P18-T02 — Performance and memory baselines

Status: LOCAL_VALIDATION_IMPLEMENTED / FULL_ACCEPTANCE_BLOCKED on codex/phase-18-functional-acceptance (second task; PR #67). Issue: #61. Docs only: every measured budget met, so per "profile before optimization" no hot path was touched.

Baselines (this host, 2026-10-01)

• Backend load smoke (test-load.sh, disposable dev PostGIS, 500 seeded stations, 20 s at concurrency 8): 52,636 requests, p95 15.1 ms (budget 300 ms), 5xx rate 0.0000 (budget 0.01) — 20× latency headroom. Fault suite 7/7 (outage refuses/fails closed, recovery, no unauthorized upload under storage outage, capped disk pressure).
• Photo memory (TestSanitizeForwardMeasuredMemory): 2 MP sanitize emits 71,731 wire bytes (target 150 KiB, cap 256 KiB) with 3,329,392 bytes heap growth vs 32 MiB hypothesis — unchanged from P15, 10× headroom, no regression.
• Android battery + assembleDebug: BUILD SUCCESSFUL (no main-thread/decode changes in this phase to profile further on-host).

Optimization decision

None: no budget breached, no new allocation hotspot found, so no code changed. Inventing an optimization without a violating profile would be speculative work against the task outline.

Validation (exact commands, this host)

bash scripts/tests/test-load.sh   # EXIT=0, log /tmp/p18-t02-load.log
go test -count=1 -v -run TestSanitizeForwardMeasuredMemory ./internal/modules/evidence/adapters/media/
./gradlew :domain:test :application:test :data:testDebugUnitTest :app:testDebugUnitTest :app:assembleDebug --no-daemon
git diff --check
bash scripts/scan-secrets.sh

Dev DB stopped after the run (named volume retained by the dev topology, no seed residue outside b000… rows it already owns).

Limits (not claimed)

• Physical Android follow-up: existing cached-home readiness test passed at 382 ms vs 2000 ms budget (P18 local exit). Complete cold-process/frame/heap/battery and iOS device matrix remains unproven; no macOS/iOS device.
• 30-minute staging matrix (100 k stations, 1 M rows): runs on provisioned staging per harness design, never here.
• No release claim by these baselines.

Publication reconciliation: PR #67 integrates the validated local slice only. The original task acceptance remains unmet; see current exit and blockers. Historical command results above are retained and are not relabeled device proof.