feat(ontology): publish O*NET content-model linkages

ARCHITECTURE.md

@@ -79,7 +79,7 @@ flowchart LR
 | `commitment_extraction.py` | Pluggable LLM derivation of a customer commitment (promise + deadline) from a post; `Null` default, `ContextualOrchestrator` real impl |
 | `temporal_expressions.py` | Pure Korean relative-time resolver for Global Ask (ADR 0150) |
 | `ask_time_axis.py` | Event-time vs ingestion-time clock choice for that window (ADR 0202) |
-| `ontology.py` | Loads the governed Turtle source tree (`lineageweave-kg.ttl` plus generated fragments), the formal OWL 2/RDFS/SKOS vocabulary for the Knowledge Graph's node/edge types and source taxonomies (ADR 0004, ADR 0252, ADR 0255) |
+| `ontology.py` | Loads the governed Turtle source tree (`lineageweave-kg.ttl` plus generated fragments), the formal OWL 2/RDFS/SKOS vocabulary for the Knowledge Graph's node/edge types, source taxonomies, and published O*NET linkages (ADR 0004, ADR 0252, ADR 0255, ADR 0256) |
 | `ontology_neighborhood.py` | Bounded typed ontology/provenance neighborhood (ADR 0184); PostgreSQL stays authoritative, OWL subclass is not an instance edge |
 | `ontology_source_cursor.py` | Opaque HMAC source-window continuation (ADR 0124); keyset pagination, never OFFSET |
 | `period_report.py` | Fit GRM/GPCM on persisted IRT rows, FIPC-select, EAP-score a period (ADR 0003 slice 3; Bock & Mislevy, 1982) |

CHANGELOG.md

@@ -8,6 +8,18 @@ All notable changes to this project are documented here. Format follows
 
 ### Added
 
+- O*NET occupation-rating source evidence now has a replay-safe PostgreSQL
+  contract with normalized release, source-table, scale, occupation, element,
+  and observation tables. Exact release/source LIST partitions fail closed;
+  nullable categories remain idempotent identities, divergent duplicates and
+  truncation fail closed, and no source value is promoted to a local weight or
+  person score. Task Ratings remain outside this content-model-element store
+  pending their own normalized identity contract (ADR 0257).
+- All eight O*NET 31.0 published linkage tables now contribute 1,417 directed
+  Ability/Essential Skill/Transferable Skill/Work Style relations to Work
+  Activities and Work Context. Every direct relation has an exact reified
+  PROV source-table assertion; no confidence, causal effect, rating, or weight
+  is inferred (ADR 0256).
 - The complete O*NET 31.0 Content Model Reference is now a deterministic
   3,006-concept SKOS fragment with exact source IDs, names, descriptions, and
   documented outline parents (ADR 0255). Its read model fails closed on

docs/adr/0256-onet-content-model-published-linkages.md

@@ -0,0 +1,44 @@
+# ADR 0256: O*NET content-model published linkages
+
+**Status:** Accepted  
+**Date:** 2026-08-27  
+**Extends:** ADR 0248, ADR 0255
+
+## Context
+
+O*NET 31.0 publishes eight explicit linkage tables connecting Abilities,
+Essential Skills, Transferable Skills, and Work Styles to relevant Work
+Activities and Work Context elements. These 1,417 source pairs provide an
+authoritative relationship layer without deriving relations from labels,
+embeddings, local judgment, or arbitrary weights.
+
+## Decision
+
+1. Publish all eight tables and all 1,417 unique directed pairs. Preserve the
+   source and target O*NET element IDs and names exactly and fail generation
+   when either endpoint is absent from the pinned Content Model Reference.
+2. Assert `:relevantWorkActivity` or `:relevantWorkContext` from the worker
+   element to the job element. Direction is preserved; the predicates are not
+   symmetric, causal, equivalent, or psychometric weights.
+3. Reify every direct assertion as an `rdf:Statement` and `prov:Entity` with
+   exact `rdf:subject`, `rdf:predicate`, `rdf:object`, and
+   `prov:wasDerivedFrom` pointing to the one pinned O*NET table that published
+   it. No unsupported confidence or generated time is added.
+4. Pin each official O*NET 31.0 JSON artifact and reproduce one deterministic
+   Turtle fragment. A duplicate pair inside a source table, endpoint-name
+   mismatch, unknown endpoint, wrong domain, or wrong row count fails closed.
+5. Keep occupation-specific ratings separate. This decision imports no
+   occupation code, data value, scale, sample size, error, confidence interval,
+   suppression flag, person trait, or locally normalized score.
+
+## Consequences
+
+The semantic layer can traverse published cognition/style-to-behavior/context
+relations with assertion-level provenance. Occupation ratings remain the next
+separate import contract because their scales, uncertainty, suppression, and
+update dates cannot be represented as these unweighted vocabulary links.
+
+## Reference
+
+National Center for O*NET Development. (2026). *O*NET 31.0 database: Content
+model linkages* [Data set]. https://www.onetcenter.org/dictionary/31.0/json/

docs/adr/0257-onet-occupation-rating-observation-store.md

@@ -0,0 +1,80 @@
+# ADR 0257: O*NET occupation-rating observation store
+
+**Status:** Accepted  
+**Date:** 2026-08-27  
+**Extends:** ADR 0166, ADR 0255, ADR 0256
+
+## Context
+
+O*NET 31.0 publishes occupation-specific ratings for content-model abilities, essential and
+transferable skills, knowledge, education, training and experience, interests,
+work styles, work activities, work context, and adjacent content-model
+domains. A rating is not merely an edge: its meaning depends on the release,
+source table, occupation, element, scale, optional response category, sample
+size, standard error, 95% confidence interval, precision-suppression flag,
+relevance flag, source update month, and domain source. O*NET publishes that
+field as a seven-character `MM/YYYY` value; coercing it to a database date
+would invent a day (National Center for O*NET Development, 2026b).
+
+Flattening those attributes into ontology predicates would erase measurement
+and provenance boundaries. Loading them into Python mathematical code would
+also violate LineageWeave's externalized-compute boundary.
+
+## Decision
+
+1. Store source releases, source tables, scales, occupations, content-model
+   elements, and rating observations in separate third-normal-form tables.
+2. Preserve published numeric values exactly as decimal observations. They are
+   source ratings, never locally estimated weights, person scores, causal
+   effects, or calibrated psychometric parameters.
+3. An observation key is release + source table + occupation + element + scale
+   + optional category. PostgreSQL `UNIQUE NULLS NOT DISTINCT` keeps an absent
+   category an honest single absence rather than replacing it with a sentinel.
+4. Partition observations first by release and then by source-table code.
+   These are authoritative lifecycle/query boundaries and require no invented
+   hash modulus. An importer must create both exact LIST partitions before
+   inserting; without them PostgreSQL rejects the row.
+5. Every insert uses the owning release/table artifact digest and idempotent
+   `ON CONFLICT DO NOTHING`. An exact duplicate is idempotent, while a row with
+   the same identity and different source values fails closed. Endpoint names
+   and scale names must match their normalized reference
+   rows; each scale definition names its owning source-table artifact, and the
+   importer rejects disagreement rather than overwriting identity.
+6. Preserve `recommend_suppress` and `not_relevant` independently. A suppressed
+   value remains stored with its warning; a not-relevant value is not converted
+   to zero. Missing `n`, error, or interval values remain null.
+7. Range and uncertainty constraints reject negative sample/error values,
+   inverted confidence intervals, malformed or future source update months, malformed source
+   digests, and values outside their declared scale bounds before persistence. Scale bounds
+   govern the published `Data Value`, not the optional response-category code. In particular,
+   O*NET's `CXP` rows store a category in `Category` while `Data Value` is the percentage that
+   endorsed it, so the authoritative `CXP` bounds are 0 through 100 (National Center for
+   O*NET Development, 2026b).
+8. This content-model-element store excludes Task Ratings, whose integer Task
+   IDs and task-statement identity require a separate normalized target table;
+   it does not reinterpret a Task ID as a content-model element.
+9. This store is immutable source evidence. Row mutation and whole-store
+   truncation fail closed. Any later aggregation, comparison,
+   temporal model, multilevel model, or occupational recommendation belongs to
+   TEPP/fast-mlsirm or another owning Rust service and must cite these rows.
+
+## Consequences
+
+LineageWeave can import the governed public O*NET content-model rating corpus without
+manufacturing semantics or embedding large production datasets in git.
+Release/source partitions localize hot imports and permit exact detach/archive
+operations. A separate API/UI ADR is still required before exposing ratings.
+
+## References
+
+National Center for O*NET Development. (2026a). *O*NET 31.0 database* [Data
+set]. https://www.onetcenter.org/database.html
+
+National Center for O*NET Development. (2026b). *Work context: O*NET 31.0 data
+dictionary*. https://www.onetcenter.org/dictionary/31.0/excel/work_context.html
+
+PostgreSQL Global Development Group. (2026). *PostgreSQL 18 documentation:
+Table partitioning*. https://www.postgresql.org/docs/current/ddl-partitioning.html
+
+PostgreSQL Global Development Group. (2026). *PostgreSQL 18 documentation:
+CREATE TABLE*. https://www.postgresql.org/docs/current/sql-createtable.html

docs/adr/README.md

@@ -32,6 +32,8 @@ decision from them.
 | [`OCCUPATIONAL_CONSTRUCT_REFERENCES.md`](../doctoring/OCCUPATIONAL_CONSTRUCT_REFERENCES.md) | [0248](0248-occupational-construct-evidence-boundary.md) |
 | [`ONET_31_CONTENT_MODEL_REFERENCES.md`](../doctoring/ONET_31_CONTENT_MODEL_REFERENCES.md) | [0255](0255-onet-31-content-model-ontology.md) |
 | [`SOC_2018_HIERARCHY_REFERENCES.md`](../doctoring/SOC_2018_HIERARCHY_REFERENCES.md) | [0252](0252-complete-2018-soc-hierarchy.md) |
+| [`ONET_31_LINKAGE_REFERENCES.md`](../doctoring/ONET_31_LINKAGE_REFERENCES.md) | [0256](0256-onet-content-model-published-linkages.md) |
+| [`ONET_RATING_STORE_REFERENCES.md`](../doctoring/ONET_RATING_STORE_REFERENCES.md) | [0257](0257-onet-occupation-rating-observation-store.md) |
 
 [0011](0011-prov-o-standard-relations.md) and [0065](0065-prov-o-provenance-boundary.md) cite the dated W3C PROV-O and PROV-DM Recommendations (https://www.w3.org/TR/2013/REC-prov-o-20130430/ and https://www.w3.org/TR/2013/REC-prov-dm-20130430/).
 

docs/doctoring/ONET_31_LINKAGE_REFERENCES.md

@@ -0,0 +1,25 @@
+# O*NET 31.0 content-model linkage evidence
+
+This supporting note records the eight official JSON artifacts governed by
+ADR 0256. It introduces no independent architecture decision.
+
+| Table | Rows | SHA-256 |
+|---|---:|---|
+| Abilities to Work Activities | 381 | `e03c3db1dbe8c7818198c943c8640cc2ddaf351972a79756b535940c84acfc52` |
+| Abilities to Work Context | 139 | `7d5fbdc3df973790f1f24f98ace1f77c6cc8dab97197444e9ebf7695b6f59348` |
+| Essential Skills to Work Activities | 110 | `35d389bb2a578d64bf49ef263f24cb5c1bb59c214bcec9045e476008023205f5` |
+| Essential Skills to Work Context | 39 | `883f99fe723610530666031da9a68d72790f04cc87e0ac808f4600f1e01b8e5f` |
+| Transferable Skills to Work Activities | 122 | `0b1755becb3100d4c19660cea2adf8d7c47a0654087560ceffd71c515fe94be2` |
+| Transferable Skills to Work Context | 57 | `4add628b7257f2a1b7b204b1ddd715b82685559d35b020e540b15ca74abb2dd6` |
+| Work Styles to Work Activities | 303 | `889071d37a3bb073854c72dc31f96a8e9079a18c4e581e6f5bf2e082f58c641b` |
+| Work Styles to Work Context | 266 | `9e3aa088aa55d1f927688c55bb77987f5825ba0fdc9da3edebcb81cbf198fa54` |
+
+The 1,417 rows contain identifiers and names only. They publish relevance
+linkages, not coefficients, ratings, causal effects, or person scores. The
+checked-in Turtle SHA-256 is
+`6caff28f75cf6bd0a78548e85d4cd5217a77bd747c094e62f59b9cd389d36489`.
+
+## APA 7 reference
+
+National Center for O*NET Development. (2026). *O*NET 31.0 database: Content
+model linkages* [Data set]. https://www.onetcenter.org/dictionary/31.0/json/

docs/doctoring/ONET_RATING_STORE_REFERENCES.md

@@ -0,0 +1,26 @@
+# O*NET occupation-rating store evidence
+
+This supporting note records the source and database capabilities governed by
+ADR 0257. It introduces no independent architecture decision.
+
+O*NET 31.0 occupation data tables publish occupation and element identifiers,
+scale identifiers and names, decimal values, optional category values, sample
+sizes, standard errors, confidence bounds, suppression/relevance flags, update
+dates, and domain sources. These fields remain source observations; they are
+not locally estimated psychometric weights.
+
+PostgreSQL declarative partitioning provides exact LIST boundaries and
+partition pruning. A unique constraint on a partitioned table includes its
+partition key; `UNIQUE NULLS NOT DISTINCT` makes a missing category one stable
+identity without a sentinel value.
+
+## APA 7 references
+
+National Center for O*NET Development. (2026). *O*NET 31.0 database* [Data
+set]. https://www.onetcenter.org/database.html
+
+PostgreSQL Global Development Group. (2026). *PostgreSQL 18 documentation:
+Table partitioning*. https://www.postgresql.org/docs/current/ddl-partitioning.html
+
+PostgreSQL Global Development Group. (2026). *PostgreSQL 18 documentation:
+CREATE TABLE*. https://www.postgresql.org/docs/current/sql-createtable.html