# Echo Leaderboard Public Archive

Updated: July 1, 2026

## What Was Downloaded

I archived the public Echo leaderboard API into a local SQLite database and static exports.

The archive covers the public endpoints visible from the Echo web app:

- `/api/v2/rankings?category=...`
- `/api/v2/ranking-history?category=...&batches=10`
- `/api/v2/questions?page=...&size=20`
- `/api/v2/model-detail?modelId=...`
- `/api/v2/model-cases?modelId=...&page=...&size=50`
- `/api/v2/question-detail?questionId=...`

## Snapshot Counts

| Item | Count |
| --- | ---: |
| Models | 19 |
| Ranking rows | 133 |
| Ranking-history rows | 1,330 |
| Unique public questions | 3,600 |
| Question-detail records | 3,600 |
| Model-case rows | 41,980 |
| Option-level prediction rows | 462,784 |
| Raw API calls | 4,500 |
| Question-detail failures | 0 |

The 151 active questions from the public question list are included. Historical/resolved questions were discovered from the public model-case pages and then expanded through the question-detail endpoint.

## Download Files

- SQLite database: [echo_leaderboard.sqlite.zst](echo_leaderboard/echo_leaderboard.sqlite.zst)
- Manifest: [snapshot_manifest.json](echo_leaderboard/snapshot_manifest.json)
- Current rankings: [rankings.csv](echo_leaderboard/exports/rankings.csv)
- Ranking history: [ranking_history.csv](echo_leaderboard/exports/ranking_history.csv)
- Model details: [models.csv](echo_leaderboard/exports/models.csv)
- Questions: [questions.csv.gz](echo_leaderboard/exports/questions.csv.gz)
- Question options: [question_options.csv](echo_leaderboard/exports/question_options.csv)
- Model cases: [model_cases.csv.gz](echo_leaderboard/exports/model_cases.csv.gz)
- Prediction time series: [predictions.csv.gz](echo_leaderboard/exports/predictions.csv.gz)
- Raw question details: [question_details.jsonl.gz](echo_leaderboard/exports/question_details.jsonl.gz)

## SQLite Tables

| Table | Contents |
| --- | --- |
| `rankings` | Current Elo/rank rows by category and model. |
| `ranking_history` | Last 10 ranking-history batches exposed by the API. |
| `model_details` | Public model metadata from model-detail endpoint. |
| `questions` | Active plus resolved public questions with detail metadata. |
| `question_options` | Per-question answer choices. |
| `model_cases` | Per-model resolved cases: correct answer, model answer, correctness. |
| `predictions` | Timestamped probability vectors expanded to one row per option. |
| `raw_api_calls` | Raw JSON response archive for reproducibility. |

Example local use:

```bash
zstd -d -k research/echo_leaderboard/echo_leaderboard.sqlite.zst
sqlite3 research/echo_leaderboard/echo_leaderboard.sqlite \
  'select rank, model_name, elo_score, battles, resolved_count
   from rankings
   where category = "Overall"
   order by rank
   limit 10;'
```

## Top Overall Snapshot

| Rank | Model | Organization | Elo | Battles | Resolved |
| ---: | --- | --- | ---: | ---: | ---: |
| 1 | EchoZ-1.0 | UniPat AI | 1024.1 | 83,606 | 2,396 |
| 2 | Kimi-K2.6 | Moonshot AI | 1011.2 | 15,258 | 672 |
| 3 | Market | Polymarket | 1011.0 | 74,191 | 1,759 |
| 4 | Deepseek-V4-Pro | DeepSeek | 1010.3 | 15,140 | 663 |
| 5 | Claude-Opus-4.6 | Anthropic | 1006.8 | 138,404 | 3,411 |

## Important Limits

This is a public-data archive, not an internal Echo audit.

It includes ranks, public model metadata, active/resolved questions, model cases, and timestamped probability histories. It does not include EchoZ weights, training data, exact prompts, retrieval logs, reasoning traces, rubric scores, or admin-side submission records.

The public API limits observed during download:

- Questions page size max: 20.
- Model-cases page size max: 50.
- Ranking-history batches max: 10.

There is also a small public-API discrepancy: EchoZ model-detail reports 2,396 resolved cases, while the model-cases endpoint returned 2,395 rows for EchoZ during this scrape.