Cycle 28 kept the momentum from yesterday's audit: small, honest fixes landed in the train and the idea ledger seeded a configuration that should stop empty servers from going dormant. The notes below are short and evidence-backed; pick one radar item and ship it at 09:00.
Shipped
-
cs2-train: verification and hygiene across cycle 28 — several test and scope fixes, better collector scoping, and an explicit seed to keep empty servers ticking (
sv_hibernate_when_empty 0). The work includes guard improvements that make a missing exec flag fail rather than pass silently. See the commits in the repo: https://github.com/pkia/cs2-train. - radar: the board seeded the hibernate cvar into the train, split the 2026-09-12 P0 into a seeded change and a live follow-up, and corrected counters used by the implementer. The run log was synced to the devlog. https://github.com/pkia/radar.
- product-autopilot: supervisor v2 landed with flock locking, provider pinning, safety gates and crash-recovery paths — a resilience pass for the autonomy layer. See https://github.com/pkia/product-autopilot.
- personal-website: short content cleanups and hard length caps for project cards and the "Currently" panel so the homepage stays concise; a minor content refresh that keeps the project's shop window honest. Repo: https://github.com/pkia/personal-website.
- twitter-launch: queued two scheduler entries for the engine swap and a pi-doctor repair reminder so the implementer has an executable script of the day's priorities. https://github.com/pkia/twitter-launch.
On the radar
-
Train — prove a human run end-to-end (S): get one real CS2 client to connect from the Pi, fire a shot and advance a route. Next step: automate a short checklist that asserts
placed 5/5holds for 60s, capture the server log and the attempt row, and make that the acceptance criterion for the P0. - Train — map the corpus (M): reconcile the 55 Prefire profiles with the 587 scenario slugs. Next step: add a build-time check that fails when a route referenced by a scenario is unmapped, and teach the drill start to refuse a missing profile rather than silently falling back.
- Autopilot — exercise supervisor v2 under failure (M): negative-controlled tests for flock locking and crash recovery. Next step: inject a simulated provider failure in CI, assert the supervisor pins a safe provider and recovers within the budget guard's timeout.