# Leftover — Cycle 3 perf harness execution

**Timestamp**: 2026-05-12T00:00:00Z (UTC+3 wallclock 2026-05-12 ~03:00)
**Reason for deferral**: User skipped the Step 15 (Performance Test) gate. Per `meta-rule.mdc`, performance tests require explicit approval; a skipped question is not approval. Defaulted to skip + record-as-leftover to avoid blocking the autodev's progress through Steps 16-17.

## What was deferred

Execution of `scripts/run-performance-tests.sh` against the cycle-3 build. The harness was made runnable by AZ-492 and would exercise:

- **PT-01..PT-06** — regression probes (latency / status checks for the existing endpoints with JWT attach, post-AZ-487 + post-AZ-494)
- **PT-07** — cold + warm region-tile distribution (cold p50/p95 vs warm p50/p95; fails if warm p95 ≥ cold p95)
- **PT-08** — UAV batch upload distribution (batch p50/p95 + per-item proxy; gated at AZ-488 NFR of batch p95 ≤ 2000 ms)

## Pre-requisites for replay

The harness needs the same env vars that the cycle-3 test run consumes:

- `JWT_SECRET` — ≥ 32 bytes; already in `.env`
- `JWT_ISSUER` — already in `.env` as DEV-ONLY (AZ-494)
- `JWT_AUDIENCE` — already in `.env` as DEV-ONLY (AZ-494)
- `GOOGLE_MAPS_API_KEY` — already in `.env`

Optionally:
- `PERF_REPEAT_COUNT` (default 20)
- `PERF_UAV_BATCH_SIZE` (default 10)

## How to replay

```bash
./scripts/run-performance-tests.sh
```

(brings up its own docker-compose stack; cleans up on exit; ~3-5 minutes for default counts)

## Why this is NOT a hard blocker

- The cycle-3 implementation report and code review verdicts already note that the perf harness was statically verified (script grep + integration-test compile + AZ-492 AC-1/AC-4/AC-5/AC-6 covered).
- The AZ-488 batch-p95 threshold was set in cycle 2 and existing integration tests do NOT regress at the cycle-3 build (Step 11 full suite all-green).
- No cycle-3 change altered the production hot paths beyond JWT validation (AZ-494 adds two string comparisons per request — sub-microsecond).
- The cycle-2 deploy also skipped this gate (Option B) without negative consequences.

## Replay obligation

Per `tracker.mdc` Leftovers Mechanism: at the start of the next `/autodev` invocation, this leftover should be replayed by running the perf harness, then either deleted (if it passes) or updated with the failure detail.