Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.mira-app.dev/llms.txt

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

Eval results can be exported from the dashboard for sharing with stakeholders, feeding into BI tools, or archiving outside MIRA.

​
Export formats

FormatBest for
CSVSpreadsheet analysis, reporting, pivot tables
JSONProgrammatic processing, long-term archiving

​
Exporting results

  1. Open the Eval Studio (Flask icon → Dashboard).
  2. Click a conversation card to open the run detail view.
  3. Click Export → CSV or Export → JSON.
  4. Choose a save location.

​
CSV schema

Each row represents one eval result for a captured agent response:
ColumnDescription
conversation_idIdentifier for the captured agent response
captured_atISO 8601 timestamp of when the response was captured
eval_nameName of the eval definition
eval_typerule, llm_judge, similarity, or metric
inputThe prompt sent to the engine
outputThe engine’s response
scoreNumeric score (0–1 for llm_judge/similarity; 1/0 for rule; raw value for metric)
passtrue or false
reasoningLLM judge reasoning text, if applicable
human_override_scoreHuman reviewer score (0–1), if submitted
human_override_noteHuman reviewer comment, if submitted

​
JSON schema

{
  "exportedAt": "2024-03-15T10:30:00Z",
  "results": [
    {
      "conversationId": "conv_abc123",
      "capturedAt": "2024-03-15T10:25:00Z",
      "evalName": "Extract parties keyword check",
      "evalType": "rule",
      "input": "List all parties named in the contract.",
      "output": "The parties are Acme Corp and Beta LLC.",
      "score": 1,
      "pass": true,
      "reasoning": null,
      "humanOverride": null
    },
    {
      "conversationId": "conv_abc123",
      "capturedAt": "2024-03-15T10:25:00Z",
      "evalName": "Summary quality judge",
      "evalType": "llm_judge",
      "input": "List all parties named in the contract.",
      "output": "The parties are Acme Corp and Beta LLC.",
      "score": 0.88,
      "pass": true,
      "reasoning": "The response accurately identifies both parties with correct legal names.",
      "humanOverride": {
        "score": 0.9,
        "reviewerNote": "Looks good, approved.",
        "reviewedAt": "2024-03-15T11:00:00Z"
      }
    }
  ]
}
Edit this page — Open a pull request