How LENITNES turns commits into accountable judgments
One proof-chained loop, two applications: [markets] produces market theses with trade safety gates; [research] produces alert-only integrity judgments graded against adjudicated record events.
● watching public commits…
Data & Detection
Continuous commit monitoring across consensus code repositories and 9 typed rule-based classifiers.
Admin-curated — the agent cannot expand its own watchlist. Full list & live status →
Heuristic pass before the LLM — each detector scores 0–100 with confidence; a commit can trip several. News only corroborates.
Urgency language (HOTFIX, urgent, critical) on release branches.
Security-sensitive paths (validation, consensus, crypto) + vulnerability class.
Diff touches consensus-critical paths — chainparams, block validation, signatures.
Versioned protocol change — soft-fork, hard-fork, BIP/EIP references.
Maintainer set changing, governance docs rewritten, signer rotations.
Dependency downgrades, unknown maintainers — supply-chain incident patterns.
Lockfile churn, deprecation notices, post-incident library swaps.
Contributor inactivity >30d or removal from CODEOWNERS files.
News sentiment corroboration only. Hard-capped below trading threshold.
The agent aggregates detector scores, diff evidence, 7-day price context, and past outcomes into one verdict.
Repos form causal chains; the agent sees upstream events in the same sector (7d window) before scoring each commit batch.