How the Clinical Pulse is computed
Radical transparency about how the score is calculated is what separates trustworthy ratings from anonymous gossip. Here is the formula, the screening rules, and the public endpoints — including a JSON-LD schema so procurement agents can query the data directly.
01The composite formula
Every review contributes a triplet of scores: Workflow, Reliability, Impact. The composite is a weighted mean, then attenuated by a freshness factor and a verification factor.
Reliability carries the highest weight because clinicians repeatedly tell us the cost of a false positive in real practice exceeds any workflow gain. Workflow and impact are equal — different clinicians weight them differently and the symmetry is honest.
02Time-decay weighting
A review of a model that has been retrained three times since submission is misleading. Weight decays stepwise — recent reviews count fully, older ones less.
| Review age | Weight | Effect on composite |
|---|---|---|
| 0–12 months | 1.00 | Full weight |
| 12–24 months | 0.50 | Half-weighted |
| 24–36 months | 0.25 | Heavily attenuated |
| 36+ months | 0.10 | Display only, near-zero weight |
03Verification weighting
Every reviewer is verified against a regulator-issued register — GMC in the UK, NPI in the US, ÄkV in Germany, equivalent bodies elsewhere. Anonymous reviews do not exist on this platform.
| Reviewer status | Weight |
|---|---|
| Verified clinician with > 500 cases on the device | 1.20 |
| Verified clinician (default) | 1.00 |
| Unverified | 0.00 — not accepted |
03bReviewer compensation transparency
Pro tier access for active reviewers is independent of review content. Verified reviewers receive Pro on their first published review and retain it while publishing three or more reviews per trailing 90 days. The perk is for being verified and active — not for the substance of any review, and never tied to a particular score, sentiment, or device.
04Screening rules
Every submission enters a 48-hour cooling-off window. During this period the review is screened for:
- Same-IP / same-institution clusters within 48 hours of each other
- Manufacturer email-domain matches (current and past employees)
- Sentiment outliers versus the device's published evidence grade — flagged for human review, not auto-rejected
- Duplicate or near-duplicate text (cosine similarity > 0.85) across the reviewer's own history
- Modified Z-score outliers on the dimension scores (threshold 3.5)
Flagged reviews are reviewed by a duty editor within 24 hours. They are either published with a note, returned to the reviewer for clarification, or rejected with a documented reason. The reviewer always sees the reason.
05Minimum sample size
A device requires 30 verified reviews before a Clinical Pulse score is publicly displayed. Below that threshold the page shows the individual reviews but no composite — a small sample is more misleading than no sample at all.
06Distribution, not just the mean
Every device page shows the full score distribution as a histogram. A composite of 78 with reviews clustered between 75–85 is very different from a 78 with a bimodal 40/95 split. Clinicians read this fluently from clinical-trial data — we trust them with the same view here.
07For agents and procurement systems
Every device page exposes a JSON-LD RatingMethodology graph and a public REST API. Procurement agents and clinical decision-support tools can query the live data without scraping.
Public endpoints
JSON-LD schema (excerpt)
{ "@context": "https://app.medtekki.no/schema/v1", "@type": "RatingMethodology", "name": "The Clinical Pulse", "version": "1.0", "scale": { "min": 0, "max": 100 }, "dimensions": [ { "id": "workflow", "weight": 0.30 }, { "id": "reliability", "weight": 0.40 }, { "id": "impact", "weight": 0.30 } ], "freshnessDecay": { "type": "stepwise", "buckets": […] }, "minimumSampleSize": 30, "coolingOffPeriodHours": 48 }
The full schema is embedded in the <head> of this page as application/ld+json — view source to see it. We publish the schema openly under CC BY 4.0; mirror it, extend it, or feed it into your own tools.
07bTiers
The catalogue ranks devices in three tiers. The point is editorial transparency: some devices we have studied closely; some we have summarised; and every FDA-cleared AI medical device is listed by virtue of being cleared.
- Hero. Devices we have reviewed in depth. Six entries today; the Auris+ treatment includes the Inside the Algorithm walkthrough, the Decision Ledger surface, and an editorial verdict.
- Featured. Devices we have curated. Filled-in clinical notes, study citations, regulatory history. Selected on the basis of (a) at least one peer-reviewed publication, (b) multi-year FDA regulatory history, or (c) at least one published Clinical Pulse review with high agreement.
- Lean.Every FDA-cleared AI medical device, sourced from the FDA's published AI-Enabled Medical Devices list. No editorial commentary; the FDA's submission detail is the ground truth.
Lean entries can be promoted to Featured (and Featured to Hero) at any time. Promotion is editorial; reviewers vote with their reading by leaving a Clinical Pulse review.
Featured tier opens with ten hand-curated entries spanning seven specialties. Expanded coverage arrives as the Phase 3 enrichment pipeline ships in mid-2026, with the orchestrator drafting per-device YAMLs for editorial review.
08What this methodology does not do
The Clinical Pulse is one signal. It does not replace the Auris Mark, which is editorial and evidence-based — see the full AurisMark Rubric (v1.0) for the criteria, principles, and worked examples. It does not replace your institution's procurement diligence. It does not capture rare-event safety signals — those live in the Safety Alerts section of each device page and are sourced from regulatory post-market surveillance, not user reviews.
If you find a flaw in the methodology — statistical, ethical, or practical — we want to hear it. Email methodology@medtekki.no. Significant changes are versioned and logged on the methodology changelog.
Methodology v1.0 — published 28 April 2026 — MEDTEK KI AS · Stavanger.