Methodology
How Sigil turns a LOOP.md into a score. Every number below is read straight from the engine, so this page can never drift from the code that runs.
The formula
A contradicted line costs 25 points each — the only thing that pulls a proven log below its integrity ceiling.
Per-line verdicts
Every applicable cross-check confirmed the line against git and run history.
The line carries no machine-checkable anchor (no commit SHA or run ID). Marked neutrally; it never lowers the score of a log that has verified content.
A source refutes the line — a missing/out-of-order SHA, a mismatched run verdict, an incoherent timestamp, or a pass that did not bank.
The run reported 'blocked' while every assertion passed — a known checker issue. Counted as passing, never as a failure, and never hidden.
Cross-checks
Each machine-checkable line is run through every applicable check against git and TestSprite run history.
The claimed commit SHA exists in git history.
Commits land in chronological order across the loop.
The claimed run ID exists in run history.
The recorded run verdict matches the line's claim.
The fix commit lands between the FAIL run and the PASS run.
A claimed banked pass is not refuted by a later failing run.
Principles
- Unverifiable lines are neutral: they lower Coverage, never the score of a log that has verified content.
- Anti-gaming: a log with zero checkable lines scores 0, not 100 — an unproven log is not a trusted one. This is not a penalty on individual unverifiable lines.
- A 'blocked' verdict with passing assertions is classified as a verdict mismatch (known checker issue), never as a failure.