KeyFuse cache-key auditor
A read-only browser lab that probes a task's declared inputs for undeclared slot dependence, prints a minimal same-key / different-output witness, and builds a repaired key from everything it detected. No storage, no network, no grading.
KeyFuse probes a task's declared inputs to find which undeclared slots its output actually depends on, then prints a minimal same-key / different-output witness and a repaired key built from everything it detected.
Everything runs in this browser over the frozen 24-task corpus: no storage, no network, no grading. Pick a task, a strategy, and a strength, then run.
Audit result
No audit yet. Pick a task, strategy, and strength, then press Run audit.
Detection is not soundness. A miss can only be reported, and a key repaired from detections is conservative, not complete.
What this is not
- A repaired key is conservative: it contains everything the audit detected, not everything that exists.
- A miss is reported, but it is not evidence that the rest of the universe is irrelevant.
- The sampled domains and the chosen strength bound what can be seen; an effect outside them can escape every arm.
- The short keys are identifiers for this lab, not a security boundary.
- Nothing here is a grade. This page stores nothing, syncs nothing, and sends nothing.