Surface Layer — Outstanding Items

Source: Consolidated from ontology/surface/docs/OUTSTANDING-ITEMS.md (2026-09-18)
Status: Living reference document; all items tracked in surface-mork-unified-projection.md
Last updated: 2026-09-22

Overview

This document tracks specific outstanding items, design decisions, deferred work, and findings that emerged during Surface-MORK unified projection implementation (Phases 0–8). For phase-level status, see surface-mork-unified-projection.md.

All items listed here are deliberately not in shipped artefacts — no README annotations, OWL declarations, shape messages, or generated modules carry these items.

1. Decisions Taken During Implementation

1.1 Signature scope and law X6 — ✅ Closed

Rationale: The design note stated conservativity (X1) unconditionally: a surface never extends source-signature consequence set. This holds for indexing (mints own terms) but NOT for promotion (restates values onto srf:promotesTo). When srf:promotesTo is an authored property, emitted triples are over the authored signature and indistinguishable from authored facts.

The MERIDIAN CSO→FBO case: fbo:hasLineOfBusiness is authored, so DimensionSourcing is producing authored-signature A-box, not an index.

Solution implemented:

• srf:signatureScope on every srf:GeneratedSurface (values: srf:LocalSignature or srf:SourceSignature)
• X1 restated to apply to local-signature surfaces only
• Law X6: promotion onto a property outside the contract’s target namespace declares exact or crosswalk-exact fidelity and is materialised rather than definitional

Sign-off: ADR-A16 addendum + ADR-A21 (composition rule across stacked surfaces).

1.2 Read paths as indexed step lists, not SHACL paths — ✅ Done

Decision: Use srf:PathStep individuals with srf:stepIndex and srf:stepDirection instead of SHACL path nodes.

Why: SHACL path syntax nests blank nodes; labels don’t survive canonicalisation without normalisation. Would have made read paths the only part of contract needing special handling.

Result: srf:PathStep gives sequence-and-inverse expressivity with a form that hashes directly, following qnt:OrderingComponent pattern. No SHACL engine needed. Alternation and zero-or-more remain out of scope.

Tradeoff: Contract cannot be handed to SHACL engine as path expression; compiler must synthesise one. Nothing currently needs that.

1.3 Derivation contract location — ✅ Closed (2026-09-25)

Resolution: Foundation declares fnd:DerivedArtefact, fnd:DerivationRun and fnd:DerivationKind, and srf:DerivedArtefact subclasses it (ADR-A92, AOR-12).

Question: Does derived-artefact contract eventually move to Foundation per ADR-A12, or does each generating layer declare its own?

Current state: srf:DerivedArtefact, srf:GeneratedSurface, srf:ReadSetEntry, srf:LawDischarge, srf:derivationAuthority, and hash properties all declared in Surface, not Foundation.

Why it matters: ADR-A12 anticipates Foundation-level fnd:DerivedArtefact; Quantification §12 and Behaviour both name it as a blocking gap. Validation reports, entailment sets, and materialised bins want the same contract.

Implication: When Foundation gains it, Surface’s classes should become subclasses, not restructured. Classes are shaped to allow this, but migration is unplanned and duplication is real in the meantime.

Related: ADR-A22 resolves ontology/governance/versioning alignment for MORK; this broader Foundation question is unaffected.

1.4 Generated-term markers are parentage, not types — ✅ Done

Design: Every generated class is rdfs:subClassOf srf:GeneratedClass; every generated relation is rdfs:subPropertyOf srf:generatedRelation.

Why: MERIDIAN’s approach (asserting owl:Class with RDF properties, creating punning) causes OWL-API toolchain issues. Parentage avoids this.

Result: Enumerating a surface inventory is a subclass query; no punning. Provenance lives on srf:GeneratedSymbol records in manifest.

1.5 Contract and example locations — ✅ Done

Layout:

• Generated packages: ontology/surface/execution/<contractKey>/
• Example contracts: ontology/surface/examples/
• Root ontology examples: ontology/examples/ (for cross-layer composition, not single-layer worked examples)

Status: Accepted; no open action.

2. Deferred with Planned Approach

2.1 srf:RangePartitionPopulation — Deferred, planned

What it is: Bucketing law X7; creates nominal bucket classes for ranges within a dimension.

Blocker: Depends on Quantification partition semantics first.

Five-step unblocking path:

1. State the partition precondition (Quantification work): qnt:PartitionRangeSet subclass or qnt:isPartition flag with shape checking disjointness and coverage
2. Settle boundary behaviour under each closure (Quantification): ranges meeting at b must be non-overlapping and complete
3. Decide what the symbol denotes (Quantification + Surface): nominal bucket class needs owl:onDatatype with facets; profile declares srf:OWL2DLEntailment or materialises
4. Naming policy (Surface): ranges need labels, or use DigestLocalName, or mint from bounds (bucket_0_100, etc.)
5. Invalidation (Surface): read-set entry under existing freshness rule; no new machinery

Effort: Most work is steps 1–2 (Quantification). Surface’s part is small once partition guarantee exists.

2.2 Stacking beyond depth 1 — Deferred, planned

Current limit: srf:StackDepthReleaseCeilingShape caps at depth 1; deleting it lifts the cap.

What exists: srf:permittedStackDepth, srf:stackDepth, srf:SurfaceSource read-set entries, srf:S10 check.

What must exist first:

Current use: FBO chain (CSO → promoted dimension → nominal index) is depth 1; works today.

2.3 srf:ExternalIndex admission criteria — Deferred

What it is: Index over an external store without materialisation into the core graph.

Status: Declared and rejected (per bhv:Proportional precedent — visibly unavailable rather than silently absent).

What’s needed:

• Way to name the store and index without substrate depending on specific store
• Statement of what “surface is fresh” means when index cannot be hashed
• Decision about whether store-native index can discharge parity (R2) at all

Timeline: Left until a deployment actually needs it.

3. Known Gaps

3.1 Foundation migration for derived-artefact contract — ✅ Closed (2026-09-25)

See §1.3 above. Resolved by ADR-A92: srf:DerivedArtefact ⊑ fnd:DerivedArtefact (applied-ontology-readiness AOR-12). §3.3 (profile identity) is no longer blocked on it.

3.2 ADR-A01 convention conflict — ✅ Closed

Resolution: Ontology header is authored in README’s turtle-spec fence, per ADR-A01. Matches Eligibility, Behaviour, and Surface practice. Foundation, Vocabulary, and Party are outliers to correct.

3.3 Profile identity is computed, not modelled — ⏳ Unblocked, not started

Current state: Profile.identity() in compiler concatenates profile IRI, generator version, canonicalisation version, entailment regime, naming normalisation, symbol mode, and permitted stack depth. Pure computation, not represented in graph.

Missing: srf:profileIdentityHash asserted on generated surface would close this.

Blocker: None since §3.1 closed (2026-09-25). A Surface change adding srf:profileIdentityHash remains to be scheduled.

3.4 Partial implementations

3.5 Compiler Limitations

Entailment regimes recorded but not acted on — Deferred

Current: Compiler always reads asserted triples. Correct for srf:NoEntailment, gap for others.

Implementation: tools/surface/compile.py::check_entailment_regime() now refuses to compile under anything but NoEntailment, closing the silent gap. Reasoner integration for other regimes still not started.

Decision needed: Choose a reasoner (owlrl for RDFS/OWL 2 RL, external EL reasoner for nominal form) and decide whether compiler materialises entailments before reading or requires caller to.

Parity cannot check DefinitionOnly forms — Deferred

Why: Parity compares assertions; DefinitionOnly surfaces assert nothing. Reports as skipped with regime named rather than silently passed.

Dependency: Same as entailment regimes item above.

Blank-node labels visible in emitted modules — ✅ Accepted Tradeoff

Design: Minted from the term they belong to (_:RoleAssignment_job-family_Sales_def_l0).

Why kept: Stable diffs for regenerated modules. Hashing no longer depends on this (canonical form relabels blank nodes by position), so this is a style choice per law srf:R3.

Canonicalisation performance — Watch item

Status: to_canonical_graph is superlinear on blank-node-bearing graphs. Read-set inputs are almost always blank-node-free (cost is sort). Large core.ttl with one blank-node per population member is the case to watch.

Action if triggered: Hash per-subject and fold rather than abandoning canonical form.

3.6 Canonicalisation contract srf-canon/1 → srf-canon/2 — ✅ Closed

Change: Switched from sorted rendering to rdflib.compare’s canonical graph. Blank-node-label independent; graph read from anywhere hashes stably.

Impact: All previously recorded hashes are stale by design. Three committed example packages regenerated per ontology/surface/execution/README.md.

Status: Examples and fixtures declare srf-canon/2 and generator 0.2.0. Runbook for future cutovers tracked under ADR-A27.

3.7 Surface-to-MORK lowering engine (ADR-A18) — ✅ Verified

Implementation: tools/surface/lowering.py implements deterministic lowering:

• lower_projection — for ProjectionContract (no direct-emit form)
• lower_contract — for Promotion/Index (when deployment mirrors to MORK)
• lower_all/link_dependencies — one mapping graph per batch with dependency edges

Verification: python -m unittest surface.test_surface -v passes 61/61 including:

• ContractLoweringTests
• LoweringDependencyTests
• Projection lowering
• Determinism checks
• MORK interop

Note on goldens: No committed lowering output; regenerate with python -m surface lower --contracts ontology/surface/examples/saas-subscription-arr-projection.ttl and commit result under ontology/surface/execution/.

4. Findings 15.1–15.5

4.1 elg:boundScheme did not exist — ✅ Closed

Fix: Implemented as option (c), elg:constrainedByContract.

Location: ontology/eligibility/spec/eligibility.ttl, ontology/eligibility/README.md, ontology/eligibility/shapes/rules.ttl.

4.2 elg:HierarchicalClosureRule unscoped — ✅ Closed

Fix: Removed from ontology/eligibility/shapes/ per eligibility-L9-replacement.md.

Status: No remaining reference in Eligibility layer.

4.3 ins: vs ino: properties — ✅ Closed

Status: Dropped; insure-o is being phased out. CSO scopes peril via hasPerilScope on TermApplication.

4.4 fnd:GovernanceState individuals undeclared — ✅ Closed

Previous status (wrong): Marked “Done” — actually was a real gap.

Fix: Now declared in ontology/foundation/vocab/foundation-vocab.ttl as part of Phase 4 MORK governance work (needed them to be real).

4.5 bhv:targetsAllowance README⇄spec drift — ✅ Closed

Fix: Hand-remediated. tools/lattice_tooling/literate_extract.py --check catches future drifts.

5. MORK Toolchain Join Assumptions — ⏳ Decision Needed

tools/surface/mork.py and tools/mork2rml.py are written against this repository’s ontologies. Before production, confirm these assumptions against a live toolchain checkout:

Open question: Whether to reuse MORK’s precedes derivation ordering for Surface’s generated-module order (class definitions → memberships → closure relations) instead of owl:imports. Currently uses owl:imports (surface consumed as unit). If toolchain expects precedes, add to manifest.

Document Status

This document is a reference for items deferred, in-progress, or requiring decisions. For phase-level status and verification checklist, see surface-mork-unified-projection.md.

Last updated: 2026-09-22
Next review: Upon SWRL verification completion (Phase 8 blocker) or Phase 9 authorization