# Experiment 1 repeat B — saturated rolling deploy

**Purpose:** second replicate for DEPLOYMENT_BLOCKED latency distribution.

**Run date/time (UTC):** 2026-04-17T17:26:31Z

**Setup:** identical to repeat A (desired=3, EXPIRES_MINUTES=5, saturated).

**Results:**
- Time from trigger to first `enable_failed`: **92.2 s**
- Spread of 3 `enable_failed` events: **217 ms**
- Convergence: **175 s**
- All 3 old tasks exited via `DEPLOYMENT_BLOCKED` + `worker_exit` (clean)
- SIGTERM events: **0**

**Collection method:** `run_saturated_deploy.sh` wrapper — 45 s post-convergence
wait + `filter-log-events` pull. Clean single capture, no re-query needed.

## Raw data

- `worker-logs.jsonl` — authoritative, via `filter-log-events`
- `service-state.jsonl`, `service-events.jsonl`, `task-state.jsonl` —
  state snapshots
- `trigger.txt`