name: review-after-each-step description: Operator wants a grounding review after EVERY step (not just big checkpoints) to prevent drift metadata: node_type: memory type: feedback
On multi-step work (e.g. the Stage-3 review sweep, phased implementations), the operator wants a review completed after each step to stay grounded and stop drift accumulating.
Why: drift compounds silently across steps -- a number, a comment, or a status that disagrees between code, the decision record, and the docs. The after-each-step review is what caught the 32-vs-16 GiB containment-overhead drift between the Phase-C code and the Phase-A record (design-decisions), before it propagated.
How to apply: after each step, before moving to the next -- (1) validate/test the change (tofu validate, harness, repo-lint); (2) grounding check against the agreed decisions + structure (does it still implement the ruling faithfully?); (3) drift sweep -- reconcile code vs the decision record vs the docs/ledger (numbers, comments, statuses, line refs must agree); (4) for non-trivial steps, an independent agent cross-check (read-only) as the operator asked for [[stage3 sweep]]. Fix what the review finds THEN proceed. Log the review outcome in the session-ledger step by step.