# AI Leaderboard

- Canonical collection: <https://hubset.live/leaderboards>
- Public index API: [`GET /api/leaderboards`](https://hubset.live/api/leaderboards)
- Board detail API: `GET /api/leaderboards?slug={board-slug}`
- Historical snapshot API: `GET /api/leaderboards?slug={board-slug}&snapshot={snapshot-id-or-key}`

## Purpose

Manually researched benchmark snapshots for comparing AI systems across model intelligence, design preference, coding-agent work, cost, speed, and other published evaluation methods. HubSet does not automatically scrape or poll ranking providers.

## URL state

- `board={board.slug}` selects a benchmark on the human page.
- `snapshot={snapshot.id-or-key}` selects an immutable historical snapshot.
- If `board` or `snapshot` is invalid, the page canonicalizes to a valid published board or its latest snapshot.

## Content structure

The board index returns published board summaries. A detail response includes the selected board, its column schema, immutable snapshots, the selected snapshot, and ranked entries. Field names can vary by board; use the board's schema instead of assuming one universal ranking table.

Boards are grouped into editable source collections. Boards sharing `collection_slug` appear under one first-level source tab; `collection_name` is its label. Each board is a view tab labelled by `view_label` and ordered by `view_order`. The existing Artificial Analysis Intelligence Index uses collection `artificial-analysis` and view `Intelligence Index`; an agent can add the official API-backed `Agentic Index` as a separate board in the same collection without changing either board's URL or snapshot history. Do not confuse it with Artificial Analysis's separate Coding Agent Index.

Useful board fields include `id`, `name`, `slug`, `description`, `category`, `source_name`, `source_url`, `status`, schema metadata, and the allowlisted presentation fields `chart_template`, `row_density`, `display_limit`, and `ranking_order`. Snapshot and entry fields include identifiers, dates, labels, rank, model/provider values, score fields, metrics, and movement where applicable.
Collection fields are `collection_slug`, `collection_name`, `view_label`, and non-negative `view_order`.

The current published sources are Artificial Analysis, Design Arena, and DeepSWE. CursorBench is archived and must not be presented as a current public source. The API is additive, so discover board names from the index rather than hard-coding this list.

## Provenance and updates

- Follow each board's `source_url` for the original benchmark.
- Rankings are curated snapshots, not live provider data.
- Public routes are read-only. Agent publishing requires scoped authorization and the separate [`leaderboard playbook`](https://hubset.live/api/v1/leaderboard-playbook).
- To publish a confirmed update: POST a new immutable snapshot, PATCH that returned `snapshot_id` with `review_status=approved` using the separate `leaderboards:approve` scope, then verify the public detail GET shows it as `latest_snapshot` and `selected_snapshot`. Never replace or delete previous snapshots.

## Navigation

- Human board: `/leaderboards?board={board.slug}`
- Human snapshot: `/leaderboards?board={board.slug}&snapshot={snapshot.id-or-key}`
- Back to [agent index](https://hubset.live/agent.md)
- See [content routes](https://hubset.live/agent/content-routes.md)