Two extended Wayland reps under increasingly aggressive
provocation conditions to find the stutter envelope referenced
in the campaign premise:
- a1prime_konsole_rep1: chromium-fourier + active konsole running
`top -d 0.5` behind, both at fullscreen-stacked geometry.
- a1prime_tiled_rep1: chromium-fourier at 600x400 tile +
konsole+top behind at fullscreen.
Both: drops_post_warmup=0, kwin %CPU median=0.00, perf zero
samples in composite/dmabuf/GL paths. Same floor result as
single-window A1.
KWin DBus scripting confirms the limitation: all top-level
windows are at fullscreen-equivalent geometry stacked, kwin's
z-order culls everything behind chrome from rendering. Even the
tiled rep didn't break the pattern.
Three interpretations: (A) predecessor's patches actually fixed
the original Wayland stutter on this workload (per
KWIN_PIVOT.md, plausible); (B) stutter conditions I haven't
matched (different content, time-of-session, load); (C) session-
state divergence. Cannot adjudicate from synthetic reps alone.
a1prime_findings.md surfaces four operator-decision options:
specific scenario from daily-driver, extended-use observation,
acceptance of (A) and matrix reframe, or continued probing.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
marfrit
298431810e