accepted 2026-05-30

Category-G Curated Walk — Execution Plan and G11 Reconciliation

Context and Problem Statement

ODR-0022 re-scoped the PDTF descriptive-layer import from a 1:1 ~935-leaf walk to category treatments (A–G); only Category G (~188 candidate names) plus the regulatory-salience allow-list receives per-leaf curation. That curation — “the curated G walk” — is repeatedly deferred to the WG, but no record says how it executes. Three concrete gaps block it:

No current execution plan. ADR-0028 did spell out a 6-step walk (datatype emission, §Q4a class-promotion, term-grain requires, definedInContext home-pass, reconciliation register, determinism) — but it was written for the old ~935-leaf flat walk and re-scoped by ODR-0022, never rewritten for the bounded Category-G curation. The ontology programme retired 2026-05-28 (ADR-0005); subsequent work lands as fresh ADRs, so this is a successor, not an amendment.
Part of the walk is silently already done. ADR-0005 item G11 (closed 2026-05-28) emitted ~17 §Q5a Quale/boolean leaves as flat datatype properties on opda:Property/opda:LegalEstate (builtForm, currentEnergyRating, councilTaxBand, ownershipType, centralHeatingFuelType, heatingType, propertyType, …). These overlap the candidate-G set, so the 188 is not a clean to-do list, and the overlap is unreconciled — the riskIndicator case already surfaced as a conflicting-domain duplicate (flat opda:Property home vs ODR-0008d opda:RiskAssessment home), fixed 2026-05-30 by re-homing it onto opda:RiskAssessment.
The walk is R2’s trigger, unsequenced. ODR-0023 R2 (the UFO-axis sub-module split) is gated on the curated G walk — only a UFO-typed leaf-set makes the Quality/Mode/Legal-estate axes countable — but nothing sequences walk → R2.

Decision Drivers

ODR-0022 §Rules (curate-per-leaf, one dereferenceable IRI per concept; reserve WG judgment for Category G + salience).
ODR-0008 §Q5a (datatype-vs-SKOS per-leaf binding), §Q6a (flat-default; hierarchy only on a named consumer query), §Q4a (class-promotion), §Q1a (reconciliation register).
ODR-0023 R2 axis framework (qualities / modes / legal-estate), ratified but spawn-gated on this walk.
ADR-0028’s totality assertion (every annotated leaf emitted-or-promoted; zero silent omissions) + the G3 ci-descriptive-roundtrip coverage gate.
The ADR-0005 §G1 principle: engineering does not silently reconcile ratified-rules disagreements — the G11∩ODR-0008d overlap is exactly such a reconciliation, and it routes through the WG.

Considered Options

Option A — a fresh successor ADR (this) that reconciles, sequences, batches, and gates the curated Category-G walk. Per-leaf modelling stays the WG’s.
Option B — amend ADR-0028 in place. Rejected: ADR-0028 is an accepted record of the retired programme; ODR-0022 explicitly permits “amend ADR-0028 or author a successor”, and a successor keeps the retired programme’s history intact.
Option C — leave the walk wholly to the WG with no plan. Rejected: leaves the G11 overlap unreconciled and the R2 trigger unsequenced; the process (reconciliation, sequencing, batching, completion gate) is engineering, separable from the per-leaf modelling judgments.

Decision Outcome

Chosen option: Option A. Five work items; the plan owns process and sequencing only — every per-leaf datatype-vs-SKOS / axis / domain / IRI decision remains the WG’s (Modelling Sub-Committee), and any ratified-rules reconciliation routes through the WG, not silent engineering.

G11 reconciliation (do first). Diff descriptive-category-binning.json (the 188 candidate-G names) against the ~17 already-emitted G11 §Q5a leaves (opda-property.ttl / opda-agent.ttl). For each overlap leaf the WG records a disposition: (a) the G11 flat emission stands (curation = confirm rdfs:domain + add dct:source schema-leaf-path (G2) + bind a SKOS range where §Q5a says so), or (b) re-home under the R2 axis split. Capture dispositions in the ODR-0008 §Q1a reconciliation register (reuse, do not invent a parallel one). The riskIndicator re-home (G11 opda:Property → ODR-0008d opda:RiskAssessment, 2026-05-30) is the first register entry and the worked template. Net remaining ≈ 188 − (overlap that stands).
Per-leaf curation procedure. For each not-yet-emitted candidate-G leaf, in order: apply §Q5a (datatype property vs SKOS-scheme range), §Q6a (flat — no rdfs:subPropertyOf unless a named query forces it), assign the UFO axis (R2: quality / mode / legal-estate), the rdfs:domain (opda:Property / opda:LegalEstate / other), the opda: local name, and the dct:source schema-leaf-path (G2). Emit via the ADR-0030 generator — no hand-edited TTL.
Batching. Curate in reviewable batches that each round-trip a form: batch 1 = the G11-overlap §Q5a Quale family (confirm/re-home — smallest, highest-certainty); subsequent batches by overlay/form (TA6, LPE1, …) so each batch flips a thin profile to enumerated under G3.
R2 sequencing. The walk produces the UFO-typed partition R2 needs; once batch 1 + a representative form-batch are typed, R2 can judge whether the axis split is “operationally load-bearing” (ODR-0008 §Q2a(a)). Fixed order: walk batches → R2 convenes → (if load-bearing) ODR-0008a/b/c spawn.
Completion gate. The G3 ci-descriptive-roundtrip gate (currently xfail) flips to PASS when every form’s leaves are enumerated + emitted, satisfying ADR-0028’s totality assertion (count(annotated leaves) == count(emitted properties) + count(class promotions); zero silent omissions). The thin-profile xfail is the live progress tracker.

Consequences

Good, because “deferred to the WG” becomes an executable, batched, gated plan with a falsifiable completion test (G3 + totality).
Good, because the G11 overlap is reconciled explicitly rather than left as latent conflicting-domain bugs (the riskIndicator duplicate was exactly that).
Good, because walk → R2 sequencing unblocks ODR-0023 R2 deterministically instead of by guess.
Bad, because the reconciliation may re-open some “done” G11 leaves (re-homing) — a one-time cost, paid to remove the silent §Q5a-vs-ODR-0008d divergence.
Neutral, because the per-leaf judgment stays WG-paced; this plan structures the work, it neither accelerates the modelling decisions nor pre-empts them.

Confirmation

The ODR-0008 §Q1a reconciliation register has a disposition for every G11∩candidate-G overlap leaf (the riskIndicator entry is present as the template).
grep confirms no candidate-G IRI is emitted ahead of its WG disposition (the generator holds Category-G IRIs unminted per ADR-0030 until curated).
G3 ci-descriptive-roundtrip flips xfail → PASS at completion; the ADR-0028 flat-default and totality CI continue to hold.

More Information

Strategy it serves: ODR-0022 §Rules.1/§Rules.6 (Category-G curated walk) — in depends-on:, not implements: (the latter is intra-corpus ADR→ADR only).
Per-leaf modelling rules: ODR-0008 §Q5a / §Q6a / §Q4a / §Q1a; ODR-0008d (the RiskAssessment home that re-claimed riskIndicator).
R2 trigger: ODR-0023 R2 (UFO-axis sub-modules, gated on this walk).
Generator + predecessor walk: ADR-0030 (the emission subsystem + gates G1–G3); ADR-0028 (the re-scoped predecessor whose home-pass / requires / totality items this plan carries forward).
Already-emitted overlap: ADR-0005 §G11 (the ~17 §Q5a leaves emitted 2026-05-28; riskIndicator re-homed 2026-05-30).
Outcome (2026-05-31): this walk executed (185/188 candidate-G emitted-or-collapsed) and was reviewed by Council session-028, ratified as ODR-0024 (the modelling rules) + ADR-0032 (the engineering + the two CI gates ci-dup-declaration / ci-category-g-coverage), and remediated to an honest 179/239 coverage. The leftover — the monetary walk (executed 2026-05-31, §G22 closed — opda:MonetaryAmount + ISO-4217 opda:CurrencyScheme + 16 per-bearer monetary properties; coverage 179→196/239), the R5-surfaced follow-on (executed 2026-05-31, §G23 closed — 40 enum/Yes-No leaves minted + the 5 R6 schemes + ownerType wired; coverage 196→236/239), and the held opda:Room leaves — is tracked in ADR-0005 §G22–G24. With G22 + G23 + G24 closed, the curated Category-G walk is COMPLETE at 239/239 — the held opda:Room/opda:Building leaves (§G24) were modelled as the anonymous by-value opda:RoomDimension structure via ADR-0033 (Council session-030; classes not minted — neither has a data-realisable IC).
R2 sequencing closed (2026-05-31): work-item-4’s sequence (walk → R2 convenes → if load-bearing, spawn ODR-0008a/b/c) ran to completion. R2 convened — Council session-029 — and judged the UFO-axis split NOT operationally load-bearing (0–6–0): the walk produced a bearer-domain (rdfs:domain) partition, not the UFO-typed one §Q2a(a) presupposed, so ODR-0008a/b/c were NOT spawned (ODR-0023 R2 updated; the gUFO typing pass that would build the typed set is deferred as ADR-0005 §G25).