Skip to content

Walkthrough: Core Release Path

This is the shortest operator path to release confidence.

Goal

Confirm the core release path, then optionally add representative policy evidence if your pipeline includes backend-swappable policy inference.

Core release path: 1. backend behavior stays within parity tolerance 2. incident diagnosis is replay-consistent 3. acceptance-gate summary gives a defensible GO/NO-GO

Specialized add-on: 4. policy backend evidence is available for representative policy pipelines

Core Command Chain

Run from retriever/:

pixi run verify-backend-parity
pixi run verify-incident-replay
pixi run python -m examples.tutorial.h_release_readiness.02_release_readiness_walkthrough

Task-oriented equivalent:

pixi run verify-backend-parity
pixi run verify-incident-replay
pixi run demo-release-readiness

Optional Specialized Add-On

Use this only if your pipeline includes backend-swappable policy inference:

pixi run demo-policy-backends

What Each Step Proves

Optional Step: Policy Backend Evidence (TUT-027, specialized)

Command: 

pixi run demo-policy-backends

Proves: - backends openpi_pi05|lerobot|mock are selected by config/interface - the same policy example and infer(example)->actions contract run across all three backends

Artifact: - logs/tutorial_policy/tut027_backend_metrics.csv

This step is specialized evidence for policy-driven pipelines, not a universal prerequisite for every Retriever deployment.

Step 1: Backend Parity Gate (TUT-032)

Command: 

pixi run verify-backend-parity

Proves: - multiprocessing and dora both run - parity checks pass under configured tolerances

Artifacts: - logs/tutorial_parity/tut032_backend_parity.json - logs/tutorial_parity/tut032_backend_parity_checks.csv

Step 2: Incident Replay Gate (TUT-033)

Command: 

pixi run verify-incident-replay

Proves: - incident root-cause candidate is detected - replay diagnosis signature matches live diagnosis

Artifacts: - logs/tutorial_incident/tut033_incident_report.json - logs/tutorial_incident/tut033_incident_checklist.md

Step 3: Release Walkthrough (TUT-029)

Command: 

pixi run python -m examples.tutorial.h_release_readiness.02_release_readiness_walkthrough

Proves: - evidence maps to acceptance gates - final decision is explicit (GO or NO-GO) with reasons

Artifacts: - logs/tutorial_release_readiness/tut029_release_checklist.md - logs/tutorial_release_readiness/tut029_release_summary.json

Decision Rule

  • GO: all gate checks pass and no required artifact is missing.
  • NO-GO: any parity/incident/acceptance gate check fails.

Fast Triage If It Fails

  • fails at Step 1: fix backend runtime/parity contract first
  • fails at Step 2: fix replay determinism or diagnosis logic
  • fails at Step 3: fix missing evidence/docs linkage and re-run