01 AI SRE evaluation, without the theatre
Can your agent find
the real failure?
Investigate a production incident, write the RCA, and watch a transparent evaluator measure root cause, evidence, remediation, and red-herring resistance.
Enter live incident ↘
NO LLM JUDGE Every point is inspectable.
02 Live investigation chamber
Follow the evidence.
Then prove your diagnosis.
Choose an incident. Inspect the alert, metrics, logs, traces, and runbook. Write an RCA—or load an example—and run the exact scoring logic used by the Python package.
SERVICE
Loading incident bundle…
Preparing deterministic evidence.
USER IMPACT—
EVIDENCE ARRAY— ITEMS
LIVE FIXTURE
CANDIDATE RCAWRITE OR LOAD A REPORT
0 WORDS
No report scored yet.
OVERALL
—
AWAITING RCA
Score a report to inspect gaps.
Score a report to inspect penalties.
Reveal the ground truth only after investigating ↘
ACTUAL FAILURE
—
03 The scoring engine
Not a vibe check.
A measurable contract.
Generate
Every scenario creates repeatable alerts, metrics, logs, traces, runbooks, noise, and hidden truth from a fixed seed.
Investigate
The candidate receives the same evidence an operator would inspect. Misleading but plausible clues are included on purpose.
Score
Terms and evidence signals map to explicit weights. Blaming a red herring removes points. No opaque evaluator decides the outcome.
Improve
Missing signals and remediation terms turn a single score into a useful explanation of what the RCA failed to prove.