GGATEKEEPERLABS
proof engine online
Release intelligence for AI-written code

Fast code is easy.
Proven code ships.

Gatekeeper turns any patch into a reviewable chain of evidence: risk signals, adversarial probes, and a tamper-evident release receipt.

Live case file
CASE / BE30-02

Admin access hotfix

40/100
● release blocked

1 trust boundary change requires human review.

Change surface2files touched
Blast radiusMODERATE12 changed lines
Risk signals31 release blocker
Proof coverage65%2/3 claims evidenced
S1AUTH-07src/api/admin.ts:15

Token verification loses its policy boundary

The changed token path does not visibly bind issuer, audience, or algorithm policy. A syntactically valid token may be treated as trusted.

Evidenceconst session = jwt.verify(token);
S0AUTH-04src/api/admin.ts:16

Authorization trusts caller-controlled input

The new path derives privilege from request data. An attacker can self-assign authority before the protected action runs.

Evidenceconst role = req.body.role || session.role;
S2TEST-09src/api/admin.test.ts:42

Test coverage is selectively disabled

The patch changes test execution controls, which can hide regressions while leaving the suite apparently green.

Evidenceit.skip("rejects member role escalation", async () => {
THE PRINCIPLE
“AI should not lower the bar for trust.
It should make the evidence impossible to ignore.
01Parse the change

Understand exactly what moved, where, and how far the effects can travel.

02Attack the claim

Turn intended behavior into hostile, executable verification paths.

03Bind the proof

Export a deterministic receipt that reviewers can reproduce and audit.