Skip to content
← BLUEHEALTHCARE
blueCatch

blueCatch

What a review process actually catches

Blinds seeded error cases into a real review queue, imports the reviewer's actual decisions, and signs a scorecard of per-class catch rates with confidence intervals — scoring the process, never a named reviewer.

Module — in active build
Request a pilot conversation Read the manual
Runs locally · CLI or web UI · Ed25519-signed receipts · no cloud account required
Who it's for

Quality and review-assurance leads.

What it is not

Not a live covert-seeding tool, not an employee-surveillance product, and it never touches medical, legal, or financial production records.

Screens

blueCatch measures a review process, not the people in it. Known error cases are declared and seeded, blinded, into an exported review queue; the reviewers' real decisions come back in; and the scorecard reports, per error class, what was caught — never blended into one number. The five screens below run that sequence in order: project, audit setup, blind export, scorecard, and the offline verifier.

blueCatch Projects screen: the suite shell's sidebar (Setup with Projects, Audit setup and Seed library — the latter two disabled with the note Open a project first — then Export, Import, Results) beside a Local projects table holding one project, blueCatch demo, with a New project button, an Offline badge in the top bar, and a footer reading Evidence first: blueCatch measures a review process, never a named reviewer, using synthetic seeded cases only.
Synthetic demo data
Everything begins with a project. Audit setup and the Seed library stay disabled until one is open. The footer records the posture on every screen: offline, synthetic seeds, and a process measured rather than a person.

Audit setup is where the review queue and the seed library are declared and hashed, before anything runs.

blueCatch Declare the inputs for this audit screen: a Review queue panel with a validated 100-item CSV, its content hash, and a paste box, and a Seed library panel below it with a validated 10-item YAML seed set and its own content hash, above a Replace and re-validate button.
Synthetic demo data
Audit setup — a 100-item review queue and a 10-item seed library, each validated locally and content-hashed before the audit can be sealed. Once every input is declared, the audit freezes and a DECLARE receipt is issued automatically, before any decision data can be accepted.

Blind queue export blends the seeds into the real queue and seals the mix.

blueCatch Freeze the inputs before evaluation screen: a sealed Run declaration panel listing the policy hash, input manifest hash, seed library hash and blind queue hash, a Composition panel reading 100 organic items and 10 seeded items with a note that seed identity is sealed until the result certificate, and a Blind queue preview table with no field marking which rows are seeded.
Synthetic demo data
Blind queue export — 100 organic items and 10 seeded items blended into one 110-row queue, hashed and sealed, with a Download blind queue CSV button. No column in the preview marks which rows are seeded; that mapping stays sealed until the scorecard, after decisions are imported.

Decision import brings the reviewers' real answers back in; Results scores them.

blueCatch Process scorecard screen: a synthetic-data banner, a headline reading This review process reliably catches 1 error class and reliably misses 1 error class, a Scored summary of 5 of 10 scoreable seeds caught, four count cards — 10 blinded seeds, 5 caught, 5 missed, leakage-rate warning NONE RECORDED — a catch-rate chart by error class showing POLICY_VIOLATION 5 of 5 marked CATCH-DETECTED and SUBTLE_MISROUTE 0 of 5 marked CATCH-BROKEN, each with a Wilson 95% confidence interval, and a Claim boundary box.
Synthetic demo data
The headline is a sentence, not a percentage. POLICY_VIOLATION 5 of 5, CATCH-DETECTED; SUBTLE_MISROUTE 0 of 5, CATCH-BROKEN — each with its own interval, never blended. A class whose interval straddled the bar would read CATCH-UNDERPOWERED instead of either. Leakage-rate warning none recorded: seeded items were a small enough share of the queue that a reviewer could not have inferred which were planted.

Download report writes the signed scorecard; Verifier checks it without trusting this UI.

blueCatch Verifier screen: a pasted signed receipt on the left showing a green Receipt verified check mark, terminal verdict SCORED, and signature, schema and pre-registration chain all reading valid, beside an Offline verification panel on the right with a paste box for any receipt JSON, a command-line example, and a Verify receipt button.
Synthetic demo data
Verifier — a SCORED receipt with its signature, schema, and pre-registration chain all valid, checked without trusting this web UI. The same JSON can be pasted into the panel on the right, or checked from the command line with no database and no server running.
Status, measured

CLI, API, and web UI exist with a 100-item synthetic demo queue. see the status page for exactly what's built.

A signed receipt proves what blueCatch ran and recorded; it does not prove an underlying real-world claim is true. blueCatch does not approve, deny, recommend, pay, retract, or alter anything a named human is responsible for deciding.