GNU Chess slim bilingual build: final native acceptance PASS

Final English: 169232 bytes, SHA256
58a4cc7c8a293913ce8069d6888cd1793aec8a05f2341a0b7276e20d27084941
Final German: 169248 bytes, SHA256
eb65acf09895f2684a10899ba104d4166217942e0d47a5144a1cfb0bf96719c4
Baseline English: 224622 bytes, SHA256
796ba70122e22276bbeac16ef55fa9bed87d525419072446df3afff6137d5c62

Matched native comparison: supplied English GeoWorks Pro 1.2, DOSBox
Staging 0.83.0, normal core 80000 cycles, VGA 640x480/16 colors,
16 MiB configured RAM and 1500 handles. Both variants used the same
read-only observer and native input workload. No guest memory was patched.

48 native searches passed: three repeats of each depths 1-4, on the start
position and automatic reply after e2e4, per variant. Every paired selected
move matched. BUSY timings cover thinking repaint and engine execution,
but end before final result repaint. They are host-emulator observations,
not hardware timing guarantees. The observer interval is nominal 2ms;
independent audit observed gaps up to 49.216ms. See min/median/max data.

Native Perf Fixed: baseline 203.6 -> 383.3KiB; slim 203.6 -> 329.0KiB.
Read-only native heap-counter deltas: 183984 -> 128384 bytes, a 55600-byte
(54.30KiB) reduction. App-owned allocation at initial idle: 271808 -> 217296
bytes. Sampled search residency peak: 271648 -> 217520 bytes. First idle
residency differs in paging state: baseline 7 code resources swapped,
slim 0; this is an observation, not a universal residency guarantee.
The whole-system Memory display was 663.0/837.3KiB before/after for
baseline and 740.1/845.7KiB for slim, reflecting variable system/cache
residency. It is not used to claim a stable total-system footprint saving.

Both 4MiB variants passed eight additional searches, native close and
ordinary icon reopen at default depth2. Fixed allocation and idle locks
remained correct. Slim4 immediate reopened owned allocation was 217888
bytes versus 217296 initially; dynamic UI allocation was 592 bytes higher.
This does not establish an invariant allocation for every lifecycle.

Independent final 2MiB pressure test passed with Chess, GeoWrite, GeoDraw,
Notepad and Calculator resident. All 29 C code resources and artwork62
actually swapped out. Artwork reloaded at another segment and repainted.
Depth 4 mouse move/reply, exact-board Undo, close/reopen and duplicate
launch refusal passed. Original application owners were fully freed
after close; duplicate launch preserved the complete Game state.
Completed app-load/search overlap was not established. The old 2MiB
baseline pressure result is historical, not repeated in this comparison.

Final German 4MiB smoke passed native labels and genuine umlauts,
no-undo and illegal-input messages, Backspace correction, keyboard and
mouse moves with replies, Undo, New Game, depth 4, close and icon reopen.
Title, button row, longer Black status and footer fit without clipping.
No German system runtime was supplied; the German application was
tested on the supplied English 1.2 system. Check/mate/draw/memory-error
messages were not triggered in this native smoke test.

Final native initial board renders pixel-identically to baseline across
all 65536 compared pixels. Independent measurement-method review passed.
All owned emulators and Xvfb servers were stopped cleanly.

Files: native-summary.json, native-summary.csv, native-searches.csv,
native-memory.csv, native-perf-method.json, final-board-comparison.json,
source-hash-bound raw run data and unmodified screenshots. Compressed
probe CSV files contain observations, not raw guest memory dumps.
Commercial runtime resources, GEOS.INI and raw RAM dumps are excluded.
