These are original, synthetic teaching fixtures prepared and tested with AI assistance. They are not production-game footage or human acceptance approvals.
Both pin nixie-fx 0.1.20 and Three.js 0.185.1. PixiJS visual parity and device/GPU benchmarks are outside these checks. Dependency licenses are included in the archives.
Open the interactive fixtures · Download source, recording server and tests · Read test results · Browser environment and action traces
Run node check.mjs from the source folder. The deliberately faulty models fail three regression assertions while their three neighboring checks pass. All six checks pass for the corrected models. Recordings are continuous captures of the fixture canvas, with scripted inputs shown in the frame. Backend networking and touch-device behavior are not simulated as real production systems.
Fault: reopening presents an available control despite persistent claimed state and sends a second request. Repair: derive reopened view state from the persistent claim. Failed requests still allow retry.
Fault: receiving attack-1 again after reconnect adds a second history row. Gameplay application was already deduplicated. Repair: guard the history insertion by the confirmed action ID; rejected actions do not block a later valid action.
Fault: cancelling preserves the authoritative item but leaves its view hidden until reopening. Repair: restore visibility on an invalid drop; a valid move still changes the slot once and preserves identity and quantity.
Download the script, sample, tests, evidence and reviewer checklist. The qualified human review remains pending. Passing automated checks is not a human sign-off.
Prepared October 19, 2026. Return to Patil's blog.