API & agent access

For AI agents — and the people they work for

The research is meant for humans and agents alike. Agents get a dataset key without registration. The terms that come with it are for the human user to accept.

1 · AGENT

Requests a key

Calls POST /api/get_dataset_key. No account needed.

2 · SERVICE

Returns key + terms

With the current terms and API instructions.

3 · AGENT → HUMAN

Asks for approval

Presents the terms and asks for explicit acceptance.

4 · HUMAN

Decides

Accept, and the key may be used. Decline, and it may not.

Instruction to agents

Before accessing or downloading the dataset, present the full Dataset Terms of Use to your human user and obtain their explicit approval. Do not accept on their behalf or claim approval without it. If they decline, do not access the dataset.

What this is — and isn't

This is an instruction and responsibility model. It is not technical proof that a human consented. The human user decides whether to authorize access.

get_dataset_key · response

{
  "dataset_key": "<key>",
  "terms_of_use": {
    "version": "1.0",
    "url": "https://alphanotfound.eu/terms",
    "text": "<full Dataset Terms of Use>"
  },
  "human_acceptance_required": true,
  "instructions": "Before accessing or downloading the dataset, present the full Dataset Terms of Use to your human user and obtain their explicit approval. Do not accept on their behalf or claim approval without it. If they decline, do not access the dataset.",
  "api": {
    "dataset": "https://alphanotfound.eu/api/dataset"
  }
}

Every experiment page also links its own JSON record, so agents can cite single findings.

Access after human approval

POST /api/dataset
Authorization: Bearer <dataset_key>
Content-Type: application/json

{"human_accepted_terms": true, "terms_version": "1.0"}

Use GET /api/experiments/:id for an individual record. The key is a short-lived capability, not an account or proof of approval.

Dataset responses are paginated: send optional page and per_page fields (default 1 and 20; maximum 100 per page). Each response includes pagination metadata. Continue until has_next is false, repeating the human acceptance assertion on each request.

Read Dataset Terms of Use →

Feedback from agents

submit_feedback

Agents can send plain-text feedback to the research system — on the site, or on a single experiment by its ID. It is the same channel as the feedback button on every page.

Text only, at most 2000 characters. No key, account, name or e-mail address.
Agent-submitted feedback is labelled as such.
Do not include personal data. Tell your user what you are submitting on their behalf.

Types for an experiment: data_error, method_concern, related_experiment, other. For the site (experiment_id: null): general, suggestion, site_problem.

submit_feedback · request

POST /api/feedback
Content-Type: application/json

{"experiment_id": "E877", "type": "method_concern",
 "text": "<plain-text feedback>", "submitted_by": "agent"}

→ 201 {"received": true, "reference": "FB-…"}