SIE.4 - Judge Cross-Source Correspondences and Their Permitted Uses

Type: Method pattern Status: Eternal alpha Normativity: Normative method guidance within SIE; examples are constructed and non-normative.

Primary working result: a QualifiedCorrespondenceSet@Use whose rows identify exact source-local endpoints, relation or incompatibility, orientation, bounded use, permitted loss, justification and evidence, source versions, counterexamples, confidence where meaningful, and accepted, rejected, unresolved, or incompatible disposition.

SIE.4:1 - Problem Frame

Use this when two or more governed sources contain concepts, types, relations, fields, codes, model elements, or claims that appear to match and a receiving use needs their relation made explicit. Typical triggers are a shared label, a crosswalk, an automated matcher result, a spreadsheet of “equivalences”, or a model-to-model mapping whose semantic premise has never been judged.

The primary EntityOfConcern is one use-qualified cross-source correspondence row. The first move is to recover both endpoint senses and ask which direct relation, difference, or incompatibility is actually supported. The first result is a set that preserves accepted, rejected, unresolved, and incompatible rows rather than forcing every candidate into an equivalence.

The payoff is a defensible semantic premise for identity work, claim composition, executable mappings, and validation. Do not use SIE.4 to construct a missing semantic model, decide cross-source identity, define transformation code, or choose the authoritative value. A local relation inside one governed source remains with its source owner unless the cross-source use makes it an integration premise.

SIE.4:2 - Problem

Correspondence candidates are cheap. Labels can be normalized, lexical similarity scored, hierarchies compared, and model structures aligned. Yet “similar”, “related”, “narrower”, “same field name”, and “safe to substitute here” answer different questions. A row can be a true relation and still lose distinctions that the receiving use requires.

When candidate generation and qualification collapse, downstream mapping code inherits an unstated ontology and loss policy. Counterexamples are discarded as data-quality defects, source versions disappear, and a reviewer cannot tell whether the relation itself or only its use is unsupported.

SIE.4:3 - Forces

Force	Tension
Automation	Matchers can surface many candidates, while their scores do not establish relation truth or permitted use.
Reuse	A general crosswalk is attractive, while relation adequacy can change with receiver, grain, interval, and accepted loss.
Precision	Exact relation kinds prevent overclaim, while sources may not support a complete formal characterization.
Coverage	Pressure favors filling every row, while unresolved and incompatible rows can be the most decision-useful result.
Direction	Transformation needs an orientation, while many conceptual relations are not symmetric or safely reversible.
Versioning	Stable mapping identifiers aid reuse, while endpoint meaning can change across source editions and profiles.

SIE.4:4 - Solution

Separate candidate discovery, direct relation judgment, and bounded-use qualification. For each row, identify exact source-local endpoints, state the relation or incompatibility and orientation, justify it with evidence and counterexamples, then decide whether that relation may support the named use under a stated loss boundary. Keep the direct relation and its use qualification independently inspectable.

SIE.4:4.1 - Pattern-Use Unfolding

Take the question from the contract. State the receiving answer and why this endpoint relation can change it.
Bind exact endpoints. Reference the SIE.2 source rows, schemes, definitions, editions, profiles, effectivity, units, codes, and local extensions. Do not map labels without their senses.
Generate candidates without promotion. Lexical, structural, instance, expert, historical, or matcher evidence may nominate a row. Record the candidate source and score where useful, but do not treat nomination as acceptance.
State the direct relation or difference. Use the narrowest warranted relation vocabulary. State direction and whether the inverse, symmetry, or transitivity is actually supported. If no positive relation is established, choose rejected, unresolved, or incompatible with reason.
Test examples and counterexamples. Include instances or cases that should satisfy the relation and cases that distinguish endpoints, including version, unit, code, granularity, lifecycle, and authority differences.
Qualify the receiving use. State which substitution, comparison, join, navigation, or transformation the row permits, what semantic loss it introduces, and which uses remain forbidden.
Preserve evidence and authority. Record justification, evidence source, reviewer or domain authority where required, uncertainty or confidence, and the owner of any unresolved domain judgment.
Assign a row disposition. Use accepted-for-use, accepted-with-loss, rejected, unresolved, or incompatible. A row can state a true broader/narrower relation yet be rejected for this use.
State dependencies and tests. Name whether SIE.5, SIE.6, or SIE.7 consumes the row, plus a test or observation that reopens it.

SIE.4:4.2 - Record the Result

Correspondence position	Required content
row identity and use	stable row identifier, receiver/use, consuming result
source endpoint	source, scheme/model, exact sense, edition/profile/effectivity, identifier
target endpoint	source, scheme/model, exact sense, edition/profile/effectivity, identifier
direct relation	relation or explicit difference/incompatibility, orientation, inverse/symmetry/transitivity qualification
use qualification	permitted operation, accepted loss, forbidden reuse, conditions
warrant	candidate origin, justification, evidence, authority, uncertainty/confidence where meaningful
challenge	positive examples, counterexamples, negative/unlike cases
disposition and continuation	accepted-for-use, accepted-with-loss, rejected, unresolved, incompatible; downstream reference and reopen condition

SIE.4:4.3 - What Changes in Practice

The team stops treating a mapping spreadsheet as a bag of equalities. A matcher result becomes a candidate; a domain judgment becomes a relation warrant; and a receiving-use decision states what the relation may safely support. Unresolved and incompatible rows remain visible to the interface and tests.

SIE.4:5 - Archetypal Grounding - Product Feature and Inspection Characteristic

In the AP242/QIF application, the lexical candidate feature ↔ characteristic is too broad. The source inventory distinguishes an AP242 product-definition feature, a QIF inspection characteristic, the plan relation that applies a characteristic to a feature, and a measured result concerning that characteristic.

Candidate row	Direct relation judgment	Use qualification and disposition
AP242 feature `F-17` ↔ QIF characteristic `C-17`	not identity and not class equivalence; the QIF plan asserts that `C-17` concerns the identified product feature under configuration/effectivity `R/E`	accepted-for-use as a directed `inspection-characteristic-concerns-feature` relation only after `SIE.5` qualifies the feature endpoint; no reverse substitution
AP242 feature type `hole` ↔ QIF characteristic type `diameter`	a diameter characteristic can characterize some holes, but the types are neither equal nor simply broader/narrower	accepted-for-use for navigation from a selected plan row with a qualified feature endpoint to its linked hole feature and diameter characteristic. Return their separate source identifiers and types with the plan relation; forbidden as a global type crosswalk
AP242 feature `F-22` ↔ QIF characteristic `C-99`	the candidate arose from a shared local label; effectivity and plan evidence do not support the relation	rejected; the interface returns unmatched rather than joining
legacy feature class ↔ current QIF characteristic class	one local extension lacks an authoritative definition	unresolved and stopped for dependent mapping; return to the extension owner

The set records endpoint editions and counterexamples such as a non-dimensional inspection characteristic and a feature absent at effectivity E. It establishes semantic premises; it does not decide that identifiers denote the same entity or specify executable joins.

SIE.4:6 - Bias-Annotation

Lens	Likely drift	Repair
Governance	A matcher or integrator silently becomes the domain relation authority.	Record the warrant, decision scope, and required domain authority for each accepted row.
Architecture	Available mapping technology limits the relation vocabulary.	State the semantic relation first; choose its carrier later.
Ontology/Epistemology	Similarity, correlation, broader/narrower relation, and equivalence collapse.	Use the narrowest supported relation and preserve uncertainty and counterexamples.
Pragmatics	Every possible endpoint pair is reviewed.	Restrict rows to relations that can change the contract or a dependent result.
Didactics	Direction arrows are read as a mandatory processing order.	Explain orientation as relation or transformation meaning, not lifecycle sequence.

SIE.4:7 - Conformance Checklist

Every row serves a named receiving use and consuming result.
Both endpoints reference exact source-local senses and applicable editions/profiles.
Candidate generation is distinguishable from relation judgment.
The direct relation, difference, or incompatibility uses the narrowest warranted kind and states orientation.
Inverse, symmetry, and transitivity are not inferred without support.
Positive examples and discriminating counterexamples are present where load-bearing.
Permitted operation, semantic loss, conditions, and forbidden reuse are explicit.
Evidence, uncertainty/confidence, and decision authority are inspectable.
Rejected, unresolved, and incompatible rows remain available to downstream interfaces and tests.
The set claims neither cross-source identity nor executable transformation by implication.

SIE.4:8 - Common Anti-Patterns and How to Avoid Them

Anti-pattern	Repair
Same label, therefore same meaning	Recover endpoint senses and test counterexamples.
Matcher score as truth	Treat the score as candidate evidence; obtain a relation warrant and use decision.
Every relation is `sameAs`	Distinguish identity, equivalence, broader/narrower, related, concerns, transforms-to, difference, and incompatibility.
One global crosswalk	Qualify rows by use, source editions, loss, and forbidden reuse.
Symmetric spreadsheet mapping	State orientation and test whether the inverse is supported.
Delete failed rows	Retain rejected, unresolved, and incompatible dispositions for receiver and validation behavior.

SIE.4:9 - Consequences

The correspondence set gives mapping and interface work explicit semantic premises and preserves honest negative results. It makes version and loss dependencies visible and allows a narrow use to proceed while other rows remain unresolved.

The cost is row-level judgment and counterexample work. Automated matchers remain useful for candidate discovery but cannot close the relation or its receiving-use qualification alone.

SIE.4:10 - Rationale

A direct relation and permission to use that relation answer different questions. Keeping them separate allows a true relation to be rejected for a loss-sensitive use and a qualified narrow relation to support one operation without being promoted globally. This is the smallest structure that protects both semantic truth and practical action.

SIE.4:11 - SoTA-Echoing

The best-known line combines explicit relation vocabularies, mapping metadata, alignment evaluation, and direct Bridge truth. The serious default is a two-column crosswalk plus confidence score. Its defect is that it hides endpoint senses, relation algebra, bounded-use loss, and counterexamples. SIE.4 adapts current mapping practice into a set whose negative dispositions and use qualification are first-class.

Source line	Adopt, adapt, or reject	Role and limit
Current FPF `F.9`	adopt	Governs each direct cross-context Bridge and separate bounded-use claim; a package row must not replace Bridge truth.
SKOS Reference	adapt	Supplies distinct mapping relation forms and cautions through their semantics; SKOS does not prove a candidate relation or authorize substitution.
SSSOM 1.0	adapt	Contributes inspectable subjects, predicates, objects, provenance, justification, confidence, and source versions; metadata completeness is not relation truth.
OAEI 2025 results	adapt	Demonstrate task- and track-dependent alignment evaluation; benchmark performance does not decide a local row.
label-only or score-only crosswalk	reject as sufficient	It can nominate candidates but cannot establish endpoint senses, relation, permitted loss, or authority.

Reopen when an endpoint meaning or edition changes, a counterexample defeats the relation or use qualification, the receiver’s loss boundary changes, or a better relation vocabulary materially changes action.

SIE.4:12 - Relations

SIE.2 supplies exact endpoint senses, editions, profiles, authority limits, and model gaps.
SIE.3 is a conditional external return when available models cannot express a required endpoint distinction.
SIE.5 consumes rows that expose a load-bearing cross-source identity question.
SIE.6 uses correspondences to decide which source claims are comparable or composable without erasing scope or conflict.
SIE.7 consumes only accepted rows under their stated use and loss conditions; it adds executable transformation semantics.
SIE.10 tests correspondence truth and use qualification independently from mapping execution.