Popper Cookbook

• What it is: companion package with runnable examples (bun run examples/NN-*.ts) for testing and demonstrating the API of the supervision engine Popper (#50). Client wrapper lib/popper-client (supervise/printResponse), types from @regain/hermes.
• Type / status / role: library (examples/cookbook) · active · lead — 10 of 13 user commits (aniashev 3).
• Active period: 2026-01-26 → 2026-01-28 — 3-day sprint.

• Languages: TypeScript.
• Frameworks/libraries: Bun (runner), @regain/hermes (supervision/HTV contracts), client for Popper API. .cursor/ (AI tooling Cursor).
• Infra/deploy: none — locally runnable scripts.
• Data: calls Popper API.

10 author commits; 20 examples, ticket-oriented (SAL-*/POP-*):

• Basic supervision flow: health (01), simple message (02), wellness vs clinical modes (03), stale snapshot (04), medication proposal (05).
• API features (SAL-604..614): full supervision test (06), audit events (07), idempotency (08), safe-mode (09), API keys (10), organizations (11), rate limiting (12), operational settings (13).
• Policy/governance (POP-013b..016): policy lifecycle (14), drift counters (15), drift baselines (16), drift triggers (17).
• RLHF and export (POP-015b, 023a/b): RLHF bundles (18), export bundles (19), TEFCA/USCDI interoperability (20).
• Volume: 13 commits in 3 days; coverage of the entire Popper API.

By the set of examples (demonstrating understanding of a complex governance API; user's authorship):

• Drift detection (examples 15–17: counters → baselines → triggers): monitoring drift of model/policies with baselines and triggers. → MLOps maturity (tracking AI degradation).
• RLHF bundles (18): collecting feedback data for fine-tuning. → ML feedback loop.
• Healthcare interoperability (20: TEFCA/USCDI): export in US health data exchange standards (TEFCA — Trusted Exchange Framework, USCDI — US Core Data for Interoperability). → Knowledge of healthcare regulatory standards.
• Idempotency / safe-mode / rate-limiting / operational settings (08, 09, 12, 13): production-grade reliability of supervision API. → Understanding resilient APIs.
• These are examples (not the engine), but show full coverage and understanding of the governance/compliance domain.

• Sessions found: 0 in the corresponding directory (verified by normalization). There's .cursor/ → development was via Cursor AI; no Claude Code sessions in this directory.
• What was done with AI: writing examples per tickets, likely with Cursor.
• AI workflow patterns: cookbook-driven development (examples as living API tests), ticket-oriented.

• 20 examples covering the supervision API end-to-end (supervision, audit, idempotency, safe-mode, policy lifecycle, drift, RLHF, export, TEFCA/USCDI).