TS => MJS. Iteration 4 - #1456

Approving at 0d28a42f. Clean iteration, and the CHANGELOG entry is there this time.

All five exported types (Sign, Compare, Cmp, Cmp1, Cmp2) were already export type in the .ts, so the typedef-visibility issue from #1453/#1454 doesn't arise here — nothing goes from private to public.

Verified rather than eyeballed:

• Every type is unchanged under a strict Equal<A,B> check, including the conditional Cmp2<A,B> (Cmp2<number,number> → number, Cmp2<string,number> → never).
• The self-referential constraint <A extends Cmp1>(a: A) => <B extends Cmp2<A, B>>(b: B) => … survives translation, which was the part most likely to break. The rejection the docstring promises still happens: min(1)("a"), max(1)("a"), cmp(1)("a"), cmp(1)(2n) all fail with TS2345: … not assignable to parameter of type 'never', and cmp({}) fails against Cmp1.
• index3 / index5 still return Index<3> / Index<5> — index3(...) assigned to 3 correctly errors with Type '0 | 1 | 2' is not assignable to type '3'.
• npm test → pass: 2354, fail: 0, total: 2354; npx tsc clean.

Follow-up: stale .f.ts paths are accumulating across the series

Not from this PR, and not blocking it — but it's now big enough to be worth one sweep. On main, 30 lines across 13 files still reference the old .f.ts path of a module that has already migrated:

AGENTS.md                                   2
fjs/types/object/todo/structurally-same.md  2
fjs/types/ordered_map/todo/at-nullable-map.md 2
… 10 more todo files                        1 each

The two in AGENTS.md are the ones I'd fix first, since agents act on that file:

• AGENTS.md:205 tells you to import Assert from fjs/asserts/module.f.ts and Equal from fjs/types/ts/module.f.ts. Both are now .f.mjs, so following that snippet literally produces a broken import — and it's exactly the snippet someone copies when writing a type-level proof.
• AGENTS.md:183 uses fjs/asserts/module.f.ts as its example of an unmigrated helper, which is now the opposite of true — asserts migrated first, in #1452.

Each iteration adds a few more, so a single grep-and-fix pass at the end of stage 1 would probably be cheaper than doing it per-PR. Might be worth a line in the migration todo so it isn't forgotten.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 9076ca8ef1