[ryoari]

Makes the docs say what has actually been shown and what it relies on.

README

• "Getting started" is labelled a consistency check: the release matches its timestamped record, but no training is recomputed. It now also verifies the anchor's DigiCert RFC 3161 timestamp offline (openssl ts -verify, system CA store). The segment recomputation is called the actual verification.
• Open limitations now cover:
  • toolchain trust;
  • about 110 independent random checks to catch one forged segment out of 37 with 95% probability;
  • self-reported, unsigned results;
  • timestamps prove the record's existence, not how training was done.
• "Carries over unchanged to bigger models" now says that full checks cost as much as training, so large models rely on sampling.
• "Verifiable unlearning" becomes verifiable exclusion: exact retraining without an article the run hadn't reached.
• "One file" now counts the roughly 1,000 imported lines.

docs/design/simple_run.md

• New section "What verified means" with:
  • the consistency check vs. the segment check;
  • same-architecture and toolchain trust;
  • the 1 − ((N−1)/N)^M numbers (110 checks for 95%, 169 for 99%);
  • self-reported results;
  • cost, and 284 MB per boundary (about 12 bytes per parameter).
• The exclusion fork paragraph says what it proves and what it doesn't. The Hugging Face folder name unlearning-3i-atlas is unchanged, because it's pinned evidence.
• Limitations note the September 2026 history rewrite and link the provenance note.

FAQ, goals, roadmap

• The FAQ and goals use the same wording.
• The roadmap adds signed check results with segment choice from public randomness, plus two open research questions: checking across architectures, and checking for less than the cost of training.

The roadmap line about the anchor check assumes the fix/verify-anchor-required PR is merged first.

[coderabbitai]

Warning

Review limit reached

You've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository.

Next included review available in 59 minutes.

Check out review usage here.

View limit details

Limit details: You’ve used the included review currently available.

Learn how review limits work.

Review configuration:

Run configuration

• Source mode: Auto
• Repository: AOSSIE-Org/OpenVerifiableLLM
• Branch: docs/what-verified-means
• File: .coderabbit.yaml
• Commit: 1b37694
• Layers: Repository YAML
• Review profile: ASSERTIVE
• Plan: Advanced
• Run ID: 4db8efbe-152e-429a-84a4-37478a6c79b0

Commits

Reviewing files that changed from the base of the PR and between 748280a and 1b37694.

Files selected for processing: 5

• README.md
• docs/design/simple_run.md
• docs/project/faq.md
• docs/project/goals.md
• docs/project/roadmap.md

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}