fix(docs): runbook must not assume levelcode_stripe_id exists

[ndemianc]

Follow-up to #406, which merged before this landed. Running step 1 on production died immediately:

PG::UndefinedColumn: ERROR:  column "levelcode_stripe_id" does not exist

The actual finding

levelcode_stripe_id is in db/schema.rb and in the model annotations, but it is not in the production database — and no migration adds it:

$ grep -rl levelcode_stripe_id db/migrate/
(nothing)

It entered the tree in b7b2cfc ("Add annotations", 2026-07-22), a commit that touched only db/schema.rb and annotation comments. An annotate run regenerated the schema from a local database where the column had been added by hand, and that snapshot was committed.

The consequence is a clean split by how each database was built:

Built by	Has the column?
db:schema:load (development, test)	yes — User.first.levelcode_stripe_id → nil
migrations (production)	no — the attribute does not exist

db:migrate will never create it, because nothing defines it. This is why it looked like a harmless dead column from the application code and only surfaced when the runbook actually queried it on production.

The fix

Step 1 now asks the database rather than assuming:

id_cols = User.column_names & %w[stripe_id levelcode_stripe_id]

That drives the known-ids scan, the per-email account map, and the pre-delete re-check, and it prints what it matched on so the operator can see which shape they are running against.

Verification

Blocks extracted from the markdown and run twice against test-mode Stripe:

• as-is → matching on: stripe_id, levelcode_stripe_id
• with User.column_names patched to hide the column, reproducing production exactly → matching on: stripe_id

Both complete without error, with identical classification (15 RETRY / 12 REVIEW_ACTIVITY / 278 other of 305) and the customer count unchanged at 321.

Not fixed here

The schema drift itself needs a decision, not a quiet edit in a docs PR: either add a real migration if LevelCode billing separation is genuinely wanted, or drop the column from db/schema.rb and the annotations if it is not. Recorded in the runbook's limits section. Worth noting the annotations in app/models/user.rb, spec/models/user_spec.rb and spec/factories/users.rb currently describe a column production does not have, including a UNIQUE index that does not exist there.

[Copilot]

Pull request overview

Updates the Stripe orphan customer cleanup runbook to be resilient to schema drift between environments by determining which Stripe ID columns actually exist on users at runtime (instead of assuming levelcode_stripe_id exists).

Changes:

• Detects available Stripe customer-id columns via User.column_names and uses them throughout Step 1 (scan + account map).
• Prints which columns were matched so operators can confirm the database “shape” they’re running against.
• Documents the underlying schema drift and calls out that it requires a separate decision/migration.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.