From “something's off” to proof it's fixed.

Clinicians decide what right looks like. Precision does the measuring.

  1. Clinicians define what right meansTheir reactions become the rubric

    Clinicians mark real outputs pass or fail and say why. Precision turns that into written criteria.

    Read the evidence
  2. Scores with error barsEvery grader is checked first

    We check each automated grader with your clinicians, then show every score with its range.

    Read the evidence
  3. Re-run any visit or callSame encounter, one change

    Record a visit once, then re-run it as often as you need. Simulated patients fill the gaps.

    Read the evidence
  4. Find the part that brokeSwap one thing, measure the change

    Swap the speech model, prompt, retriever or model on the same visit. Causes come ranked by effect size.

    Read the evidence
  5. Fixes that hold, with a paper trailGates, approvals, ledger

    A fix is accepted only if it holds on new cases. Exact rules become hard checks. Every step is logged.

    Read the evidence
  • Failure types in clinical words
  • Rubrics clinicians can edit
  • A blind set to check against
  • How often each grader is wrong
  • Ranges on every comparison
  • Repeat runs, not one lucky run
  • Reproducible failures
  • Simulators scored for realism
  • No live patient contact
  • Ranked causes
  • Speech vs model vs prompt
  • Evidence an engineer can act on
  • Fixes proven on held-out cases
  • Hard gates for doses and escalation
  • A ledger reviewers can read

Panels show illustrative data.

The evidence

How strong is each layer in healthcare?

We say where it's proven and where it isn't.

Layer 1

Specify

Clinicians mark real outputs pass or fail and say why. Precision turns that into written criteria.

Evidence: Template exists

HealthBench shows the template: 262 physicians wrote its criteria. Clinician time is the scarce part, so we pick which cases to label.

Layer 2

Measure

We check each automated grader with your clinicians, then show every score with its range.

Evidence: Best evidence

Generic AI graders catch invented content but miss omissions: near chance on omissions (0.50 to 0.63) in published tests.

Layer 3

Replay

Record a visit once, then re-run it as often as you need. Simulated patients fill the gaps.

Evidence: Partial

Text and FHIR replay have precedents. Voice, accents and Indian languages do not, so simulators must be fitted to your own consented data.

Layer 4

Diagnose

Swap the speech model, prompt, retriever or model on the same visit. Causes come ranked by effect size.

Evidence: The opening

No healthcare study attributes scribe errors this way. Reading logs finds the faulty step only 14% to 52% of the time.

Layer 5

Repair and prove

A fix is accepted only if it holds on new cases. Exact rules become hard checks. Every step is logged.

Evidence: Design, not yet proven

Nothing here replaces a clinical decision. Precision proposes and blocks under your rules; a named person approves every change.

Ground rules

Two rules that never bend.

Check before you score

Every grader is tested with your clinicians first, and shows its own error rate.

A person approves every change

Precision measures, proposes and blocks. It never makes a clinical decision.

Meet the 18 agents See what we measure

Questions

The loop, explained.

More? hello@precision.example

What is the five-layer closed loop?

Clinicians react to real outputs, graders are checked against them, visits are replayed, causes are found by swapping one part at a time, and fixes are accepted only if they hold on new cases. Every step is logged.

Why swap parts instead of reading logs?

Methods that read a failed run and name the faulty step are right only 14% to 52% of the time. Swapping a part on the same input is an experiment, so the answer comes with an effect size.

Who decides what counts as correct?

Your clinicians. Precision proposes criteria from their reactions, shows where its graders disagree with them, and reports the graders' error rates.

Which layers are proven today?

Measurement has the best evidence. Replay works for text and FHIR but not yet voice or Indian languages. Diagnosis by swapping parts has no healthcare study yet. Repair is a sound design with little independent evidence.

Know your healthcare AI is safe to trust
before it reaches a patient.

Book a 30-minute call. We'll look at your product and tell you what we would measure first.