> ## Documentation Index
> Fetch the complete documentation index at: https://docs.mutual-dissent.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Transcript Schema

> JSON transcript format — full field reference with examples

Transcripts are saved as JSON to `~/.mutual-dissent/transcripts/YYYY-MM-DD_shortid.json`.

## Top-Level Fields

| Field            | Type                  | Description                            |
| ---------------- | --------------------- | -------------------------------------- |
| `transcript_id`  | string (UUID4)        | Unique debate identifier               |
| `query`          | string                | Original user query                    |
| `panel`          | string\[]             | Model aliases that participated        |
| `synthesizer_id` | string                | Model alias used for synthesis         |
| `max_rounds`     | integer               | Configured reflection round count      |
| `rounds`         | DebateRound\[]        | All debate rounds in order             |
| `synthesis`      | ModelResponse \| null | Final synthesized response             |
| `created_at`     | string (ISO 8601)     | Debate start time (UTC)                |
| `metadata`       | object                | Stats, routing config, experiment data |

## DebateRound

| Field          | Type             | Description                                   |
| -------------- | ---------------- | --------------------------------------------- |
| `round_number` | integer          | 0 = initial, 1+ = reflection                  |
| `round_type`   | string           | `"initial"`, `"reflection"`, or `"synthesis"` |
| `responses`    | ModelResponse\[] | All model responses for this round            |

## ModelResponse

| Field           | Type              | Description                                                |
| --------------- | ----------------- | ---------------------------------------------------------- |
| `model_id`      | string            | Full OpenRouter or vendor model ID                         |
| `model_alias`   | string            | Short alias (e.g. `"claude"`)                              |
| `round_number`  | integer           | Round this response belongs to                             |
| `role`          | string            | `"initial"`, `"reflection"`, or `"synthesis"`              |
| `content`       | string            | Full response text                                         |
| `timestamp`     | string (ISO 8601) | When the response was received (UTC)                       |
| `token_count`   | integer \| null   | Total tokens used                                          |
| `input_tokens`  | integer \| null   | Input/prompt tokens                                        |
| `output_tokens` | integer \| null   | Output/completion tokens                                   |
| `latency_ms`    | integer \| null   | Response time in milliseconds                              |
| `error`         | string \| null    | Error message if call failed; null on success              |
| `routing`       | object \| null    | Routing decision (see below)                               |
| `analysis`      | object            | Scoring metadata (populated when `--ground-truth` is used) |

### routing object

| Field            | Type    | Description                                                  |
| ---------------- | ------- | ------------------------------------------------------------ |
| `vendor`         | string  | Provider vendor (e.g. `"anthropic"`)                         |
| `mode`           | string  | Routing mode in effect: `"auto"`, `"direct"`, `"openrouter"` |
| `via_openrouter` | boolean | Whether OpenRouter handled this request                      |

## metadata object

| Key                    | Description                                                                          |
| ---------------------- | ------------------------------------------------------------------------------------ |
| `version`              | Mutual Dissent version that produced the transcript                                  |
| `resolved_config`      | Panel, synthesizer, rounds, routing, and provider keys active for this debate        |
| `stats`                | Token totals, per-model breakdown with input/output split and cost, `total_cost_usd` |
| `scores`               | Ground-truth scoring results (present only if `--ground-truth` was used)             |
| `experiment`           | `ExperimentMetadata` (present only if set via Python API)                            |
| `panelist_context`     | Per-panelist context strings (present only if set via Python API)                    |
| `source_transcript_id` | ID of the source transcript (replay transcripts only)                                |
| `replay_config`        | Synthesizer override and additional rounds (replay transcripts only)                 |

## Example

```json theme={null}
{
  "transcript_id": "a1b2c3d4-...",
  "query": "What are the risks of agentic AI?",
  "panel": ["claude", "gpt", "gemini", "grok"],
  "synthesizer_id": "claude",
  "max_rounds": 1,
  "rounds": [
    {
      "round_number": 0,
      "round_type": "initial",
      "responses": [
        {
          "model_id": "anthropic/claude-sonnet-4-6",
          "model_alias": "claude",
          "round_number": 0,
          "role": "initial",
          "content": "...",
          "timestamp": "2026-03-04T14:00:00+00:00",
          "token_count": 412,
          "input_tokens": 28,
          "output_tokens": 384,
          "latency_ms": 1243,
          "error": null,
          "routing": { "vendor": "anthropic", "mode": "auto", "via_openrouter": false },
          "analysis": {}
        }
      ]
    }
  ],
  "synthesis": { "role": "synthesis", "round_number": -1, "..." : "..." },
  "created_at": "2026-03-04T14:00:00+00:00",
  "metadata": {
    "version": "0.1.1",
    "stats": {
      "total_tokens": 8432,
      "total_cost_usd": 0.0124,
      "per_model": { "claude": { "tokens": 2100, "cost_usd": 0.0031, "calls": 2 } }
    }
  }
}
```