Soundness Coverage
Fuzz a circuit's constraint system to find signals an attacker could set freely — the underconstrained inputs that break soundness.
Run inputs
The full session is JSON-encoded and sent to the stateless endpoint — no upload, no storage. Edit the JSON to point at your own constraint system, or click Try Me to load the canonical underconstrained fixture.
- Targets
- 0
- Underconstrained
- 0
- Well constrained (so far)
- 0
- Coverage
- —
- Candidates
- 0
- Violations
- 0
- Elapsed
- —
Per-target heatmap
One tile per target signal. Red tiles flag underconstrained signals — those are real audit findings; green tiles only indicate the fuzzer failed to disprove soundness within the configured candidate budget.
No run yet — load a sample above and click Run coverage.
Exhibits
Each row is a single candidate witness that satisfies every constraint while disagreeing with the honest baseline on the listed signals.
| # | Signals diverged | Honest → Candidate | Constraints | Note |
|---|
Caveats
The fuzzer works in a small reduced field for tractability — BN254 / BLS12-381 arithmetic is delegated to the proof-system layer when an exhibit is forwarded. The constraint shape is R1CS (a * b = c). Inputs are capped at 4096 constraints, 256 target signals, and 4096 candidates per run.
Sign in to run soundness coverage
Use the Sign In button in the navigation bar.