You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Daemon witnesses: make the warm check reach its hold deterministically #160
Follow-up from #131's merge gate (Limit L3). The held warm-up, re-warm and "works past the bound" witnesses need the warm daemon's check to reach its hold within the CLI's 3 s test bound. That fails loudly, but it is a timing dependence (R57; C-30, now M). Deterministic form: give the CLI's bound a driver-written release file, next to the existing WAIT_MARK, so the driver releases the bound only after the hold is reached.
Seen in CI: #133 at 5fc4b9a, run 44436324847, job Gap report end-to-end oracle (determinism, 4/8). The unit determinism: constructed history: a check through the daemon that works past the bound recorded different traces at observation 8 (incremental, step test). In the first of its two runs only, the CLI printed oracle-gaps: note the package's daemon sent nothing for 3 s; this check runs cold, which is the timing dependence that this issue describes. Main 30928ee's determinism shard that holds this unit passed in its run (44435834785), and #133's previous head 9bb7c74 passed it too.
#133 adds synchronous work before a warm check's first beat (hashing the session's warm-time reads for inputsChanged and the unit identity). On the e2e fixtures that is +15 to +35 ms for a whole unchanged daemon check, against the 3 s bound, so it can add only a little to this rate. The deterministic release that this issue proposes would close it.
Activity
KyleAMathews commentedon Nov 19, 2026
Seen in CI: #133 at 5fc4b9a, run 44436324847, job
Gap report end-to-end oracle (determinism, 4/8). The unitdeterminism: constructed history: a check through the daemon that works past the boundrecorded different traces at observation 8 (incremental, steptest). In the first of its two runs only, the CLI printedoracle-gaps: note the package's daemon sent nothing for 3 s; this check runs cold, which is the timing dependence that this issue describes. Main 30928ee's determinism shard that holds this unit passed in its run (44435834785), and #133's previous head 9bb7c74 passed it too.#133 adds synchronous work before a warm check's first beat (hashing the session's warm-time reads for
inputsChangedand the unit identity). On the e2e fixtures that is +15 to +35 ms for a whole unchanged daemon check, against the 3 s bound, so it can add only a little to this rate. The deterministic release that this issue proposes would close it.This comment was written by Isaac.