spot-blindness

Algebraic blindness is using a type too generic to carry the meaning of its own values — bool, bare string/int, blind Maybe/Option/null, untagged tuples. The fix is almost always the same move: name the cases, put them in a type, let the compiler hold the line.

When the detector flags a site:

1. Read checklist.md (same directory) — it has the symptom tells, the 7-question checklist, the fix toolbox, and the "when to stop" guardrails.
2. Identify which fix applies (boolean param → enum; multiple flags → one state type; blind nullable → named result; primitive obsession → newtype; ambiguous predicate → meaningful return type).
3. Run the litmus test first. A type earns its place only if it encodes a distinction that (a) actually exists in the domain, (b) is currently lost or violable, and (c) costs less to maintain than the bugs it prevents. If you can't name the bug the precision prevents, leave the simple type — don't gild.
4. Apply the smallest fix that removes the blindness. It usually deletes validation code rather than adding it.

The detector only flags symptom shapes then asks an LLM to judge; it is not authoritative. If a flag is a false positive (self-evident boolean, speculative variant, harmless newtype), say so and move on.