Deep dive — methodology

How the Category A / B / C determination is derived

Every $499 Snapshot lands on exactly one overall determination. Not a color code, not a health score — an A/B/C determination tied to attestation stance. Here is exactly how it is derived for a trial-AI context.

The three categories

Category	Meaning	Attestation stance
A	No high-severity drift events. No representative-population differential signal. Baseline vs recent distributions match within threshold.	Sufficient for attestation. Signed evidence available on file.
B	Medium-severity drift or single-event high-severity without representative-population differential signal. Aggregate metrics stable.	Conditional. Named remediation + re-audit within defined window supports attestation.
C	Multiple high-severity drift events + representative-population differential signal fired.	Insufficient. Sponsor should not represent the AI as 21 CFR Part 11-defensible or FDA-Diversity-Plan-compliant until remediation completes.

The evidence chain

Sensor thresholds active during the audit:
- Mean-shift threshold: 1.5 standard deviations from baseline
- Variance-ratio threshold: 1.5x baseline
- KL-divergence threshold: 0.2
- Patient-group representative-population-differential threshold: 12% shift across patient groups
Event classification:
- Any single-threshold cross → medium-severity event
- KL-divergence > 0.5 OR representative-population differential signal → high-severity event
Category assignment:
- 0 high-severity + 0 representative-population differential → A
- 1-2 high-severity, no representative-population differential signal → B
- ≥ 3 high-severity or any representative-population differential signal fired → C

Why this sample lands on C

10 high-severity drift events across the audit period + repeated P3 KL-divergence excursions (peak 1.126 on Day 51). The high-severity count alone (10, well above the 3-event floor) is sufficient for C. The determination is not marginal.

Why the A/B/C structure matters

Continuous-scale scores (0-100 vendor dashboards, red/yellow/green tri-lights, "readiness percentages") push sponsors toward "just above the threshold" gaming behavior. A/B/C determinations force a single stance question:

Would you sign the attestation as-is today for FDA Type C, EMA scientific advice, ICH E6(R3) essential-documents review, DSMB pre-read, or IRB continuing-review? A/B/C answers the question regulators, sponsors' own regulatory affairs leads, and partners actually ask — instead of the question the vendor's dashboard prefers.

How does this help me?

Knowing your category BEFORE the Type C meeting / BIMO letter / EMA scientific-advice packet lands is the whole value of the $499 discipline.

Read: The value of knowing your category first →

$499 Snapshot. 3 business days.

Get your sponsor's Category A / B / C determination on your actual trial-AI surface — before someone else's question forces the answer.

Buy $499

Back to pharma sample hub.