Independent certification simulation

Certification intelligence across the full lifecycle.

A connected workspace follows evidence from job analysis through release, monitoring, investigation and refresh. It demonstrates how review gates can control what moves forward.

SIMULATED DATA SIMULATED REVIEW

Certification Intelligence workspace showing ten lifecycle stages, ten thousand synthetic attempts and clearly labeled simulated review

Every record and every review decision on this page is simulated. Practitioner, panelist, candidate, response, reviewer, approval and disposition records were pre-generated for this demonstration. They are not human approvals and do not validate a real credential.

10Lifecycle stages

10,000Historical synthetic attempts

600,000Historical simulated responses

500Separate successor pilot attempts

Design → Validate → Build → Set Standard → Pilot → Analyze → Launch → Monitor → Investigate → Refresh

Follow evidence through each gate.

Define the work

Synthetic practitioner evidence supports proposed tasks and domain weights. A simulated reviewer disposition controls whether the JTA can feed the blueprint.

Build and set the standard

The approved simulated JTA weights constrain form assembly. Twelve simulated judges complete three Modified Angoff rounds before a separate simulated approval records the standard.

Launch and monitor

Historical data covers 10,000 synthetic attempts and 600,000 simulated item responses. Monitoring can open an investigation without changing prior scores or evidence.

Refresh and revalidate

An approved simulated revision produces a successor item. That item receives its own 500-attempt synthetic pilot and exact revalidation before a refreshed release can advance.

Walkthrough

Certification Intelligence in action.

This walkthrough shows an independent demonstration using simulated data and simulated reviewer decisions.

Behind the build

Agent orchestration and engineering

I designed the architecture and coordinated scoped engineering tasks through Codex and Claude Code, with defined responsibilities, review checkpoints and automated acceptance checks. The platform combines model-assisted interpretation with deterministic simulation and statistical analysis, generating thousands of candidate attempts without thousands of model calls. Versioned artifacts, reproducible runs and explicit decision gates keep the workflow auditable.

The walkthrough combines my DemoForge workflow for screen capture, timed highlights and captions with my Voicebox narration runner for resumable local narration using Voicebox.

Simulation and analysis: GPT-5.6 Luna through Codex produced a recorded investigation synthesis from analysis evidence. Seeded statistical simulation generated the 10,000-candidate population; deterministic calculations produced the metrics. The public demo replays stored model outputs and simulated reviewer decisions.
Walkthrough production: Qwen3-TTS 1.7B generated narration locally through Voicebox. Playwright captured the working application, FFmpeg assembled the video, and faster-whisper supplied speech-alignment checks.

Model roles: GPT-6 Astra for engineering and orchestration; GPT-5.6 Luna for bounded investigation synthesis; Claude Fable 5.1 for targeted code, UX and visual review. Playwright, FFmpeg and automated tests handled repeatable verification.

Engineering foundation: Node.js, TypeScript and Express; a Supabase/Postgres data model; an isolated Railway demo deployment; versioned evidence, repeatable seeds and automated tests.

Walkthrough and deployment

The interactive public deployment is ready.

The public Certification Intelligence app is available for interactive review alongside this walkthrough.

Open experiment