Install it on a real website.

Choose training and search preferences.

Use the generator on the HUMIA homepage. During this pilot, HUMIA v0.3 training maps to AIPREF train-ai, and HUMIA search_retrieval maps to AIPREF search.

allow → y
deny  → n

Publish the HUMIA manifest.

Place the generated JSON at the predictable public endpoint:

https://YOUR-DOMAIN/.well-known/humia.json

The public pilot baseline remains HUMIA v0.3. Do not publish the experimental v0.4 repository prototype as your stable manifest.

Merge the robots additions safely.

Do not replace an existing robots.txt. Preserve all current crawler rules. Merge Content-Usage into the intended robots group and add the experimental HUMIA discovery line.

User-agent: *
Content-Usage: train-ai=n, search=y

Humia: https://YOUR-DOMAIN/.well-known/humia.json

Validate the public HUMIA baseline.

Run the public validator first. It checks the current HUMIA v0.3 deployment. A validator PASS is not an IETF certification or standards endorsement.

Open the public validator →

Run the experimental semantic checker.

From the HUMIA repository branch experiment/v0.4-aipref:

python3 experiments/v0.4/check_live.py https://YOUR-DOMAIN

A successful pilot ends with:

RESULT: PASS — HUMIA and AIPREF are discoverable and semantically consistent for this experiment.