ResearchCall

Survey workbench · dry run · no calls
Test mode — example data, not a real study
All eight stations are open. Ethical locked settings stay hidden. Network disabled · fixture transport · no real calls.

3. Conversation and ethics frame

12 settings shown · an agent asks 8 · 2 are recorded without effect

Text in double quotes is spoken word for word. Everything outside the quotes is rephrased by the agent — a behaviour measured in a real call.

Some settings on this station are kept but not yet read by anything. They are marked, because a control that quietly changes nothing is worse than one that is missing.

The first sentence discloses the automation and the commissioning body.

First spoken block, quoted word for word.

Named verbatim in every call: an automated assistant is calling on their behalf.

Named in the disclosure sentence that opens every call, before consent.

Named verbatim at the end of every interview; without it no live call starts.

Named at the end of every interview; a live run without it is refused.

Spoken in every call before consent is asked; the long notice stays alongside it.

Spoken in every call before consent, and checked against the transcript.

The text names data categories, recording or transcription, the deletion route and the contact point in full.

Spoken as an opening block; the one-sentence version is what the floor says.

The origin is stated transparently and has to match the sampling frame.

Spoken block: where the number came from.

No persuasion; the answer stays voluntary and only feeds the nonresponse analysis.

Adds the refusal question and a schema field that stores the answer.

Only an explicit callback agreement may be carried into the contact rules.

Offers a later call and re-queues whoever accepts, up to the callback limit.

The warm opening may be phrased freely; the disclosure in the first sentence stays intact.

Spoken opening, phrased freely by the agent.

The closing contains thanks and an opportunity to ask questions.

Spoken closing, phrased freely by the agent.

Only reference policy files that already apply.

Nothing reads this value yet: Policy files are named but not read; there is no policy reader yet.

The level steers the AI disclosure in the later publication.

Nothing reads this value yet: The pipeline's L1-L5 catalogue is not in this repository, and inventing its wording would be worse than leaving it unused.