[Factory] PR candidate: WG-MOT9XTAO-3QNU

docs/operations/production-smoke.md

@@ -0,0 +1,74 @@
+# Function Factory Production Smoke Test
+
+## Purpose
+Verify that the Function Factory production environment is healthy and capable of executing a full pipeline from signal ingestion through synthesis. This procedure does **not** change runtime code.
+
+## Prerequisites
+- Access to `ff-pipeline` and `ff-gateway` in the production environment
+- Valid authentication tokens / network access
+- A text editor or runbook to record IDs and results
+
+## Procedure
+
+### 1. Health Checks
+Query the following `ff-pipeline` debug endpoints and confirm healthy responses.
+
+| Endpoint | Expected Healthy Output |
+|---|---|
+| `GET /version` | Returns current application version string (e.g., `1.2.3`). |
+| `GET /debug/health` | Status `200 OK`; body indicates overall system health. |
+| `GET /debug/arango` | Status `200 OK`; confirms connectivity to ArangoDB. |
+| `GET /debug/ai-test` | Status `200 OK`; confirms AI/LLM backend reachability. |
+
+Record any anomalies. All four must pass before proceeding.
+
+### 2. Start a Dry-Run Pipeline
+Submit a smoke-test signal through `ff-gateway` configured for **dry-run** mode.
+
+- Ensure the pipeline configuration specifies `dryRun: true`.
+- Capture the returned **`pipelineId`** and **`signalId`**.
+- Note the **`workGraphId`** generated after pipeline initiation.
+
+### 3. Approve the Pipeline
+Using the administrative interface or API:
+- Locate the pipeline by its `pipelineId`.
+- Execute the approval action so synthesis may proceed.
+
+### 4. Confirm Synthesis-Passed
+Monitor the pipeline until the synthesis stage completes.
+
+- Verify the stage status is **`synthesis-passed`**.
+- Record the **`workGraphId`** (if not already captured).
+- Record the **Gate 1 result** (pass / fail / pending) as shown in the pipeline state.
+
+### 5. Optional: Real-Mode Docs-Only Smoke
+To validate model-backed execution without risking code mutation, run one **real-mode** pipeline restricted to **docs-only** scope:
+
+- Submit a signal with `dryRun: false` and scope limited to documentation changes only.
+- This confirms that LLM inference, cost tracking, and artifact generation function in production while guaranteeing no runtime code is modified.
+- Record the resulting `pipelineId`, `signalId`, and final verdict separately.
+
+### 6. Final Verdict
+Based on the above observations, assign one of the following verdicts:
+
+- **PASS** - All health checks green, dry-run completes end-to-end, optional real-mode docs-only succeeds (if performed).
+- **FAIL** - Any health check fails, pipeline stalls, or synthesis does not pass.
+- **DEFERRED** - Infrastructure issues block execution; escalate before release.
+
+## Recording Sheet
+
+| Field | Value |
+|---|---|
+| Pipeline ID | |
+| Signal ID | |
+| WorkGraph ID | |
+| Gate 1 Result | |
+| Final Verdict | |
+| Operator | |
+| Timestamp | |
+
+## Success Criteria
+- An operator can follow this document manually without referring to source code.
+- Expected healthy outputs are listed for each health endpoint.
+- The docs-only real-mode smoke is explained as a way to validate model-backed execution without mutating runtime code.
+- No runtime code is changed by executing this procedure.