# Seed, ODIN and Lean

> Discover completed attempt exports and their separate human-review workflow.

Source: https://math-api.qmindlabs.ai/docs/guides/families

## Check availability

First call `GET /api/math/families`. When `enabled` is false, the current catalog is not available. Previous-generation Math remains available through the existing task endpoints.

[GET/api/math/families](https://math-api.qmindlabs.ai/docs/api-reference/tag/math-families/GET/api/math/families.md)

## Three families in one Math workspace

| Family | Public evaluation | Private review evidence |
| --- | --- | --- |
| Seed | Problem | Final answer and verification |
| ODIN | Problem | Rubric and independent review records |
| Lean | Lean statement and context | Lean proof, evaluation and rubric |

All source buckets remain private. A public export is a shareable payload within an authenticated API, not a public bucket. A successful attempt or accepted export is not a human approval. Counts represent exports, not distinct mathematical problems.

## Read the current catalog

```sh
curl --fail-with-body 'https://math-api.qmindlabs.ai/api/math/catalog/evaluations?family=lean&page=1&page_size=25' \
  -H "Authorization: Bearer $MATH_API_KEY"
```

[GET/api/math/catalog/evaluations](https://math-api.qmindlabs.ai/docs/api-reference/tag/math-families/GET/api/math/catalog/evaluations.md)

[GET/api/math/catalog/attempts](https://math-api.qmindlabs.ai/docs/api-reference/tag/math-families/GET/api/math/catalog/attempts.md)

Use `evaluations[].id` for evaluation routes. Failed attempts and attempts with zero exports appear separately in the operator attempt history. Reviewers see only assigned evaluations. A PDF is not required for these evaluations.

## Review a verified revision

[GET/api/math/catalog/evaluations/{task\_id}](https://math-api.qmindlabs.ai/docs/api-reference/tag/math-families/GET/api/math/catalog/evaluations/%7Btask_id%7D.md)

[GET/api/math/catalog/evaluations/{task\_id}/evidence](https://math-api.qmindlabs.ai/docs/api-reference/tag/math-families/GET/api/math/catalog/evaluations/%7Btask_id%7D/evidence.md)

[POST/api/math/catalog/evaluations/{task\_id}](https://math-api.qmindlabs.ai/docs/api-reference/tag/math-families/POST/api/math/catalog/evaluations/%7Btask_id%7D.md)

Read `evaluation.version` before each workflow action. Assignments, comments, decisions, reopening and archival are versioned and audited. Decisions bind to the imported evidence revision. A source reimport cannot reset an existing review. Private evidence requires `files:read` and current assignment or operator access.

## Keep previous work accessible

Previously imported Math belongs to the previous ODIN generation. Its IDs, reports and reviews remain on `/api/math/tasks` and their existing routes. Use the dashboard generation selector to switch. Current catalog IDs must not be sent to previous task routes. The current catalog does not launch compute, replace source artifacts or fabricate pass rates.
