Certification Readiness
Bounty / certification targets, each traced upward to the derived and true invariants that evidence it.
Overall readiness
82% of referenced invariants enforcedAI authorship
99.5% of lines AI-authored · target 90% ✓ metBy model
| Model | Lines added | % of lines | Commits |
|---|---|---|---|
| Claude (unversioned) | 7,327,035 | 92.6% | 217 |
| Claude Opus 4.7 | 294,188 | 3.7% | 246 |
| Claude Opus 4.8 | 214,534 | 2.7% | 733 |
| Claude Opus 4.6 | 41,863 | 0.5% | 27 |
Lines added per non-merge commit (git log --numstat) attributed to models via Co-Authored-By: trailers; lock files and binary diffs excluded. Co-authored lines split equally when multiple distinct models appear on the same commit; commits without an AI trailer count as human-only.
Sync to tip from verified anchor
evidencedImport a verified Mithril snapshot or replay from genesis, then follow forward to the current tip with byte-identical, replay-equivalent state.
Block production
evidencedForge a valid block: leader check, KES-signed real header, single canonical envelope, self-accept before serve.
N2N / N2C protocol coverage
partly evidencedClosed mini-protocol grammar with byte-identical round-trip across the N2N and N2C taxonomies.
Private testnet readiness
mostly evidencedOperator-pass evidence on a private testnet, with a closed evidence manifest schema.
Tx validity agreement
evidencedPer-transaction validity verdicts agree with the reference node across the Conway corpus.
Block validity agreement
evidencedBlock-level validity verdicts agree with the reference node.
Consensus tip agreement
partly evidencedFork-choice selects the same tip as the reference node under the same inputs.
Power-loss recovery
evidencedRestart after a crash produces byte-identical state to a clean run (replay-equivalent recovery).
Memory measurement
mostly evidencedBounded, measured memory footprint on the authoritative paths.