[jonathanprozzi]

Important

@kamescg — review ready. Stacked on #39 (codex/p16d-classifications-parity); review after #39 and retarget as #39 and #36 land.

To ratify here: the no-identifier fallback envelope. The reference builds a validated P1 envelope with no identifier when no rung fires, the ladder is null, or a canonical URL exceeds 220 characters; the train used to refuse it. The anchor result's iid, scheme and class are therefore nullable and consumers must narrow.

Coordination: this PR and #40 both touch packages/primitives/src/types.ts (whichever lands second takes IdentityClass | null). One harness line changed: encode proofs register identified anchors only.

Receipt: byte parity on all 52 agreeing anchor vectors; 19 ledger entries retired (anchor proof now 52 agreements / 6 divergences / 23 OPEN); harness 1027 pass / 0 fail.

Child PR of the v1.1.0-alpha umbrella (#15). Part of the parity series: each PR in the series makes a train package behave exactly as the reference implementation's in-tree copy does today, so the published packages can replace those copies without a behaviour change.

Current (train)

@0xintuition/primitives built identity anchors whose bytes differed from the reference's for the same input in several families: raw evidence was normalized before hashing, derived recipe values (geohash, year, hashed text) were not materialized under the recipe key, sameAs was dropped from identity envelopes or kept unsorted, descriptive fields leaked into scheme envelopes, and a classification with no identifier at all was refused where the reference builds a P1 fallback envelope. The conformance harness recorded 26 anchor divergences for these.

Need

Byte parity of every anchor vector with the reference builder, the no-identifier fallback envelope as the reference builds it, and the conformance ledger moved to match.

Reconciliation

src/anchor.ts and src/types.ts:

• Headers: declared classification context and type, then the canonical identifier; no context override table.
• Raw evidence preserved before NORM-1 (punctuation and padding kept); finite numeric scalars stringified as the reference does.
• Recipe order: direct-only recipes keep sorted keys; recipes with transforms keep declaration order.
• Derived values materialized under the recipe key: geohash at the declared precision, the four-digit year from a date, keccak16 of the NORM-1 text.
• sameAs as the last identity field: strings verbatim, exact-deduplicated, default-sorted, omitted when empty or not an array.
• Inclusion: scheme envelopes omit descriptive fields; gen1 envelopes include only the producing rung's evidence; the identity path skips whole-map validation.
• Account floor: the weak platform:@handle form emits P1, the strong form keeps P0 (consistent with feat(iid, iid-spec): floor weak account handles below anchor eligibility, with the spec amendment and conformance rule (P16g) #35).
• Fallback: no firing rung, a null ladder, or a canonical int:url: value over 220 characters produces a validated P1 envelope without identifier; iid, scheme and class are null on that result, so consumers that assume every successful anchor carries an IID must narrow. Unknown, missing or invalid fields still fail.
• Serialization: compact JSON in insertion order, UTF-8 hex, the atom id over the exact bytes.
• Conformance harness: support/proofs.ts registers encode proofs for identified anchors only, since a no-identifier envelope has no URI to encode (the encode inventory pinned in the manifest is unchanged).
• Conformance ledger: 19 anchor entries retired because both sides now agree (17 builder defects and the two fallback-envelope rulings); the six "reference adopts the train's builder" entries stay, with their fingerprints intact. README enumeration: anchor proof 52 agreements / 6 divergences / 23 OPEN (was 33 / 25 / 23).

Parity

Matches the reference implementation's anchor builder byte for byte on all 52 agreeing vectors (src/__tests__/anchor-parity.test.ts pins every one: JSON, hex, atom id, IID or null, profile). The six remaining anchor divergences are the cases where the reference is expected to adopt the train's field order; they are documented in the ledger, not changed here. The primitives README now describes the reproducible canonicalization and hashing; the IID spec should carry that as a section in a later cut.

Held out

• No version bump; the release PR freezes versions for the whole family at once.
• The 23 OPEN anchor capabilities in the harness stay OPEN.
• src/types.ts also changes in feat(iid-spec, iid, iid-ladder, iid-registry): register provider-local identifiers as class D with the reference's grammar, typing and P1 floor (P16f) #40 (the class union widens to include class D); whichever lands second takes IdentityClass | null.

Evidence

• packages/primitives: typecheck, lint and build green; 192 tests (the 52 literal parity pins, 19 red-first regressions that failed before the builder change, and the pre-existing suite).
• Conformance harness on top of test(integration): reconcile merged conformance gates and enforce them in CI #36 and feat(classifications, iid): bring the catalogue, ladders and helpers to parity with the reference implementation and derive one podcast GUID per feed (P16d) #39: 1027 pass, 23 todo, 0 fail; README table equals the harness's own enumeration.
• Root registry:check (14 packages), guard:supply-chain, pack:dry-run pass.
• Independent review: the conformance pass found every item conforming (one wording fix in the harness README, applied); the adversarial pass ran the reference builder on 15 agreeing vectors, four builder mutations (each turning named parity tests red), the fallback edges (220 vs 221 characters, null ladder, all rungs missing, required field missing), every reader of an anchor's identity fields in the repo, and recomputed 10 retired ledger cases. No findings.

Review notes

Base codex/p16d-classifications-parity (#39), which sits on fix/pr15-integration-gates (#36); retarget as those land. Read the ledger diff first (19 retirements), then src/anchor.ts top to bottom, then the parity test. The one judgement call is the fallback envelope (the reference builds it; the train used to refuse it); it is in effect by default and open to ratification here.