Build the rule.
Test the boundary.

Download and extract Courier Lab. The archive contains the reviewed teaching code, source exercises and separately labeled author controls. It contains no learner submissions or private assessment bank.

Download Courier Lab beta (.zip)

Install Node.js 22 or later, open a terminal in the extracted courier-lab folder, and run:

npm start

Open the full /learn URL printed in the terminal. Keep the terminal running. No package installation is needed to study the course or run the source exercises. Your records stay in that local workspace; export them before moving or clearing it.

Want to try a small example first? Build a movement validator in your browser. This optional model teaches the authority and latency tradeoff; it has a different protocol from the full course.

1. Authority and distance. Repair a proximity guard without trusting client position.
2. Input ordering. Preserve valid sequences, expiry and pending actions.
3. Independent tests. Observe authoritative effects through real HTTP requests.
4. Owner time. Account for processed ticks, pauses and reconnects.
5. Owned-file integrity. Compare an inventory with an independently approved reference.
6. Causal replay. Reconstruct a bounded trace from its checkpoint.
7. Detector evaluation. Measure case-level errors and keep evaluation separate from tuning.
8. Retention and privacy. Keep only evidence needed for a stated question.
9. Compatible recovery. Restore a coherent version and account for lost state.
10. A new mechanic. Implement and test an atomic repair transaction with safe retries.

This is an original owned toy-game teaching system, not production anti-cheat security or a job qualification. It does not target third-party games or install drivers. A denied action is not evidence of a person's intent.

Technical and AI learner-surface reviews support the tested behavior. They do not establish human learning, retention or accessibility for every user. Allow 12–23 active hours as an unvalidated planning range, with source work likely spread across sessions. Exact historical generator model IDs remain unavailable; the provenance record preserves that gap and the reviewed source identity.