Resolve X4 source lag under active view rendering and native/Flatpak performance gap

## Current release qualification, 2026-09-12

This is now a shared X4 active-playback problem, not only a native/Flatpak gap.
At 2256x1504 under 40 seconds of high-rate pan, published GitHub0.3.0 measures
250–255 completed redraws/sec but only26.7–27.2 source advances/sec from29.97fps
footage. Restoring the exact accepted48e package reproduces254.899/27.225.
The independently built signed0.3.0 package measures253.949/27.600, with3.14s
reported lag and completion p99/max15.198/25.173ms. None is a combined
full-source/240-capacity pass. ONE X2 signed0.3.0 measures314.724/29.975 with
p99/max10.886/24.094ms. Source lag and frame-time spikes are not accepted
tradeoffs. Earlier successes below remain evidence, not a current guarantee.

An existing-lifecycle-probe run localizes the sustained deficit to the serial
filtered stitch/graphics stage, not delayed UI events: redraw age p99/max is
0.056/0.259ms and pump duration p99/max0.025/0.090ms. Worker starts are
consecutive and average35.32ms apart, matching about28.3 source advances/sec.
The composite start-to-panorama span grows from roughly10–12ms before pan to
32.68ms mean during pan; temporal filter-submit-complete grows from11.23ms
idle median to15.17ms pan median. Existing timers cannot distinguish worker
CPU/import work from shared GPU/driver contention. No speculative fix follows
from that remaining ambiguity. Logging and host conditions also differ, so
233.199 redraws/sec in the lifecycle run is not directly an overhead estimate.

Current evidence: `scratch/installed-capacity/release-030-*` and
`scratch/merge-readiness-20260912/release-030-x4-lifecycle-classifier.md`.
Release qualification remains open in #187; the public notes disclose the
limitation. Preserve the accepted picture, source cadence and history while
addressing performance. No broader Studio optimization RE is needed.

## Historical measured concern at issue opening

The owner-accepted GPU stitching release path is the Flatpak. The native build
is slower under uncapped active playback on the same Radeon 760M machine.
Keep this discrepancy visible after the shared-stitcher delivery closes #184;
at that checkpoint the installed cohorts below met the average capacity target.
The later release qualification above supersedes that unconditional status.

Matched 40-second X4 Air interval, 2256x1504, source29.970fps, on battery:

- Native source48e1d741:217.949 completed redraws/s,28.575 source advances/s;
completion-spacing p99/max15.409/24.189ms, maximum reported lateness1.834s.
- Unchanged preceding installed baseline at the same interval:267.224 redraws/s,
normal29.975 source advances/s.
- Installed source48e1d741:280.499 redraws/s,29.975 consecutive source
advances/s, p99/max13.813/23.463ms; no reported drops/starvation/audio underruns.
- Installed ONE X2:318.024 redraws/s,29.975 source advances/s,
p99/max9.729/23.788ms; no reported drops/starvation/audio underruns.

This rules out battery state alone, not native compiler, loader or runtime/driver
differences. Do not claim a cause from those numbers. Capacity is not physical
scanout or a guarantee that every redraw fits4.17ms; the panel is60Hz.

## Follow-up boundary

Use the existing authenticated actual-player harness and saved interval. First
compare source, executable/linker provenance, FFmpeg and GPU runtime identities;
then isolate the smallest differing boundary with matched controls. Preserve
source cadence, maps/color, world-coordinate temporal filtering and the owner-
accepted picture. Do not reopen broad Studio optimization RE or silently select
slower seam refresh, stale maps or a new color smoother.

Frame-time tails remain part of every report, not erased by the throughput
average. No new algorithm change or accepted quality/performance tradeoff is
authorized by opening this tracking item.

Evidence stays local under `scratch/installed-capacity/world1147-x4-native-01`,
`world1147-installed-baseline-01`, `world1147-installed-candidate-01` and
`world-x2-installed-candidate-01`. Public summary: `docs/MERGE_READINESS.md`.
Related #182, #184; camera/GPU coverage remains #88.

Start with `docs/MERGE_READINESS.md` and the evidence artifacts in `scratch/merge-readiness-20260912/release-030-x4-lifecycle-classifier.md`, `scratch/installed-capacity/world1147-x4-native-01`, `world1147-installed-baseline-01`, `world1147-installed-candidate-01`, and `world-x2-installed-candidate-01`. Then run the existing authenticated actual-player harness with the saved 40s/2256x1504/29.97fps interval to reproduce native vs Flatpak X4 behavior. Compare source advances/sec vs redraw/sec and p99/max completion/lateness in the same view, and finish by confirming evidence shows restored source cadence without introducing accepted quality/performance tradeoffs.

由索引模型根據 Issue 內容生成。