You’re exploring an example. Saved results for a support assistant. No live AI calls.Use your own project
SUPPORT ASSISTANT / EXAMPLE REPORT

See what your tests catch.

12 rules checked against 9 tests. Start with the gaps below.

Download report
JSON data Full HTML report ↗
5 / 12 rules

Removal detected

Tests caught the missing instruction

4 / 12 rules

Removal missed

Tests passed with the instruction missing

3 / 12 rules

No test linked

Add a test for these rules

START HERE

5 high-priority rules need review.

The clearest place to start is a test that kept passing after its rule was removed. Open it to see the before-and-after evidence.

Your rules

Needs attention first. Select a rule to see its result.

12 of 12 rules · sorted by priority
How this report was calculated

Each linked test was repeated 3 times before and after removing one instruction. 9 of 12 rules have a linked test (trace coverage: 75%). 5 of 12 had removals detected reliably (causal rule coverage: 42%).

These are precomputed results from a deterministic example. They explain the method, not the performance of a live model. Read the method and thresholds →