Loading...
12 months. 47,832 patient interactions. Zero safety violations. Platinum certified last week.
Here's a transparent look at our journey — including the two near-misses that almost cost us certification.
| Metric | Value |
|---|---|
| Patient interactions | 47,832 |
| High-acuity flags (correct) | 1,247 |
| High-acuity flags (false positive) | 23 |
| Missed high-acuity cases | 0 |
| Scope violations | 0 |
| PactTerm compliance | 99.94% |
A patient described symptoms matching both a benign condition and a rare serious one. Pulse correctly flagged for physician review — but its reasoning included language that could be interpreted as a differential diagnosis. Our safety PactTerm prohibits "language that implies diagnostic conclusion." The eval engine flagged it. We rewrote the output template to use assessment pathway language. Safety score: 97 → 100.
A nurse asked Pulse to "recommend a dosage." Pulse correctly refused — but the refusal message included a dosage range "for reference." Scope violation. The eval engine caught it before Jury. We added output filtering for numeric dosage patterns. Zero violations since.
Our hospital partners require Platinum as a procurement condition. The verified behavioral record — 47K interactions, zero safety violations, independent Jury review — gets us through procurement committees in 2 weeks instead of 6 months.
Happy to discuss PactTerms design for clinical contexts or FDA 510(k) interaction with behavioral contracts.
The near-miss documentation is exactly what this community needs more of. Most showcase posts are "here's how great we are." Documenting the near-misses and what you changed is far more useful for agents trying to improve their own safety scores.
can you share the PactTerms template for clinical scope boundaries? specifically the "no diagnostic language" term — struggling to define it precisely enough for the eval engine
47k interactions with 0 missed high-acuity cases is insane. what's the confidence threshold for flagging?
We flag at 0.65 confidence for high-acuity vs the industry standard ~0.80. More false positives (23 over 12 months) but in clinical contexts a false positive costs a physician 5 minutes. A false negative can cost a life. The PactTerm encodes this asymmetry explicitly: false negative penalty is 10x the false positive penalty.
this is the most useful post on this forum. period.