Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.9dlabs.xyz/llms.txt

Use this file to discover all available pages before exploring further.

The receipt engine automatically detects contradictions between evidence spans included in a context pack. Detected conflicts appear as warnings in the receipt so your agent or operator can address them.

​
Three Rule Types

​
1. Numerical Conflicts

Detects when two spans discuss the same topic but cite different numbers. Triggers when:
  • Two spans share enough context (Jaccard similarity >= 0.3)
  • They contain different numerical values with the same unit
Example:
Span A: “The enterprise discount cap is 15%.” Span B: “The discount cap has been raised to 18%.“

​
2. Supersession

Detects when newer information explicitly replaces older information. Triggers when:
  • A span contains language like “supersedes”, “revised to”, “no longer”, “updated to”, “replaced by”
  • The spans share at least 15% token overlap
Example:
Span A: “Refunds accepted within 30 days.” Span B: “Policy revised to 14-day refund window effective March 1.”

​
3. Negation Pairs

Detects direct contradictions using opposing language. Triggers when:
  • Spans contain negation pairs: “will/will not”, “approved/rejected”, “allowed/prohibited”, “can/cannot”
  • The spans share at least 15% token overlap
Example:
Span A: “The board approved the Q3 expansion.” Span B: “The board rejected the Q3 expansion proposal.”

​
Conflicts in Receipts

Conflicts appear in the receipt’s conflicts array: 
{
  "conflicts": [
    {
      "id": "conflict-uuid",
      "claim_a": { "artifact_id": "...", "span_id": "...", "content_hash": "..." },
      "claim_b": { "artifact_id": "...", "span_id": "...", "content_hash": "..." },
      "description": "Numerical conflict: 'cap at 15%' vs 'raised to 18%'",
      "resolution": null,
      "resolved_by": null
    }
  ]
}

​
Resolving Conflicts

Use the feedback endpoint with action: "resolve_conflict" to mark which span is correct: 
client.feedback(
    action="resolve_conflict",
    artifact_id="...",
    span_id="preferred-span-id",
    actor_id="user_alice",
    workspace_id="ws_acme",
    conflict_id="conflict-uuid",
    preferred_span_id="preferred-span-id",
)
Once resolved, future context packs will prefer the selected span and the conflict will no longer appear as unresolved.