Provable green
Every LOOP.md line is checked, not trusted.
VERIFIED
Deterministic loop verification
The trust layer forAI coding loops.
Sigil cross-checks your agent's log, git history, and CI runs — line by line — into one deterministic Loop Integrity Score. The log proves itself.
Trusting an AI coding loop is hard
when most tools just take the agent's word for it.
So we made it simple to compare how Sigil verifies versus what every other tool assumes.
Other tools
Trusts the agent's self-report
No cross-check against git history
Screenshots and vibes, not verdicts
“Green” means the agent said so
The log can't prove itself
Sigil
Cross-checks three independent sources
Every line carries a SHA + run ID
Per-line: verified / unverifiable
“Green” means CI actually passed
A self-proving LOOP.md
agent log
git history
CI run
Cross-checked, not trusted
The agent's log, the git history, and the CI run are reconciled against each other — line by line — so no single source gets the last word.
Reproducible by design
Same inputs, same score — every single time.
3
SOURCES
98
INTEGRITY
0
CONTRADICTED
Line-by-line
one verdict per line, with its SHA
Built on
GitHubTestSpriteNeonVercelNext.jsPostgresSSELOOP.mdGitHubTestSpriteNeonVercelNext.jsPostgresSSELOOP.md
LIVE ON GITHUB
98/100
Loop Integrity
PRODUCT
Verify, replay,
and prove every loop
Line-by-Line Verification
Every LOOP.md line is checked against its git SHA and TestSprite run verdict, then resolves to verified, unverifiable, or contradicted. The methodology is public.
Live Replay
Scrub the whole loop end to end — time-to-green, banked-suite growth, and per-fix diffs — streamed live over SSE as each run lands.
Badge & Season Index
Ship a README badge and a public proof page, then see how loops break across the season with failure fingerprints. Celebration, never policing.
How it works
From plan to green
01
02
03
04
05
Step
01
Plan
Write the plan file first. No implementation before a failing test exists.
Plan file
Claim Your Badge01/05
Red
Create the failing test and confirm it fails for the expected reason.
Failing testExpected red
Claim Your Badge01/05
Implement
Build straight from the failure bundle — the exact assertion that is red.
From the red
Claim Your Badge01/05
Commit
The commit carries the run ID, so the LOOP.md line proves itself.
SHARun ID
Claim Your Badge01/05
METHODOLOGY