Truth Slots

A truth slot records the currently accepted value or values for a governed claim within a security envelope. Use this guide when a graph value looks outdated, two sources appear to disagree, or your application needs to explain why a value is current. The complete MCP investigation uses /mcp/full: list_truth_slots is also available on Agent and Maintenance, but get_truth_slot_history is Full-only. Keep the same principal and space throughout. Replace example IDs with actual returned IDs. HTTP examples assume local SKIP_AUTH development; add a bearer token when authentication is enabled.

List truth slots

Query truth slots filtered by subject entity, predicate, or staleness.

MCP Tool
curl

{
  "subject_entity_id": "entity:550e8400-e29b-41d4-a716-446655440000",
  "predicate_id": "works_at",
  "limit": 25
}

Tool: list_truth_slots

curl -s "http://localhost:3000/v1/truth-slots?limit=25"

MCP list_truth_slots returns data.items. Each item contains the slot fields directly, plus conflict_count and has_conflict; the list response has no nested slot field. REST returns full slots in results, with total and cursor, without the MCP summary fields:

Field	Description
`slot_key`	Key for the slot type, claim identity, and security envelope
`subject_entity`	The entity this truth is about
`predicate_id`	The predicate (e.g., `works_at`, `located_in`)
`current_values`	The currently accepted value(s) with claim IDs
`stale`	Whether the slot needs re-evaluation
`conflict_count`	Claim count in the associated conflict group
`has_conflict`	Whether an unresolved conflict exists

Filter for stale slots

Stale means the slot needs recomputation after its inputs changed. It does not by itself establish that the current value is false. Set stale_only: true to find slots awaiting evaluation.

MCP Tool
curl

{
  "stale_only": true,
  "limit": 50
}

Tool: list_truth_slots

curl -s "http://localhost:3000/v1/truth-slots?stale_only=true&limit=50"

Investigate conflicts

Use get_truth_slot_history with a returned slot ID to inspect its conflict group and available assertion references. This response is an inspection view, not an exhaustive immutable history of every past decision.

MCP Tool
curl

{
  "id": "truth_slot:550e8400-e29b-41d4-a716-446655440000"
}

Tool: get_truth_slot_history

curl -s "http://localhost:3000/v1/truth-slots/$SLOT_ID/history"

Set SLOT_ID to the complete returned ID for the HTTP example. The MCP history response contains:

Field	Description
`slot`	The truth slot record
`conflict_group`	The conflict group if one exists, including all `claim_ids`
`assertions`	Assertion references the handler could resolve and authorize

REST history returns slot and an optional conflict_group; it does not return the assertions array. All reads use the authenticated space. Add a bearer token when authentication is enabled.

Interpret the result

First inspect the predicate’s semantics with get_predicate. Equal values can corroborate each other. Different values can coexist when the predicate permits multiple values, or form a conflict when it does not. Claims in different security envelopes are isolated and do not adjudicate against one another. An LLM adjudicator, when configured, can corroborate, select a winner, allow coexistence, supersede a value with validity dates, or request review. Without an adjudicator, configured deterministic policies provide the fallback. Unknown semantics do not justify automatically superseding a value.

Observation	Next question
Several current values	Does the predicate allow coexistence?
A superseded claim	Which evidence supports the replacement and its time window?
`has_conflict: true`	What values actually differ, and is adjudication pending?
`stale: true`	Did input changes enqueue recomputation, and is the worker progressing?
Empty history assertions	Are the claim references graph records rather than assertion IDs, or are reads restricted?

Use search_assertions with the known entity_id to inspect the entity’s resolved claims independently of the history helper. Follow each relevant claim to its source before presenting a winner as fact. A confidence score is part of the extraction or adjudication output, not an independent probability that the claim is true. See Truth maintenance for the slot model and Follow updates and conflicting claims for a worked investigation.

Inspect a single assertion

Drill into an individual assertion to see its evidence, resolution status, and linked entities.

MCP Tool

{
  "id": "assertion:550e8400-e29b-41d4-a716-446655440000"
}

Tool: get_assertion

Advanced: pagination

Both list_truth_slots and list_assertions support cursor-based pagination. Pass the cursor value from the previous response’s next_cursor field to fetch the next page.

{
  "predicate_id": "works_at",
  "limit": 25,
  "cursor": "550e8400-e29b-41d4-a716-446655440000"
}

Next steps

Ingesting data — Feed documents into the extraction pipeline
Building graphs — Create entities and relationships manually
Event streaming — Monitor extraction and resolution events

Getting started

Core concepts

By example

Guides

List truth slots

Filter for stale slots

Investigate conflicts

Interpret the result

Inspect a single assertion

Next steps

​List truth slots

​Filter for stale slots

​Investigate conflicts

​Interpret the result

​Inspect a single assertion

​Next steps

List truth slots

Filter for stale slots

Investigate conflicts

Interpret the result

Inspect a single assertion

Next steps