Semantic Bridge Evidence¶

Part IV — Semantic bridge evidence¶

8. Mapping routes and structural distinctions¶

8.1 Route framework¶

The final route framework is:

Route	Evidence-based description
1	Same meaning and compatible graph shape; eligible for reviewed direct alignment.
2	Meaning can be preserved only through conditional or structural transformation.
3	Publishable fact without a first-class target; use `schema:PropertyValue`.
4	Inapplicable, private, B2B-only or deliberately excluded fact.

8.2 Direct correspondence versus structural transform¶

The central technical distinction is whether the source and target assertions have the same graph shape.

A direct property equivalence can entail the same relationship under another property. It cannot:

convert an object record into a string;
inspect one property’s literal value to select another output property;
create a new PropertyValue node with several fields;
aggregate or format several values into one output.

Those operations require a rule or transform.

8.3 Allergen correction¶

The real Web Vocabulary v1.16 source described gs1:allergenStatement as a datatype property with range rdf:langString and a definition indicating one textual string. The real GDSN allergen structure is a two-hop coded record. Therefore:

the structured chain is not soundly equivalent to gs1:allergenStatement;
a rule may derive a text statement from the structured chain if governance approves the vocabulary lookup and formatting;
the LOOM output identified a separate flat GDSN attribute, gdsn:a-1463136308, as a direct lexical candidate for gs1:allergenStatement;
the flat statement and structured allergen records are related but distinct source representations.

8.4 Nutrient mapping¶

The source baseline mapped four nutrient codes to schema.org nutrition properties:

GDSN code	Target property
`ENERC`	`schema:calories`
`FAT`	`schema:fatContent`
`PRO-`	`schema:proteinContent`
`NA`	`schema:sodiumContent`

The mapping is conditional on the code in a structured record and is therefore Route 2. Calcium (CA) had no corresponding property in the examined schema:NutritionInformation set and used Route 3 in the worked example.

8.5 GPC generic projection¶

Most tested GPC Brick facets had no Web Vocabulary candidate. The generic publication pattern is:

{
  "@type": "schema:PropertyValue",
  "schema:propertyID": "gpc:20000031",
  "schema:name": "Type of Cheese",
  "schema:value": "GOUDA"
}

This preserves the source identifier and readable label without asserting a native target property.

8.6 Publication allow-list¶

A mapping does not automatically authorise publication. Route 4 is a separate policy decision. The source evidence repeatedly distinguishes:

internal or recipient-specific GDSN detail;
public product facts;
marketplace-owned commercial facts;
inapplicable cross-category shape slots.

The bridge package therefore requires both semantic mappings and a publication allow-list.

9. SSSOM governance model¶

9.1 Why a mapping registry is required¶

The source evidence shows that mapping knowledge cannot safely remain in label-comparison scripts or hard-coded queries. A governed registry is needed to retain:

candidate source;
review status;
mapping predicate;
confidence;
cardinality;
structural assumptions;
source and target releases;
reviewer and decision date;
rejection rationale;
compilation status.

9.2 Representative SSSOM seed records¶

The following rows are based on the real LOOM output and reconciliation. They remain candidates until formally reviewed.

subject_id	predicate_id	object_id	justification	cardinality	status note
`gdsn:a-1292425203`	`skos:closeMatch`	`gs1:gtin`	`semapv:LexicalMatching`	n:1 context	Three sibling GDSN GTIN attributes also matched; packaging level requires review.
`gdsn:a812659001`	`skos:closeMatch`	`gs1:ingredientStatement`	`semapv:LexicalMatching`	1:1 candidate	Tool-corroborated; range and domain review still required.
`gpc:20002867`	`skos:closeMatch`	`gs1:sharpnessOfCheese`	`semapv:LexicalMatching`	1:1 candidate	Term correspondence corroborated; value-set equivalence remains unreviewed.
`gdsn:a-1463136308`	`skos:closeMatch`	`gs1:allergenStatement`	`semapv:LexicalMatching`	1:1 candidate	More structurally plausible direct candidate than the coded chain.
`gdsn:a2008988543`	`skos:closeMatch`	`gs1:nutrientBasisQuantity`	`semapv:LexicalMatching`	1:1 candidate	Corroborates the earlier direct-source finding.

9.3 Promotion and compilation¶

A candidate is not promoted merely by changing skos:closeMatch to owl:equivalentProperty. The review shall establish:

semantic equivalence or the intended weaker relation;
domain and range compatibility;
datatype and language compatibility;
graph-shape compatibility;
cardinality and packaging-level meaning;
target publication role;
regression examples.

Approved SSSOM records are the source of truth. OWL axioms and rules are compiled outputs.

9.4 Mapping cardinality¶

The GTIN result demonstrates why cardinality metadata matters. Four GDSN attributes matched one target term. A reviewer must determine whether the source attributes represent:

the same concept at different packaging levels;
distinct roles that should map to narrower targets;
context-specific aliases;
false lexical matches.

Collapsing them without review could publish the wrong identifier.

9.5 Rejected mappings¶

Rejected candidates should remain in the SSSOM set with rationale. This prevents repeated rediscovery and supports audits when source labels or definitions change.