Review Route Guide

Guide Maps

graph TD
  question["Review question"] --> summary["Summary commands"]
  question --> verify["Verification report"]
  question --> release["Release review"]
  question --> recovery["Recovery review"]
  question --> tour["Tour or walkthrough"]

flowchart LR
  ask["Name the question precisely"] --> route["Choose the smallest honest route"]
  route --> inspect["Read the saved outputs or command result"]
  inspect --> escalate["Escalate only if the question is still open"]

Use this guide when the capstone now has enough commands and bundles that route choice could become its own source of noise. The goal is not to memorize targets. The goal is to match each review question to the smallest honest evidence surface.

Fast route selection

If the question is about...	Start with
stage ownership and declared versus recorded edges	make stage-summary
promoted population facts	make profile-summary
promoted scoring behavior	make model-summary
promoted inventory and training metadata	make manifest-summary
borderline decisions near the threshold	make threshold-review
concrete misclassifications	make review-queue and publish/v1/predictions.csv
whether the promoted contract currently passes	make verify
saved contract evidence plus enforcement logic	make verify-report
downstream trust in the promoted release boundary	make release-review
remote-backed restoration after local loss	make recovery-review
first-read orientation through the repository	make walkthrough
broader executed proof across repository state	make tour
strongest built-in local bar	make confirm

Escalation order

1. Start with one summary command when the question is narrow.
2. Use make verify when the question becomes contract validity.
3. Use make verify-report, make release-review, or make recovery-review when the answer must survive later inspection.
4. Use make confirm only when the whole repository story is under pressure.

What this guide prevents

• defaulting to confirm when a smaller route would answer the question more cleanly
• using release review when the question is still only about stage truth
• using one promoted metric as a substitute for profile, model, or record-level review
• treating recovery review as if it replaced ordinary release review