGV JS license header.

Deliberately excluded because they're out of scope for a scientific-lib release by a small team: fuzzing, Windows, concurrency benchmarks, Docker, accessibility testing. The user's constraint — "don't add things that would be unreasonable to test" — shaped this.

Every added item is either a one-line grep/script or a clear yes/no human check. No make-work.

Test plan

  • Doc-only change; pytest unaffected
  • grep -n '^## ' audits/AUDIT_CHECKLIST.md confirms the new sections land where expected

🤖 Generated with Claude Code

…ffline, logging, deps, license)

Extends audits/AUDIT_CHECKLIST.md with sections the v18 version skipped
but that actually matter for a scientific library release:

- §13 Scientific determinism — same input twice → same output,
  known-good ±0.006 CPU non-det band across machines.
- §14 Genomics edge cases — variant near telomere, soft-masked FASTA,
  multi-allelic, indels (error before model run, not after), chrM/chrY.
- §15 Offline / air-gapped — predict/analyze_gene_expression without
  network; grep for CDN references in shipped HTMLs.
- §16 Logging hygiene — no HF_TOKEN or AWS keys in committed outputs
  (grep pattern provided).
- §17 Dependency supply chain — env.yml pins, pip-audit, per-oracle
  editable install.
- §18 License / attribution — LICENSE, third-party model credits,
  bundled vendor JS license headers.

Deliberately excluded: fuzzing, Windows support, concurrency, Docker,
accessibility — out of scope for a scientific-lib release.

Every added item has either a one-line grep/script or is a yes/no
human check. No new audit-for-the-sake-of-audit items.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@lucapinello lucapinello merged commit 2200ddf into chorus-applications Apr 21, 2026
1 check passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant