Phase 3B · Calibration
Benchmark & calibration
Validate exact, near-exact, paraphrase, attributed quotation and original-writing behavior against a known synthetic source using the same live retrieval pipeline as real submissions.
Pass rate
—
target 100%
Retrieval recall
—
target 100%
Precision
—
reportable evidence precision
False-positive rate
—
target 0%
Cases passed
—
controlled ground truth
Original false positive
—
must remain NO
3B.1 readiness
—
5/5 + recall 100% + FPR 0%
Controlled cases
| Case | Detected | Citation | Confidence | Evidence signals | Raw / adjusted | Result |
|---|---|---|---|---|---|---|
Run the controlled benchmark to generate calibration evidence. | ||||||