When something on JetPatient isn’t real production data, we say so. When the AI is sampling, you see a sample badge. When a clinic’s outcome cohort hasn’t reached the publish threshold, we show pending — not fake numbers. This page is the single source of truth for what every badge means, exactly when you see it, and what we’re doing about it.
Every JetPatient AI surface (Brief, Risk Engine, Scribe, Pathway, Navigator, Copilot, PREDICT, JetVision) renders one of three badges on every response.
The response on the page was generated by the live Claude model against the actual prompt you submitted. No caching, no substitution. This is the “normal” state for every AI surface once the production Anthropic API key is configured.
Someone (often you, a moment ago) sent an identical input to the same model within the last few minutes. To save you a wait and the model a call, the cached response is returned. Content is byte-for-byte the same as if the model had just computed it.
The live AI is unavailable (no API key configured, upstream error, or you’re in a preview environment), so a curated sample response is shown. Samples are crafted to be representative of what the live model would produce for a generic case, but they are not specific to your data and should not be used to make decisions. Every sample badge is honest: we’d rather show a labelled example than silently fail.
When a clinic doesn’t have enough completed JetPatient cases yet, we will not substitute demo data into a real profile. We say pending and explain when the cohort will publish. Patients comparing their own recovery against the cohort see the same honest rules in Patient Passport » Cohort Mirror — never a misleading trend on too little data.
JetPatient publishes recovery curves and outcome cohorts only when the underlying case count clears MIN_PUBLISH_COHORT = 10. Below that, the panel says “we’ll publish once 10 cases are on file” instead of showing demo numbers behind a small SAMPLE pill (the pre-Block-39 behaviour that we reversed because it was a credibility risk). Block 42 added the patient-PRO collection loop that feeds this threshold; Block 43 added the patient-side Cohort Mirror that uses the same data + same threshold — anonymized, with the patient’s own submissions overlaid.
For clinics: JCI status, surgeon licences, malpractice records, and sanctions checks were validated against the issuing authority. The supporting documents are stored in the audit log. A clinic only carries the Verified badge while credentials remain in good standing.
Information has been entered into the system (by a clinic during onboarding, by a patient during intake) but the verification workflow has not yet completed. Treat as the submitter’s claim. The audit trail records who entered the data and when.
Quote-checkout, BNPL prequalification, and payment flows use a consistent vocabulary so a patient never has to guess what stage they’re in.
Created but not sent. Edit freely; nothing here is visible to the other side of the conversation until you explicitly hit Send.
Delivered to the recipient. They’ll see it the next time they open JetPatient or check their notifications.
Both sides have signed off. The next step in the flow is unblocked (payment hold, case creation, deposit capture, etc.).
Quotes and prequalification offers have explicit validity windows. Once past, you’ll need a fresh version — one click to regenerate, no data lost.
The system is healthy; the specific action can be retried. Persistent failures are routed to JetPatient ops via the same telemetry that feeds the status page.
Across Blocks 13, 35, and 39, the team established the same pattern in three different surfaces: when the live system can’t produce a real answer, return a clearly labelled sample, never a silent failure. Block 39 went further and reversed the pre-existing pattern of substituting demo cohort data into real clinic profiles with a small SAMPLE pill — on real clinic profiles, the only acceptable rendering is real data or an honest empty state.
The same state that drives the badge a visitor sees is recorded in the audit log (NS.AUDIT) and the AI usage telemetry (NS.AI_USAGE). The admin SLA dashboard at /admin/ops.html shows the live-vs-sample rate per endpoint, so the team knows when sample-mode is happening more than it should.
You found this page by clicking the tooltip on a badge somewhere on the platform. The tooltip is automatic for every badge on every page that includes assets/js/jp-trust-badge.js. The labels in this explainer match the labels in the code — if we add a new state, this page is updated in the same PR.
The publish thresholds (MIN_PUBLISH_COHORT = 10 for clinic outcomes, the Trust Score “methods note” rules) are chosen to balance signal against premature publication. They will tighten over time as the network grows and the law of large numbers does more of the work. Methodology notes for each threshold are linked from the Trust & Methods hub.
If you see numbers or AI output on JetPatient that don’t carry a badge but should, please email trust@jetpatient.com. We treat missing-badge reports as P1 because they break the contract on this page.