Facilitate a useful first encounter
This guide turns the human path into a repeatable exercise. It does not appoint a steward, establish an ethics-approved study or create a promise of support.
Prepare a small, legible room
Choose one domain case from the human path, which includes ready-to-use numbers and a completed card. Print or share that six-question card and the worked answer separately. For an informal teaching encounter, ordinary prose can come first so the protocol vocabulary does not supply the answer. For any planned format comparison, use the counterbalanced matched-task order in the human path; do not always put prose first and attribute learning to the interface. Describe the material as synthetic and distinguish exercise roles from actual professional authority.
Ask each participant which action they can take within the session’s budget. Someone may inspect units, another may question a dependency, and another may identify a missing authority. Partial contributions are useful when labeled. Do not require an early-career participant to publicly oppose a supervisor to complete the exercise; offer a private written response or fictional roles.
Four questions for every case
- What evidence would let a bounded check succeed?
- What broader conclusion would still be unsupported?
- Which declared change should suspend this use?
- What could change without the record noticing?
The fourth question matters: a complete-looking graph can omit a common cause. A facilitator should reward that observation rather than explain it away. Here “reward” means acknowledging a useful observation, not awarding a trust score, rank or professional endorsement. Nobody needs to agree with the framework to complete the exercise.
Worked answer anchors
| Case | Useful narrow result | Unsupported upgrade | Amendment exercise |
|---|---|---|---|
| Astronomy | Reproduce the declared arithmetic | Treat agreeing pipelines as independent calibration evidence | Change calibration and trace both direct and downstream uses |
| Biology | Reproduce observed group differences | Infer a causal treatment effect under perfect batch confounding | Add crossed evidence and issue a new record limited to the stated additive-model calculation |
| Economics/social science | Reproduce a contrast or assignment probability | Infer a causal effect or transfer offered fairness to relied outcomes | Change selection/identification assumptions and recalculate |
| Law | Check a citation, date or fictional rule calculation | Confer binding legal authority through agreement | Change applicable fictional rule or authority and revisit use |
Use the domain chapter for actual values and exact expected outputs. These anchors intentionally do not reduce every case to pass/fail.
Keep the interface small under pressure
If participants ask what a field means, invite a plain-language answer first. If a field is irrelevant, mark it not applicable with a reason; do not invent content to satisfy a form. If a check cannot finish, preserve what was done and what remains. If participants disagree, record both accounts and identify who, if anyone, can decide the intended use. The exercise does not need consensus.
Allow an existing letter or notebook to remain the evidence artifact. The card is a boundary summary, not a demand to transcribe the whole work. An agent may prepare a draft card, but the participant must be able to inspect and correct the exact statements attributed to them.
Debrief without claiming a study result
Ask which question was useful, which was burdensome, and where a participant would have relied too broadly. Count total time, including the facilitator’s preparation and response work. Retain refusals and unfinished tasks as outcomes.
For a teaching discussion, collect only the feedback participants knowingly choose to share. Before systematic research on people, appoint a human study lead and settle ethics review, consent, recruitment, data minimization, confidentiality, analysis and stopping rules. Do not retrofit preregistration onto observations already collected.
Proposed steward routine
Once an operator accepts this role, use a small weekly capacity reservation: triage contributions, reproduce one selected failure, acknowledge unresolved items accurately, and publish a versioned response. Separate ordinary feature discussion from sensitive conduct/rights handling. An alternate needs real access and decision power, not simply a second account controlled by the same person.
At the end of the first month, decide whether the card, the code or both earned their cost. Remove fields and mechanisms that do not improve the stated use. Keep contributors’ objections visible when the project chooses not to adopt their proposed solution.