docs backend evidence maps MAP 29

MAP-29 — Measured performance, compatibility and operational budgets

Phase commit: 1a6a1add (this evidence commit; docs-only revisions after it do not change code).

Status: LOCAL_DONE Validation: PASS for the declared local proxy budgets; PERFORMANCE_GATE PARTIAL Integration: INTEGRATION_PENDING Gate: PERFORMANCE_GATE PARTIAL — the local 7-item proxy matrix passes, but the full MAP-29 matrix still has owed external/device/DB/frame/network items listed below. This is not a national capacity, commercial SLA, physical-battery or release certification.

Contract written before measurement: MAP_29_PERFORMANCE_CONTRACT. Raw sanitized samples: raw-perf.txt. Static receipt: static.txt. Reproduction recipe: scripts/maps/measure-map29.sh.

Environment

• Owned emulator-5554: Android 15/API 35, fingerprint google/sdk_gphone16k_x86_64/emu64xa16k:15/AE3A.240806.043/12960925:userdebug/dev-keys, kernel page size 16,384, app ABI executed arm64-v8a through the host ARM translation, native GLES path, debug app. Host Linux x86_64; JDK 21 (Temurin 21+35); Gradle 8.10.2.
• Admitted regional fixture in app-private files: Brazil_Parana_East.mwm SHA-256 d1e9b2ce81cc10335b6b082a64d78a8241390f42c262e2832f400c981b200a3e, World.mwm 77a35c16…, WorldCoasts.mwm d42a71b1…. Synthetic Curitiba route endpoints; no real GPS/route trace in the receipt.
• Release static candidate: universal unsigned APK 78,396,538 bytes = 74.76 MiB (78.40 MB), SHA-256 feadd9f34d60c26b57a2025e96a1b799c04b52c90da9d47c9784a430363536e0. Existing 80 MiB ceiling unchanged. Debug runtime APK SHA-256 18f283c6…; test APK bc04e1ed….
• Source measured: parent d766dbb4 plus the MAP-27–29 code. The exact tree is the phase commit carrying this file; evidence/docs were added after the measurements and do not change code. No production gate is inferred from this local tree.

Local proxy matrix (3 samples)

Metric	Samples	Declared local budget	Result
Engine init to native readiness	1149, 752, 754 ms	≤ 15,000 ms	PASS
Real 1,800 m route build	410/86/75; 347/87/72; 217/78/74 ms	≤ 5,000 ms each	PASS
Process PSS after routes	351,481; 350,505; 349,163 KiB	≤ 524,288 KiB (512 MiB)	PASS
Native route cancel	0, 0, 0 ms; 0 callbacks after cancel	≤ 1,000 ms, no stale result	PASS
App cold launch am start -W	1218, 1384, 1242 ms (one extra warm sample discarded)	≤ 4,000 ms	PASS
Release universal APK	78,396,538 bytes / 74.76 MiB	≤ 80 MiB	PASS
ABIs + 16 KB packaging	arm64-v8a + armeabi-v7a; every packaged .so LOAD 0x4000; zipalign -c -P 16 4 PASS; no .mwm assets; manifest minSdk 26	supported ABIs, 16 KB, no test maps	PASS (static)

Notes:

• PSS is the Debug.MemoryInfo.totalPss of the instrumentation/app process; it includes test and engine overhead. It is an emulator proxy, not a low-end physical budget.
• Engine init is readiness, not "time to useful map content"; content/frame pacing remains OWED. The MAP-28 UI test separately waits for rendering and passes functionally.
• One earlier perf attempt crashed in the emulator framework after logging init/routes/PSS but before the cancellation step (MonitoringInstrumentation.queueIdle/JIT SIGSEGV). The final direct-cancel runs above all passed. The failed run is not counted.
• The cold launch is app launch on Home, not map content. The extra WARM sample is retained in raw-perf.txt and explicitly discarded; the three COLD samples are the evidence.
• Static ELF alignment is not runtime execution. API 30/API 35 results are not API 26 proof.

Physical reference profile (4 KB, informational)

The same MapPerformanceInstrumentedTest was run 3 times on the authorized phone VKCQGAUCGYAMR4W4 (POCO/duchamp, Android 16/API 36, arm64-v8a, PAGE_SIZE 4096), with the admitted regional fixture and no network provider:

Metric	Samples	Same local budget	Result
Engine init	439, 354, 363 ms	≤ 15,000 ms	PASS (proxy)
Real 1,800 m route	125/46/51; 82/55/57; 85/59/64 ms	≤ 5,000 ms	PASS
Process PSS after routes	276,902; 275,743; 277,633 KiB	≤ 512 MiB	PASS (proxy)
Cancel	0, 0, 0 ms; 0 callbacks	≤ 1,000 ms	PASS
Cold app launch	1250, 1210, 1125 ms	≤ 4,000 ms	PASS

Raw: raw-phone-perf.txt. This is a 4 KB physical profile, not a 16 KB runtime proof; the emulator S2 remains the 16 KB profile. It also does not establish frame pacing, battery/GPS autonomy or low-end/national coverage.

Functional regressions used

• EvTripRealEngineInstrumentedTest OK (3.037 s), EvTripUiInstrumentedTest OK (20.339 s), EvTripSectionAccessibilityTest OK (2.133 s) on the reference emulator. These prove the MAP-27/28 real-core/UI journey; they do not prove frame pacing, battery or national coverage.
• :domain:test :application:test, targeted data/app map/saved/ledger/EV tests, and :app:lintDebug :data:lintDebug :maps:engine:lintDebug pass on the measured source.

Owed / not claimed

• Frame time/pacing/jank for Compose and the native SurfaceView, time to useful map content, repeated rotation/navigation/process-death leak growth. MAP-14's historical 21–27 FPS and SurfaceFlinger queue observations are context only.
• Network bytes/download resume/304/cancel behavior on the current head. MAP-24's 9 → 9 regional proof remains historical evidence, not this run's measurement.
• PostGIS density/concurrency, cursor pagination, EXPLAIN/index checks, job backpressure and regional generation resource ceilings. No SQL/job/schema change was made in this batch and no disposable PostGIS service was used.
• API 26 runtime and clean ARMv7 JNI runtime on this head. Static packaging is not runtime proof; the only attempted ARMv7 profile died with the emulator during a UI run.
• Physical GPS/CPU/battery autonomy, low-end native hardware, dense/sparse national regions, staging CDN behavior, commercial SLA and release certification.
• Release runtime validation; only the release APK static/size/alignment receipt is claimed. Debug runtime results do not transfer to release.

PERFORMANCE_GATE therefore remains PARTIAL, not PASS. No waiver is recorded.