Twinned from https://github.com/cfsh/ninefold/issues/1094 by tools/beads/import_github.py's reconcile pass.
games/explore/probes/explore-pause.json fails reproducibly on nf-dev-sg:
``` [probe] expect_prop PlayerCell '-40,0' FAILED — it is '-10,0' ```
The failing step is a 6-second { "hold": "explore_left", "ms": 6000 } right
after resuming from the pause menu (having visited the Controls page and
changed PauseControlsPercent a few times). The player ends up 30 cells
short of the expected position.
Bisected via a throwaway git worktree while merging #1090/#1092 (which
surfaced this as a UI-gate "new failure" against a stale 13-commits-behind
baseline — it isn't new):
- Fails identically at 9500af5 (before #1090, #1089, #1092 all merged).
- Fails identically at f12ecda (right after #1090, before #1089 changed
this same probe's back-navigation step to joybtn: b).
- Fails identically at current main.
Same exact numbers every run (-40,0 expected → -10,0 actual), not
randomly-varying jitter, which argues against a simple one-off host-load
frame hitch — though a *systematic* slowdown on this box's headless
llvmpipe rendering during a long wall-clock hold is still consistent with
a consistent shortfall. Reproduce: ./probe.py games/explore/probes/explore-pause.json.
Not filed against any specific PR — the bisection above rules out #1090,
#1089, and everything each depends on back to at least 9500af5. Whether
the root cause is a genuine gameplay bug (movement speed regressed further
back in history) or a probe/environment issue (the hold's wall-clock
duration not producing the expected number of processed frames under this
box's rendering load) is undetermined — worth investigating either way
since it's failing 100% reproducibly, not flaking.
None.
No comments.