verify-completeness

Verify completeness

A cleanup or verification pass fails the moment it starts to feel procedural. The instinct "I fixed all instances of X" is exactly where singletons survive: the one match that did not fit the mental pattern, the capitalized variant, the synonym you never searched for. The reliable method is slower per pass and faster to actually-done. Do not summarize-then-apply. Enumerate, tick off each spot individually, and iterate the search until it comes back empty.

When to use

• Removing every occurrence of a token, secret, deprecated API, or debug hook.
• Applying a mechanical fix reported as a list of file:line findings.
• Any sweep where one survivor is a real defect: security, compliance, a public share, a rename that must be total.

When not to

• A single, known edit at a known location. There is nothing to enumerate.
• Exploratory work where "good enough for now" is an acceptable stopping point.

Procedure

1. Enumerate into atomic checklist items. Turn the search result into one line item per exact spot -- each file:line, each match, its own box. Do not collapse "12 hits in service A" into a single item. You tick a box only after that specific location is fixed and re-read. Twelve hits means twelve boxes.

2. Fix, then re-run the SAME search. After a batch of edits, run the identical comprehensive search again. If it returns anything at all, you are not done -- fix the new items and search again. Repeat until the search returns empty. The empty result, not your sense of having handled it, is the completion signal.

3. Search case-insensitively. A case-sensitive sweep silently skips Token, TOKEN, and token when you searched for one form. Pass the ignore-case flag (grep -i, rg -i) by default on cleanup work.

4. Expand the pattern as variants surface. Every pass teaches you a new spelling: a hyphen vs underscore, an alias, a wrapper name, a comment form. Add it to the pattern and re-run from the top. The search you finish with is almost never the search you started with.