Skip to content

Business Index Semantic Object Model

This document defines the narrow product model used by Business Index. It is an implementation of current SOSA concerns, not a replacement for the SOSA, SOM, or SI specifications.

First object families

Type Product meaning
client An organization or person receiving work from the business
project A bounded body of work associated with a client or internal goal
decision A choice that changes or constrains subsequent work
commitment An explicit promise, obligation, deadline, or agreed deliverable
problem An issue, obstacle, defect, or business condition requiring attention

Each durable object has an opaque identity independent of files and models. Its meaning changes through immutable versions. A version records its label, meaning statement, structured attributes, confidence, epistemic state, occurrence or validity time when known, system-learned time, extraction rule version, and supporting evidence.

Evidence

An evidence span belongs to an immutable source content version and records:

  • source asset and content hash;
  • character start and end offsets;
  • the preserved passage and passage hash;
  • a structural locator when available;
  • extraction run, method, model, and rule version;
  • confidence, evidence strength, and review state.

A generated explanation is never stored as evidence.

Identity and aliases

Canonical keys and aliases are normalized only for matching. Display values are preserved. Identity resolution records which labels and aliases were compared, the outcome, confidence, extraction run, and supporting evidence.

Client identity may use a single normalized canonical or evidenced alias match, including a versioned initialism derived from an evidenced corporate name. Other object families require both matching identity text and an existing relationship to a resolved neighbour. A context-free collision is retained as a duplicate candidate for review.

Merges redirect one durable object to another and create a feedback event. Splits create a new durable object, move only explicitly selected aliases or versions, and create a feedback event. Neither operation deletes the earlier record.

Relationships

Relationships are directional, versioned, evidence-backed, and independently reviewable. The initial product vocabulary is:

  • client_has_project
  • project_has_decision
  • project_has_commitment
  • project_has_problem
  • client_requested_change
  • decision_affects_project
  • decision_created_commitment
  • commitment_supersedes_commitment
  • decision_supersedes_decision
  • problem_resolved_by_decision
  • related_to

related_to is permitted only as an unreviewed inference and ranks below more specific connections.

Epistemic and lifecycle states

Epistemic state is one of observed, inferred, confirmed, rejected, or corrected. Model extraction produces inferred; direct source-derived asset facts may be observed; only user action produces confirmed, rejected, or corrected states.

Lifecycle state is one of active, superseded, merged, split, obsolete, or rejected. Lifecycle changes preserve prior versions and their evidence.

Semantic representations

Lexical text and embeddings are representations of an object version. They record provider, model, dimensions, representation version, and creation time. They may be rebuilt or removed without modifying the object, version, evidence, relationship, or feedback records.