[hugomallet]

Why

First brick of the AJV 8 track: formatErrors becomes shape-agnostic so the future AJV 8 bump (separate PR, v1 branch) and a potential Standard Schema adapter only have to swap the validator, not the error plumbing. AJV 8 replaced dataPath (dot/bracket notation) with instancePath (JSON Pointer, RFC 6901).

What

• formatErrors reads error.instancePath when it is a string, and falls back to the legacy error.dataPath otherwise (the active branch with the bundled AJV 6).
• New exported helper pointerToFieldPath(pointer) — JSON Pointer to dot-separated field path, with RFC 6901 escape decoding (~1 before ~0, so ~01 correctly decodes to a literal ~1).
• The required concatenation is now conditional (field ? field + '.' + missing : missing), removing the orphan-leading-dot dance at the root.

error input	field (before)	field (after)
dataPath: 'arr[0].input'	arr.0.input	arr.0.input (unchanged)
dataPath: '' + required input	input	input (unchanged)
instancePath: '/arr/0/input'	n/a (undefined behavior)	arr.0.input
instancePath: '' + required input	n/a	input
instancePath: '/a~01b'	n/a	a~1b

Non-breaking, proven

• The 5 pre-existing dataPath tests are untouched (the test file diff is purely additive) and still pass — strict non-regression under AJV 6.
• 11 new tests: mirror instancePath set (same cases as the dataPath set), pointer edge cases (root pointer, ~0/~1 escapes and their decoding order, required inside an array item), plus direct pointerToFieldPath unit tests with literal expectations.
• Each new test was mutation-checked: swapping the ~0/~1 decoding order, dropping the slice(1), making the required concat unconditional, and weakening the typeof check to truthiness each make the corresponding tests fail for the expected reason.
• Full suite: 113 tests before, 124 after, all green. yarn lint and yarn type-check green.

[hugomallet]

Tooled self-review (claude + review53 skill) — posted as comment, author and reviewer are the same account.

Verdict: LGTM, no correctness issue found. ESLint + tsc green on changed files, 124/124 tests pass, CI fully green (build + types React 16–19).

Verified beyond the suite:

• RFC 6901 edges (tested locally): lone / → '', /a//b → a..b, /0 → 0, /~01 → ~1. All decode per the RFC; the / vs root-pointer collision (both → '') is a representational limit of dot paths, not a decoding bug.
• dataPath equivalence: ran the old master implementation differentially against the new code on 8 dataPath cases, including dataPath: '[0]' + required — outputs strictly identical, bug-for-bug (both keep the pre-existing leading dot on root-array dataPath, .0.x).
• Dispatch predicate: typeof instancePath === 'string' is the right choice — truthiness would misroute AJV 8's root ''.

Minor, non-blocking:

1. CHANGELOG overstates the export. pointerToFieldPath is exported from helpers.js only; the package exports map exposes just dist/index.js, which does not re-export it, so it is not user-visible. Reword ("internal helper") or re-export from the public barrel if intended.
2. Neither-shape error objects now always throw (formatted.dataPath.replace on undefined). Old code threw too, except required errors which produced garbage "undefined.x". Loud failure on degenerate input is fine — just noting the divergence.
3. Inherited cross-shape quirk: root-array dataPath '[0]' → .0 (leading dot) vs instancePath '/0' → 0. Dies with the dataPath branch in PR 2 — no action.
4. Optional: pin lone / and empty-segment (/a//b) behavior with two tests.

[hugomallet]

Review comments addressed in 889c461 (single round-trip):

1. CHANGELOG — reworded: pointerToFieldPath is described as a new internal helper (groundwork for AJV 8), since it is not reachable through the package exports map. No public export added in this PR.
2. Degenerate input — an error with neither instancePath nor dataPath no longer throws: the missing path is treated as the root (dataPath ?? ''). Test added ({keyword: 'required', params: {missingProperty: 'x'}} → field 'x', no throw); mutation-checked (dropping the ?? '' makes it fail with the exact TypeError).
3. Optional pointer tests added — lone-slash '/' → '' (root collision documented in the test comment) and '/a//b' → 'a..b'; both mutation-checked (filtering empty segments / dropping slice(1) makes them fail).

The cross-shape [0] quirk is deliberately left for PR 2 as agreed.

Suite: 124 → 127 tests, all green; lint and type-check green.