Appearance
What this is
How Performix is built and how a diagnostic flows — from precomputed psychometrics to an executive decision surface.
Shape
A Next.js (App Router) app on Vercel (auth-protected via Clerk), with Drizzle/Postgres, Stripe billing, and an MCP surface (@modelcontextprotocol/sdk). It is not a standalone tool — it composes capabilities owned by other repos. API routes live at src/app/api/<namespace>/<verb>/route.ts.
Data flow (a diagnostic)
POST /diagnose
→ reincarnation item selection (toolbox spoke) — protected-feedback items, not a survey screen
→ CAMS scoring (IRT / MaxDiff / Monte Carlo EVPI–EVSI / Wilson intervals) — measurement, precomputed
→ binding constraint identified
→ executive surface: readout · interpretability · conformal gating (consumes Principia methods)AI assists around this (schema mapping, evidence extraction, ETL) — it is never in the scoring path.
Vendored contracts (consumer, not author)
| Consumed capability | Owner | Use |
|---|---|---|
cams-diagnostic items + scoring | toolbox reincarnation | the live diagnostic (real in prod) |
| competency-rollup + perception-gap | Principia methods | the Executive Readout |
| universal job schema / O*NET | toolbox job_family_agent | job-spec context |
Algorithms never live here — only vendored contracts; the toolbox is the canonical author of cross-cutting contract changes (see ~/devplane/docs/MULTI-REPO-WORKFLOW.md).
The interaction surface
Executive-facing primitives over precomputed measurement: Executive Readout, Interpretability Rules, Conformal-Interval Confidence Gating, and the Coaching Door (coaching-sim + coaching-nudge). Dense operator/admin views follow the P233 Operator-Console visual discipline; consumer diagnostic views use the light-surface default.
Honest scope
The CAMS diagnostic is real and live-validated; other surfaces vary from shipped to scaffold — trust the AGENTS.md wiring table. Performix's correctness depends on the upstream toolbox/Principia services it consumes.