The runner measured elapsed_s with time.time(), i.e. CLOCK_REALTIME, which
keeps advancing while a machine is suspended. The box running bk9 was suspended
for three days, so any run spanning that would have reported elapsed_s inflated
by ~72 h -- silently, since the fail count and score are unaffected and nothing
else would look wrong.
time.monotonic() is CLOCK_MONOTONIC on Linux, which stops during suspend
(CLOCK_BOOTTIME is the variant that does not). That measures the time a run
actually had a CPU, which is what the column is for.
Nothing is corrupted yet -- no run has completed since the suspend, so the
timing column is still only the 39.12 baseline, measured on a box that stayed
awake. Noted in the module docstring so that caveat travels with the number.
Refs homemaker-py-bk9.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MJ84Feep79Hhm3E4zZJmnB
The runner never passed --checkpoint-every, so each run's only output landed at
the end. The longest single run in the first baseline took 62 h; losing the
machine at hour 61 lost all of it. That flag was added during this session
precisely for this case and then not wired into the one script that needs it.
Default is budget/20 -- 25000 evals at the 500k baseline budget, roughly every
3 h for the slowest programme. --checkpoint-every overrides it.
Also documented --slots: it is the number of concurrent runs, each single-worker
(one worker per run avoids b8g's parallel non-determinism), so it should match
the core count.
Verified end to end with a 400-eval run: the .dom.checkpoint appears alongside
the output, and the score/record/push path still works. The smoke-test artefacts
were removed rather than committed.
Refs homemaker-py-bk9.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MJ84Feep79Hhm3E4zZJmnB
record_and_push used `git add -A`, so each completion committed the entire
working tree. On the live local run that meant one commit carrying 49 files
and ~1.97M insertions -- three still-running programmes' partial
coldstart-*.dom, plus unrelated evolved-*.dom -- under a message naming
only programme-house seed 0. In-flight artefacts were being recorded as if
they were results, attributed to the wrong run.
Now stages and commits exactly this run's .dom/.log/.score/.fails plus the
results TSV, via `git commit --only <paths>` so it holds regardless of what
else is staged and a concurrent edit elsewhere cannot ride along.
Note for the run currently in progress: it has the old code loaded, so it
will keep sweeping until restarted. The committed .dom files for
harbor-house, maple-court and health-centre are mid-run snapshots, not
results -- their logs show 75k/87k of 500k evals -- and should be
disregarded until those runs report their own rows.
First real result is in: programme-house seed 0 @ 500k, finish collapse
28 -> 1 fails, final 1 fail (0 hard / 1 soft), 19400s.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MJ84Feep79Hhm3E4zZJmnB
Cold-start re-baseline after the DESIGN.md 38.10/38.11 objective
change. Single worker (avoids homemaker-py-b8g), scored by the
shipped scorer from the programme directory.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MJ84Feep79Hhm3E4zZJmnB
