sql-backfill-pipeline-scaffoldlisted

# SQL Backfill Pipeline Scaffold Use when recovering a corrupted value requires walking a chain across two or more systems (e.g. a local DB → a join key into another system's log table → the original value extracted from a logged JSON response), and the mutation will eventually be handed to someone else (a DBA, another team) to run against production. ## Inputs Ask for, or infer from the conversation, before scaffolding: 1. What was corrupted, and the chain of systems/DBs needed to recover the correct value (name each system and the join key between consecutive hops). 2. Roughly how many rows are affected. This determines which mutate shape to generate — see "Verify/mutate split" below. If unknown, ask; do not default to the small-batch shape silently. 3. Who runs the final mutation — you, in this session, or a hand-off to someone else. A hand-off raises the bar on self-guarding. ## Scaffold shape ``` docs/ plan.md # what's corrupted, the recovery chain, current status resume-prompt.md # "paste into a fresh session" prompt with a living status section phases/ phase1-<name>.sql # RUN AGAINST: <DB> banner comment at the top of every phase file phase2-<name>.sql ... scripts/ batch-ids.py # --table/--column flags; one generalized helper, not one per phase command-steps.md # file -> DB target -> input -> output, plus exact shell commands ``` - Every `phaseN` SQL file starts with a `-- RUN AGAIN