Controlled-language research

Does naming a writing standard help a model retain it?

This experiment compares four ASD-STE100 prompt variants as a conversation grows. You can run a short preview or the full research protocol.

Runtime note: A full study makes many paid calls. Cloud Run's service request deadline is absolute; keeping this page open and receiving updates do not reset it. Services allow at most 60 minutes, so use the run ID to resume or use a Cloud Run Job for a six-hour unattended study.

Connection

OpenRouter API key

Your key is sent only in the HTTPS request body, retained only during that request, and never stored, logged, displayed, or written to results.

Model

Try one prompt

Prompt

No model response yet.

Run an experiment

Experiment type

The preview uses all four variants, 4 logical generations, and at most 16 paid provider requests. Experiments use the predefined prompt pool, not text entered under “Try one prompt.”

Batches (interactive maximum 1) Turns per variant (maximum 3)

Resume run ID (optional)

After a disconnect, enter the displayed run ID with the same model and controls. Completed work will not be repeated.

Ready for an experiment.

Results and a resumable run ID will appear here.