IRI and Identity Policy (historical)

Status: Historical record. Not normative, not a reference profile, and not guidance. Superseded by: IRI and Identity Patterns and ADR-A82, which supersedes ADR-A51 Body removed: 2026-09-23, following iri-patterns-post-3866b21-review.md finding D7. The disposition is in iri-patterns-post-3866b21-remediation.md.

What this document was

The P0.1.3 deliverable: one proposed platform IRI profile for LATTICE, reviewed under ADR-A51. It specified a urn:lattice:{tenantId}:{scope}:{family}:{localName} lineage grammar, content revision and event IRIs beneath it, four entity minting strategies (natural-key, derived-hash, surrogate-claimed, surrogate), a content-hash procedure, a skolemization rule and a graph-name validator. The review history showed that a framework cannot mandate one identifier scheme for every adopter, and ADR-A82 replaced the profile with the framework-neutral pattern catalogue in iri-identity-patterns.md.

The body was removed rather than kept as “historical context” because it still read as guidance, and parts of it contradict the current catalogue. Section numbers cited by ADR-A51 and by the ADR-A51 review records refer to the removed body, which remains in version control:

git show 3323b85:docs/architecture/iri-policy.md

Known defects in the removed body

Anyone reading the historical body should treat these as errors, not as options:

• Hash-only literal canonicalisation. Its §3 canonicalised literal lexical forms before hashing while storing the original terms. Two different stored graphs could then share one content IRI. iri-identity-patterns.md §8.2 forbids this.
• Contradictory skolemization. Its §5 said skolem IRIs are deterministic within a content revision while prescribing random UUIDv4 surrogate IRIs for nodes without a natural key.
• Lowercase rule versus its own example. Rule 7 made the canonical lexical form all-lowercase, while the natural-key example …/Policy/POL-000123 is mixed case.
• Invalid UUID example. The surrogate example 01j9z8q3-k4m5-n6p7-… is not a UUID: it contains non-hexadecimal characters.
• Delimiter concatenation. The derived-hash form hashed v{schemeVersion}|{ns}|enc(keyTuple), a separator join around an undefined enc. iri-identity-patterns.md §7.2 requires a self-delimiting tuple encoding for the whole input.
• “At least 128 bits”. An open-ended digest width lets two implementations mint different IRIs for one key. iri-identity-patterns.md §7.4 requires an exact width.
• Status claims. The body described its rules as “normative”, surrogate-claimed as the “Recommended default”, and its validator as the “single source of truth”. None of these hold.