Assertions and claims

An assertion is a statement about an entity, value, relationship or classification. Semogram retains its subject, predicate, value/object, confidence, validity, provenance and review state. The governed claim system adds precise evidence links, revisions, decisions, verification and derivations.

A definition says “Order has a delivered property.” An assertion says “this particular Order was delivered.” They are different records. A proposed statement is not a verified fact.

Statement types and origins

The origin is also recorded: source observation, extracted claim, human assertion or derived assessment. Assertion type describes the statement's shape; origin describes how it was produced. Model output and human testimony are not interchangeable evidence.

Create a human assertion

You need a Semogram account, access to the project and permission for the assertion workflow. In project Assertions → New assertion, use this bounded example only after a reviewer has actually checked the delivery. The subject is a real entity ID; the predicate is the exact model property from the operations model.

Inspect the saved assertion ID and Proposed status. Review it separately before approving. The form does not need an ontology-store plugin to persist this human assertion.

Human assertion versus captured evidence

The current basic creation form submits a human assertion. Its source/citation/free-form evidence fields are not a substitute for a verified capture-and-selector link; the current create path does not preserve those fields as governed evidence links. Do not rely on filling them to prove a receipt has been preserved. Inspect claim details and actual linked captures.

For pipeline-extracted claims, the assertion materializer requires evidenceLinks pointing to real captures and precise selectors. Use the supported evidence collection/upload workflow first; do not fabricate capture IDs or hashes. Evidence and verification explains that contract and its limits.

Confidence has several meanings

All scores are bounded between 0 and 1; unknown values remain null. Forecast probability requires a separate claim confidence in the governed contract. A score of 0.9 is neither an approval decision nor a measured 90% system accuracy.

Inspect through an assistant

A connected assistant can inspect existing governed claims using claim_list and claim_get; these require query read permission. They are not generic assertion-create/review tools. Human review stays in its supported workflow.

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "claim_get",
    "arguments": {
      "projectId": "<PROJECT_UUID>",
      "claimId": "<ASSERTION_UUID>"
    }
  }
}

The claim detail includes evidence, history, revisions, derivations and verification records. Inspect those records alongside status. Creating a claim does not materialize an ontology fact or rewrite a source table; queries use it only through a configured asserted binding.