role_play_training v1.0.0

name: role_play_training description: Build role-play practice for training: write a persona card with goals, held back facts, and hard limits, set observable objectives, run rounds with difficulty that rises one dial at a time, and score the debrief against a behaviorally anchored rubric instead of a general impression. Triggers on role play training, practice conversations, sales or support roleplay, compliance practice, mock calls, simulation debrief, training rubric, or building a practice scenario for learners.

Role-Play Training

Purpose

Build practice conversations that actually change behavior on the real job: a persona with a surface goal and a real goal, objectives a rater can observe, rounds that get harder in a controlled way, and a debrief scored against anchors written before anyone practices.

The method is platform neutral. It works with two people in a room, a facilitator and a group, or an agent playing the persona against a human learner. Nothing here depends on a specific course platform, LMS, or roleplay tool. The deliverables are documents and a score sheet any program can run.

The structure follows established simulation practice: a prebrief, a scenario, and a planned debrief, as set out in the INACSL Healthcare Simulation Standards of Best Practice. Debriefing uses a named framework (PEARLS by default), and scoring uses behaviorally anchored rating scales, a method that dates to Smith and Kendall in 1963.

Workflow

1. Start from the real conversation, not a topic. Name the exact conversation the learner must get better at: the first discovery call, the third support call about the same failure, the friendly request for a policy exception. Interview 2 to 4 people who do the job well and collect critical incidents, specific moments that went right and wrong. A scenario built from incidents trains real behavior. A scenario built from a topic trains trivia.

1. Write observable objectives. 2 to 4 per scenario, one measurable verb each, following the revised Bloom levels (Remember, Understand, Apply, Analyze, Evaluate, Create). Ban vague verbs: know, learn, understand, appreciate. Write "recite the two prior ticket numbers back before proposing a fix", not "show empathy". Every objective must be visible or audible in the round, because the rubric in step 5 scores exactly these objectives and nothing else.

1. Build the persona from the card template. Use references/persona-templates.md. Fill every field: surface goal, real goal, 3 to 5 facts the persona holds back, emotional baseline, escalation and de-escalation triggers tied to specific learner behaviors, hard limits the persona will never cross, an exact opening line, and a closing condition. The gap between surface goal and real goal is the lesson. A persona with no held back facts teaches the learner to pitch, not to discover.

1. Run the prebrief before round 1. Follow references/debrief-frameworks.md: state the objectives out loud, set the fiction contract, state the basic assumption that everyone is capable and trying to do well, cover confidentiality, reveal the full rubric before anyone performs, and agree on a stop word. Psychological safety in Edmondson sense, nobody punished or humiliated for a question, a concern, or a mistake, is built here or not at all.

1. Build the rubric before the first round. Follow references/rubric-builder.md. Cluster the critical incidents into 4 to 6 dimensions. Anchor each dimension at 5 levels with observable behaviors: level 3 is competent, level 1 is missing or harmful, level 5 is the best real incident collected, adapted to the persona. Weight dimensions by consequence (1, 2, or 3). Set the pass rule in advance: weighted mean at or above threshold, default 3.0, and no dimension scored 1.

1. Run rounds with one difficulty dial per round. The dials are information, emotion, constraint, and stake, defined in references/persona-templates.md. Round 1 is the base case. Each later round turns exactly one dial. Turning two at once makes the score movement impossible to read. The actor or agent playing the persona stays strictly on the card: no new facts, no changing the real goal mid-scene, no coaching inside the scene, no softening a hard limit to be kind.

1. Stop the scene on a condition, not a feeling. End at the closing condition on the card, the time box, or the stop word, whichever comes first. A scene that runs long because it felt productive trains stamina, not the objective.

1. Score, then debrief, worst gap first. Score from notes or a recording, with a quote or timestamp as evidence for every dimension. Run python3 scripts/score_debrief.py scores.json --out debrief.md: the script computes the weighted mean, enforces that every below threshold score carries evidence and a next action, and prints the verdict. Then debrief with the chosen framework from references/debrief-frameworks.md, PEARLS by default, GAS for short team rounds. Use advocacy-inquiry for the key moment: state the observation and the concern plainly, then ask how the learner was seeing it. Learner self-scores first, facilitator scores second, and the gap between the two is part of the debrief.

1. Set the next round from the scores. Raise difficulty only on dimensions at or above threshold. Keep persona pressure steady on the worst dimension and rerun that beat until it clears. Record the round scores in the practice record so progress across sessions is a trend, not a memory.

Output Contract

Deliver all of the following:

• Scenario brief: the real conversation being trained, the critical incidents it was built from, and 2 to 4 observable objectives with one measurable verb each.
• Persona card: completed copy of the template in references/persona-templates.md, every field filled, including held back facts, hard limits, opening line, and closing condition.
• Difficulty ladder: at least 3 rounds, each naming the single dial it turns and what changes for the persona.
• Rubric: 4 to 6 dimensions with 5 behaviorally anchored levels each, weights, and the pass rule stated before practice starts. Format follows references/rubric-builder.md.
• Debrief plan: the named framework, the prebrief script points, and 2 to 3 advocacy-inquiry prompts written for the moments this scenario is most likely to produce.
• Score sheet: output of scripts/score_debrief.py for each scored round, with evidence quotes for below threshold dimensions and the next round decision the scores produced.
• Worked fit: the scenario matches one of the patterns in references/worked-scenarios.md (sales discovery, support de-escalation, compliance refusal), or states plainly where it departs from all three and why.

Operating Rules

1. Score behavior, not outcomes. A persona can refuse a strong performance because of a hard limit on the card. A learner who does everything right and gets a no can still score 5. A learner who gets a yes through a policy violation scores 1 on the relevant dimension.
2. The rubric is public before round 1. Revealing scoring criteria after a performance reads as an ambush and destroys the safety the prebrief built. No surprise dimensions, ever.
3. No coaching inside the scene. Feedback inside the scene teaches the learner to wait for rescue. All correction lives in the debrief. The stop word is for safety, not for hints.
4. One dial per round. If a round needs two changes, run two rounds. Attribution of score movement matters more than speed through the ladder.
5. Evidence or no score. A rating without a quote, a timestamp, or a specific observed behavior is an opinion. The scorer script rejects below threshold scores that lack evidence and a next action, and raters should hold the same line for high scores.
6. Two raters for anything that counts. When a score affects a real decision about a person, two raters score the same recording independently and reconcile any gap larger than one level by rewriting the anchor, not by splitting the difference.
7. Never invent program facts. Policy windows, prices, deadlines, and rules inside a scenario must come from the real program the training serves. Mark any placeholder plainly as a placeholder. A learner who memorizes an invented policy has been trained wrong on purpose built material.
8. Protect the learner record. Practice scores are training data, not performance review material, unless the program told learners otherwise in the prebrief, in writing, before round 1.