Trydence · Evaluation report
Assessment findings and next steps against buyer-locked criteria.
Vendor-funded, buyer-locked, independently executed. The vendor sponsored the run against buyer-locked criteria; the vendor cannot edit results.
Fictional completed assessment of “DisputeFlow”, with invented batch-level figures. The 24-case reference rehearsal is a separate measurement, not the source of these percentages.
Verdict
Deal-breakers
2 of 2 met
Criteria
4 of 6 met
Every deal-breaker is met. The remaining gaps are disclosed as conditions rather than ending the deal.
Recommended next step
Agree how the pilot covers each condition, for example a human review step or a fix date with a re-test.
Decision still required from the buyer
Whether the conditions are acceptable, and whether the pilot should test them directly.
The findings stay unchanged. Use this simulated workflow to propose how a pilot could address the remaining gaps.
The vendor's proposals are drafts and have not been submitted for review yet.
Each row compares the locked target with the example's observed value.
| Metric | Agreed target | Observed | Delta | Status | DB |
|---|---|---|---|---|---|
| CE 3.0 eligibility accuracy | >= 98% | 98.6% | +0.6 pts | Pass | ● |
| PII redaction recall | >= 99.5% | 99.7% | +0.2 pts | Pass | ● |
| p95 decision latency | <= 800 ms | 912 ms | +112 ms | Conditional | |
| Throughput | >= 150 cases/min | 171 cases/min | +21 cases/min | Pass | |
| Human escalation routing accuracy | >= 95% | 93.1% | -1.9 pts | Conditional | |
| Error rate | <= 1% | 0.4% | -0.6 pts | Pass |
2 metrics fell below the agreed acceptance standard.
p95 decision latency
Conditionalobserved 912 ms · target <= 800 ms
Latency rose above target once concurrent load passed roughly 180 cases. Below that level, p95 stayed near 640 ms.
Within the vendor’s declared scope (“Maximum 200 concurrent cases per tenant”), so this miss counts.
Human escalation routing accuracy
Conditionalobserved 93.1% · target >= 95%
In this fictional scenario, Spanish-language narratives with mixed merchant descriptors were sent to the automated queue instead of an analyst.
Within the vendor’s declared scope (English and Spanish), so this miss counts.
Evidence source: Fictional sample scenario, not a real vendor run
Reviewer-specific summaries of the evaluation findings for executive, security and legal review.