synthesis.md

Synthesis API

POST /v1/synthesize produces grounded Markdown for an object, folder, collection, or root scope.

The core supplies preferred text. Synthesizers do not read .osii and must not perform their own corpus traversal.

Request

{
  "api_version": "v1",
  "request_id": "synth-88",
  "scope": {
    "scope_type": "collection",
    "scope_id": "calibration-runs",
    "documents": [{
      "file_id": "sha256-a",
      "filename": "run-a.txt",
      "text": "Calibration drift was 0.4%...",
      "metadata": {"representation": "edited"}
    }]
  },
  "expert_context": "Focus on deviations from procedure Q-17.",
  "config": {"maximum_words": 400}
}

Response

{
  "api_version": "v1",
  "request_id": "synth-88",
  "processor": {"...descriptor": "..."},
  "markdown": "## Findings\nCalibration drift remained below 0.5%...",
  "citations": [{
    "file_id": "sha256-a",
    "char_start": 0,
    "char_end": 31
  }],
  "metadata": {"input_documents": 1},
  "warnings": []
}

Claims should be tied to citations whenever possible. Character offsets refer to the exact document text sent in the request.

See examples/synthesizer.py.