Reference MCP Submit Correction

submit_correction

Submit a scholarly correction of something the agent said.

Arguments

Argument	Type	Required	Description
interaction_id	integer	yes	From ToolResponse.interaction_id of the response being corrected
claim	string	yes	The specific claim the agent made that is wrong
correction	string	yes	What the correct interpretation is
evidence	string	no	Scholarly citation supporting the correction
scholar_id	integer	no	Links to scholars table for attribution

The interaction_id is the integer in the interaction_id field of any ToolResponse. Every response from semantic_search, summarize_artifact, and interpret_token includes one.

What happens

1. A new row is created in annotation_runs with source_name='user-correction' and method='agent-hypothesis-correction'. If scholar_id is provided, the run is attributed to that scholar.

2. A row is created in interaction_feedback recording the full correction payload: the original claim, the correction, and the evidence.

3. Any cached agent_outputs that cited the corrected annotation are marked for regeneration. The next call to summarize_artifact for the same P-number will regenerate rather than return the cached result.

4. The correction is returned with its assigned IDs.

Permanence and attribution

Corrections are permanent. They do not delete or overwrite existing data. The original interpretation remains in the database with its provenance intact. The correction layers on top as a new annotation run.

If scholar_id is provided, the correction is attributed in the annotation_runs.scholar_id column. Corrections are discoverable via the annotation run history.

Example

# After interpret_token returns an incorrect reading:
submit_correction(
    interaction_id=44,
    claim="The agent read lu₂-gal as lugal[king]N",
    correction="This sign sequence in this context should be read as lugal[owner]N — the administrative genre and parallel texts from Nippur support the 'owner' sense",
    evidence="Steinkeller 1987, FAOS 9/2, p. 62",
    scholar_id=1234
)

Response

{
  "interaction_id": 44,
  "feedback_id": 17,
  "new_annotation_run_id": 88,
  "message": "Correction recorded. The next request that re-loads any agent output citing the corrected annotation will regenerate."
}