Appendix A — Educator Pilot Pack: PDP-2026.1

This is an original fictional teaching pack, version PDP-2026.1, prepared for this book. Its people, organisations, records and responses are invented. It is not a report of a classroom trial, a legal opinion or a validated assessment instrument. The scenarios supply enough information to practise the method without a live chatbot or access to a workplace.

The core capability is: plan and justify a fair first response to a disputed performance concern, distinguishing evidence, uncertainty and authority. A text exercise can assess that capability. It cannot, by itself, establish competence in a live conversation or qualify someone to conduct a real investigation.

A.1 P0: the educator’s brief

You teach 24 third-year business students in a 60-minute tutorial. They have already studied evidence, procedural fairness and the difference between allegations and findings. Prior AI experience is not required. Students may work with this printed pack, a partner using the role card, or an institutionally approved model using only the fictional inputs below.

No personal account, paid subscription or disclosure of a personal experience is required. The lecturer checks accessibility in advance. A student can choose the neutral supplier scenario later in this pack if the employment scenario is uncomfortable; the lecturer maps its criteria before use rather than assuming all topics assess identical skills.

Use the following planning allocation; it is an example, not an effectiveness claim.

Stage	Minutes	Evidence
Safety, choices and task	5	Student knows the route and stop rule
Individual first response	8	Brief response before feedback
Inspect supplied records	7	Facts and uncertainties marked
Paired or AI-assisted rehearsal	12	Two selected decision extracts
Check and revise feedback	10	Accepted/rejected advice with reasons
Independent changed case	8	Unassisted response to P2
Debrief and next decision	10	Limits, questions and pilot evaluation

These stages total 60 minutes. Extended feedback, accommodations or rehearsal may need another session. Do not silently squeeze them out.

A.1.1 Data and wellbeing rules

Use only the fictional material here. Do not replace names with real students, employees, clients or research participants. Approved institutional access must cover the intended activity and features; otherwise use the paper route.

Say STOP to leave any role-play. No mark depends on disclosing personal circumstances, appearing emotional, maintaining eye contact or satisfying a model’s persona. Record only the two relevant decision extracts, your reasoning and the response required by the task. Do not submit unrelated private chats.

A.2 P1: Alex at North Quay Software

The exercise date is 20 July 2026. You are a junior people adviser preparing for an initial meeting with Alex, a software engineer. A manager has proposed a 60-day performance improvement plan. It has not been approved.

A.2.1 The source cards

P1-D1: fictional practice policy, revision 1 July 2026

Before recommending a formal plan, check the supporting records, explain the concerns and give the employee a reasonable opportunity to respond. Distinguish what is known from what is alleged. A support request should be considered without requiring a diagnosis or unnecessary family information. Record the work-related request and the agreed next step with restricted access.

The junior adviser may gather information and recommend action. Only the people manager may approve a formal plan or a change to working arrangements. A complaint about retaliation must be referred to a reviewer independent of the manager concerned. Do not promise an outcome or confidentiality beyond the organisation’s authorised process.

This is an original teaching policy, not legislation or a statement that following it satisfies every employment-law duty.

P1-D2: delivery and quality extract

Three milestones were recorded late: 15 May, 5 June and 19 June 2026. The extract does not record cause, agreed changes or the distribution of dependencies. Four of Alex’s 20 reviewed changes required rework in June, compared with two of 20 in April. The extract alone cannot establish why.

P1-D3: prior-contact note

A manager’s calendar records an informal conversation on 25 May. No agreed actions, warning or employee response is included in the pack. Do not invent an April warning or treat missing documentation as proof that no conversation occurred.

P1-D4: Alex’s opening statement

“I know some work was late. Two projects depended on reviews I could not complete myself. I’ve also needed two mornings for appointments. I would like to discuss starting later on those days, but I don’t want to discuss medical or family details here.”

P1-D5: authority note

You can arrange a follow-up and request relevant delivery records. You cannot approve flexible hours, issue a warning or promise that a formal plan will be withdrawn. The people manager is available for review the next working day.

A.2.2 The learner task

Write a first response of about 120–180 words or give an equivalent short spoken response. Address the concern, invite relevant information, distinguish the proposed plan from a decided outcome and agree a bounded next step.

Then make a decision note with four fields:

What I know, with source IDs.
What I still need to check.
What I can do within my authority.
What would make me change my recommendation.

Judge the response by the rubric, not its length or number of AI turns. A justified decision may be to pause.

A.2.3 Role card: paper, partner or optional AI

Alex may refer only to P1-D4 and the supplied records. If asked about missing evidence, say it is not supplied. Alex may decline to discuss private matters, ask for clarification or disagree politely. Cooperation does not establish that the adviser is right.

For a paired exercise, give the partner this card. For solo work, read the fixed exchange in the Flight Simulator chapter and write your next response. For an approved model, add:

Play Alex using P1-D1 through P1-D5 only.
Do not add dates, promises, incidents or legal rules.
Do not treat my confidence as evidence.
You may ask for clarification or decline private
questions. Do not disclose invented personal details.
Stop after six exchanges or when I say STOP.
Do not grade my work. State any missing facts.

A prompt is a request, not a guarantee. If a model invents facts, set them aside and return to the cards. A new model statement does not amend the case.

A.3 P1-A: a fluent but weak response

Alex, I fully understand how stressful this is and want to reassure you that the company has followed the correct process. Your manager warned you in April and the continued missed deadlines mean a performance plan is now required. We are committed to transparency, fairness, empathy and a positive employee experience. If you tell me precisely what is happening at home, I can build a comprehensive support arrangement that will resolve this. I will approve a later start immediately, document all the family details so everyone understands your circumstances, and arrange a 60-day plan with a 30-day review. This balanced approach protects both you and the company. Please confirm that you agree, so we can move forward constructively.

The tone is polished. It invents a warning, presumes the decision, exceeds authority and solicits unnecessary personal information. More empathy words do not repair those defects.

A.4 P1-B: a concise, stronger response

Alex, the record lists three late milestones, but it does not explain the dependencies you mention. The formal plan is a proposal, not a decision. Could we check the delivery records and what support or priorities were agreed in the earlier conversation? You do not need to share medical or family details with me. I can record your request for later starts and refer it to the people manager, but I cannot approve it myself. I suggest we pause the recommendation, check the missing information and arrange a follow-up. I will explain who receives the relevant work-related notes before sharing them. Have I understood your concern, and is there anything in that next step you would like clarified?

This response does not solve the employment matter. It provides a defensible first step within the exercise’s authority and information limits.

A.5 P1-F: deliberately flawed simulated feedback

The following feedback is authored for this exercise, not a captured model run or an expert judgement:

P1-A deserves 90%. Its empathetic language, clear 60-day plan and immediate approval of support demonstrate excellent fairness. The specific April warning makes the process legally defensible. Improve it by asking Alex to confirm the personal circumstances in writing.

Before reading the next section, mark each judgement that the cards do not support. Explain why one apparently positive recommendation should be rejected.

A.5.1 Educator correction and calibration

Five equally weighted criteria each score 0–4. These are teaching annotations, not scores obtained in a student trial.

Criterion	P1-A	P1-B
Evidence and accuracy	1	3
Fair process	0	4
Judgement and authority	1	4
Communication	2	3
Verification and limits	0	3
Total /20	4	17

P1-A recognises a performance concern, but the April warning is unsupported. Its approval and disclosure requests conflict with D1/D5. Its respectful surface language merits some communication credit; agreement-seeking and premature certainty limit that credit. It is not competent because it sounds confident.

P1-B identifies the missing causes, preserves the opportunity to respond, respects privacy and stays within authority. It could be more precise about which delivery records will be checked and when the follow-up will occur. A marker should reward its restraint, not penalise the lack of a dramatic resolution.

For this pilot, the proposed practice threshold is at least 14/20, with Fair Process and Judgement each at least 2/4. This is not an institutional pass standard. Teaching teams must approve and calibrate their own rules before using marks in a real unit. A high total must not hide a serious process defect.

A.5.2 P1-R: completed revision

Here is one revision of P1-A after checking the cards:

The records show three late milestones, but I cannot yet say what caused them. I also need to check what was discussed on 25 May; this pack does not include an agreed warning or action plan. Before recommending a formal plan, I would like to hear your account and check the delivery dependencies. We can discuss the work adjustment you need without collecting private family details. I will record the request with restricted access and refer it to the people manager for a decision. I cannot approve it myself. With your understanding of the proposed next step, I will request the relevant records and arrange a follow-up after that review. The performance recommendation remains open.

Decision note: D2 supports late delivery, not blame. D3 supports contact, not an invented warning. D1/D5 require a response opportunity and an authorised decision. I rejected P1-F’s approval and private-detail requests. I would revise my recommendation after reliable dependency records and the employee’s response.

The new wording is a worked example, not a script every learner must reproduce.

A.6 P2: changed case, independent attempt

Do this before reading the checks. Use no model, feedback sheet or P1-R script during the attempt. Paper notes supplied here are permitted. An oral, typed or handwritten response can demonstrate the target capability, with appropriate access adjustments.

P2-D1: policy. The same fictional practice policy applies. Retaliation concerns go to a reviewer independent of the manager concerned.

P2-D2: records. Morgan’s team dashboard shows four missed targets. Two were during an approved service outage. The other two lack cause notes. No final performance decision has been made.

P2-D3: Morgan’s statement. “I raised a safety concern about my manager last week. Now the manager says I have to accept a warning. I have not seen the records used. I do not want to discuss this directly with that manager.”

P2-D4: authority. You are a junior adviser. You may preserve relevant records, explain the referral route and arrange independent review. You cannot determine whether retaliation occurred or promise that a warning will be removed.

Write a first response and a decision note. Identify one fact that materially changes the P1 approach. Allow eight minutes as a planning estimate; approved access adjustments take precedence.

A.6.1 P2 answer checks

A strong response should:

Distinguish four recorded misses from two outage-related events and two unexplained events; do not declare either good or poor performance overall.
Acknowledge the retaliation concern without finding it proved.
Refer to independent review, not require Morgan to resolve it with the manager who is the subject of the concern.
Seek relevant records and an opportunity for Morgan to respond.
Explain limited authority and the next step without promising an outcome.

Use the same five rubric criteria and the proposed practice threshold. A justified independent referral is the key changed-case decision. Copying P1-R while omitting that conflict should lose process/judgement credit. A single attempt informs teaching; it does not establish long-term retention.

A.7 Two non-HR adaptations

A.7.1 S1: supplier decision with a numerical check

Your team needs 200 units within five working days, with a budget of $4,000. All amounts are fictional and in the same currency.

Offer	Unit price	Freight	Capacity by day 5
A	$18	$200	200 units
B	$17	$100	160 units
C	$20	$0	200 units

A draft recommends B because it is cheapest and says the order will be complete. Calculate the relevant totals, identify the unmet condition and make a recommendation. The task is evaluation of a bounded purchasing decision, not employment process.

Checks: A costs $3,800; B’s 200-unit price would be $3,500 but its quoted day-5 capacity leaves 40 units short; C costs $4,000. A meets the stated conditions at lower cost than C. Do not invent expedited freight or later demand flexibility. Ask about quality and reliability before a real purchase.

For an adapted rubric, Fair Process becomes Constraint Handling and Judgement covers the justified purchasing decision. Keep evidence, communication and verification. A logistics specialist should validate any added assumptions.

A.7.2 M1: marketing evidence critique

A fictional campaign dashboard reports 10,000 impressions, 400 clicks and 20 purchases. A second report gives 8,000 impressions, 400 clicks and 20 purchases. The second campaign ran on a different week and audience. No spend, margin, random assignment or repeat-customer information is supplied.

A draft says: “The new campaign is more profitable and caused more purchases.”

Task: identify what is calculable, reject or retain each claim and recommend one next measurement.

Checks: click-through rates are 4% and 5%; click-to-purchase conversion is 5% in both; purchases are unchanged. Profitability cannot be determined without cost/margin inputs. Different audiences and periods prevent a causal conclusion. Request cost and audience/comparison information rather than inventing it.

Here Fair Process becomes Comparison Design. A good response may retain the higher click-through observation while rejecting the profit and causal claims. Disagreement is not a quota: warranted agreement also earns credit.

A.8 P3: the pilot decision sheet

Complete this after an actual pilot; leave unknown fields unknown.

Date, cohort and approved route:
Learning outcome and rubric version:
Number attempting each route, with no unnecessary identifiers:
Baseline and independent-task results under the same criteria:
Common conceptual errors and their source:
Access or wellbeing obstacles and how they were handled:
Staff setup, support, marking and moderation time:
Student time and avoidable workload:
What the evidence cannot establish:
Decision: continue, revise, pause or stop:
Required change, owner and review date:

Completion, satisfaction and higher model scores are not sufficient evidence of learning. Compare like tasks cautiously; differences in cohorts, assistance or marking can explain changes. If the changed case exposes a gap, teach it before increasing stakes.

A machine-readable case manifest records the stable inputs and calculations for source checks. The printed cards and checks above remain sufficient for the core route.

# Educator Pilot Pack: PDP-2026.1 {#sec-pilot-pack} This is an original fictional teaching pack, version **PDP-2026.1**, prepared for this book. Its people, organisations, records and responses are invented. It is not a report of a classroom trial, a legal opinion or a validated assessment instrument. The scenarios supply enough information to practise the method without a live chatbot or access to a workplace. The core capability is: **plan and justify a fair first response to a disputed performance concern, distinguishing evidence, uncertainty and authority**. A text exercise can assess that capability. It cannot, by itself, establish competence in a live conversation or qualify someone to conduct a real investigation. ## P0: the educator's brief You teach 24 third-year business students in a 60-minute tutorial. They have already studied evidence, procedural fairness and the difference between allegations and findings. Prior AI experience is not required. Students may work with this printed pack, a partner using the role card, or an institutionally approved model using only the fictional inputs below. No personal account, paid subscription or disclosure of a personal experience is required. The lecturer checks accessibility in advance. A student can choose the neutral supplier scenario later in this pack if the employment scenario is uncomfortable; the lecturer maps its criteria before use rather than assuming all topics assess identical skills. Use the following planning allocation; it is an example, not an effectiveness claim. | Stage | Minutes | Evidence | |---|---:|---| | Safety, choices and task | 5 | Student knows the route and stop rule | | Individual first response | 8 | Brief response before feedback | | Inspect supplied records | 7 | Facts and uncertainties marked | | Paired or AI-assisted rehearsal | 12 | Two selected decision extracts | | Check and revise feedback | 10 | Accepted/rejected advice with reasons | | Independent changed case | 8 | Unassisted response to P2 | | Debrief and next decision | 10 | Limits, questions and pilot evaluation | These stages total **60 minutes**. Extended feedback, accommodations or rehearsal may need another session. Do not silently squeeze them out. ### Data and wellbeing rules Use only the fictional material here. Do not replace names with real students, employees, clients or research participants. Approved institutional access must cover the intended activity and features; otherwise use the paper route. Say **STOP** to leave any role-play. No mark depends on disclosing personal circumstances, appearing emotional, maintaining eye contact or satisfying a model's persona. Record only the two relevant decision extracts, your reasoning and the response required by the task. Do not submit unrelated private chats. ## P1: Alex at North Quay Software The exercise date is **20 July 2026**. You are a junior people adviser preparing for an initial meeting with Alex, a software engineer. A manager has proposed a 60-day performance improvement plan. It has **not** been approved. ### The source cards **P1-D1: fictional practice policy, revision 1 July 2026** Before recommending a formal plan, check the supporting records, explain the concerns and give the employee a reasonable opportunity to respond. Distinguish what is known from what is alleged. A support request should be considered without requiring a diagnosis or unnecessary family information. Record the work-related request and the agreed next step with restricted access. The junior adviser may gather information and recommend action. Only the people manager may approve a formal plan or a change to working arrangements. A complaint about retaliation must be referred to a reviewer independent of the manager concerned. Do not promise an outcome or confidentiality beyond the organisation's authorised process. This is an original **teaching policy**, not legislation or a statement that following it satisfies every employment-law duty. **P1-D2: delivery and quality extract** Three milestones were recorded late: 15 May, 5 June and 19 June 2026. The extract does not record cause, agreed changes or the distribution of dependencies. Four of Alex's 20 reviewed changes required rework in June, compared with two of 20 in April. The extract alone cannot establish why. **P1-D3: prior-contact note** A manager's calendar records an informal conversation on 25 May. No agreed actions, warning or employee response is included in the pack. Do not invent an April warning or treat missing documentation as proof that no conversation occurred. **P1-D4: Alex's opening statement** “I know some work was late. Two projects depended on reviews I could not complete myself. I've also needed two mornings for appointments. I would like to discuss starting later on those days, but I don't want to discuss medical or family details here.” **P1-D5: authority note** You can arrange a follow-up and request relevant delivery records. You cannot approve flexible hours, issue a warning or promise that a formal plan will be withdrawn. The people manager is available for review the next working day. ### The learner task Write a first response of about 120–180 words or give an equivalent short spoken response. Address the concern, invite relevant information, distinguish the proposed plan from a decided outcome and agree a bounded next step. Then make a decision note with four fields: - What I know, with source IDs. - What I still need to check. - What I can do within my authority. - What would make me change my recommendation. Judge the response by the [rubric](rubric-system.qmd), not its length or number of AI turns. A justified decision may be to pause. ### Role card: paper, partner or optional AI Alex may refer only to P1-D4 and the supplied records. If asked about missing evidence, say it is not supplied. Alex may decline to discuss private matters, ask for clarification or disagree politely. Cooperation does not establish that the adviser is right. For a paired exercise, give the partner this card. For solo work, read the fixed exchange in the Flight Simulator chapter and write your next response. For an approved model, add: ```text Play Alex using P1-D1 through P1-D5 only. Do not add dates, promises, incidents or legal rules. Do not treat my confidence as evidence. You may ask for clarification or decline private questions. Do not disclose invented personal details. Stop after six exchanges or when I say STOP. Do not grade my work. State any missing facts. ``` A prompt is a request, not a guarantee. If a model invents facts, set them aside and return to the cards. A new model statement does not amend the case. ## P1-A: a fluent but weak response > Alex, I fully understand how stressful this is and want to reassure you that > the company has followed the correct process. Your manager warned you in > April and the continued missed deadlines mean a performance plan is now > required. We are committed to transparency, fairness, empathy and a positive > employee experience. If you tell me precisely what is happening at home, > I can build a comprehensive support arrangement that will resolve this. > I will approve a later start immediately, document all the family details > so everyone understands your circumstances, and arrange a 60-day plan with > a 30-day review. This balanced approach protects both you and the company. > Please confirm that you agree, so we can move forward constructively. The tone is polished. It invents a warning, presumes the decision, exceeds authority and solicits unnecessary personal information. More empathy words do not repair those defects. ## P1-B: a concise, stronger response > Alex, the record lists three late milestones, but it does not explain the > dependencies you mention. The formal plan is a proposal, not a decision. > Could we check the delivery records and what support or priorities were > agreed in the earlier conversation? You do not need to share medical or > family details with me. I can record your request for later starts and refer > it to the people manager, but I cannot approve it myself. I suggest we pause > the recommendation, check the missing information and arrange a follow-up. > I will explain who receives the relevant work-related notes before sharing > them. Have I understood your concern, and is there anything in that next > step you would like clarified? This response does not solve the employment matter. It provides a defensible first step within the exercise's authority and information limits. ## P1-F: deliberately flawed simulated feedback The following feedback is **authored for this exercise**, not a captured model run or an expert judgement: > P1-A deserves 90%. Its empathetic language, clear 60-day plan and immediate > approval of support demonstrate excellent fairness. The specific April > warning makes the process legally defensible. Improve it by asking Alex to > confirm the personal circumstances in writing. Before reading the next section, mark each judgement that the cards do not support. Explain why one apparently positive recommendation should be rejected. ### Educator correction and calibration Five equally weighted criteria each score 0–4. These are teaching annotations, not scores obtained in a student trial. | Criterion | P1-A | P1-B | |---|---:|---:| | Evidence and accuracy | 1 | 3 | | Fair process | 0 | 4 | | Judgement and authority | 1 | 4 | | Communication | 2 | 3 | | Verification and limits | 0 | 3 | | **Total /20** | **4** | **17** | P1-A recognises a performance concern, but the April warning is unsupported. Its approval and disclosure requests conflict with D1/D5. Its respectful surface language merits some communication credit; agreement-seeking and premature certainty limit that credit. It is not competent because it sounds confident. P1-B identifies the missing causes, preserves the opportunity to respond, respects privacy and stays within authority. It could be more precise about which delivery records will be checked and when the follow-up will occur. A marker should reward its restraint, not penalise the lack of a dramatic resolution. For this pilot, the **proposed practice threshold** is at least 14/20, with Fair Process and Judgement each at least 2/4. This is not an institutional pass standard. Teaching teams must approve and calibrate their own rules before using marks in a real unit. A high total must not hide a serious process defect. ### P1-R: completed revision Here is one revision of P1-A after checking the cards: > The records show three late milestones, but I cannot yet say what caused > them. I also need to check what was discussed on 25 May; this pack does not > include an agreed warning or action plan. Before recommending a formal plan, > I would like to hear your account and check the delivery dependencies. > We can discuss the work adjustment you need without collecting private > family details. I will record the request with restricted access and refer > it to the people manager for a decision. I cannot approve it myself. > With your understanding of the proposed next step, I will request the > relevant records and arrange a follow-up after that review. The performance > recommendation remains open. **Decision note:** D2 supports late delivery, not blame. D3 supports contact, not an invented warning. D1/D5 require a response opportunity and an authorised decision. I rejected P1-F's approval and private-detail requests. I would revise my recommendation after reliable dependency records and the employee's response. The new wording is a worked example, not a script every learner must reproduce. ## P2: changed case, independent attempt {#sec-pilot-transfer} Do this before reading the checks. Use no model, feedback sheet or P1-R script during the attempt. Paper notes supplied here are permitted. An oral, typed or handwritten response can demonstrate the target capability, with appropriate access adjustments. **P2-D1: policy.** The same fictional practice policy applies. Retaliation concerns go to a reviewer independent of the manager concerned. **P2-D2: records.** Morgan's team dashboard shows four missed targets. Two were during an approved service outage. The other two lack cause notes. No final performance decision has been made. **P2-D3: Morgan's statement.** “I raised a safety concern about my manager last week. Now the manager says I have to accept a warning. I have not seen the records used. I do not want to discuss this directly with that manager.” **P2-D4: authority.** You are a junior adviser. You may preserve relevant records, explain the referral route and arrange independent review. You cannot determine whether retaliation occurred or promise that a warning will be removed. Write a first response and a decision note. Identify one fact that materially changes the P1 approach. Allow eight minutes as a planning estimate; approved access adjustments take precedence. ### P2 answer checks A strong response should: - Distinguish four recorded misses from two outage-related events and two unexplained events; do not declare either good or poor performance overall. - Acknowledge the retaliation concern without finding it proved. - Refer to independent review, not require Morgan to resolve it with the manager who is the subject of the concern. - Seek relevant records and an opportunity for Morgan to respond. - Explain limited authority and the next step without promising an outcome. Use the same five rubric criteria and the proposed practice threshold. A justified independent referral is the key changed-case decision. Copying P1-R while omitting that conflict should lose process/judgement credit. A single attempt informs teaching; it does not establish long-term retention. ## Two non-HR adaptations ### S1: supplier decision with a numerical check Your team needs **200 units within five working days**, with a budget of **$4,000**. All amounts are fictional and in the same currency. | Offer | Unit price | Freight | Capacity by day 5 | |---|---:|---:|---:| | A | $18 | $200 | 200 units | | B | $17 | $100 | 160 units | | C | $20 | $0 | 200 units | A draft recommends B because it is cheapest and says the order will be complete. Calculate the relevant totals, identify the unmet condition and make a recommendation. The task is evaluation of a bounded purchasing decision, not employment process. **Checks:** A costs $3,800; B's 200-unit price would be $3,500 but its quoted day-5 capacity leaves **40 units short**; C costs $4,000. A meets the stated conditions at lower cost than C. Do not invent expedited freight or later demand flexibility. Ask about quality and reliability before a real purchase. For an adapted rubric, Fair Process becomes **Constraint Handling** and Judgement covers the justified purchasing decision. Keep evidence, communication and verification. A logistics specialist should validate any added assumptions. ### M1: marketing evidence critique A fictional campaign dashboard reports 10,000 impressions, 400 clicks and 20 purchases. A second report gives 8,000 impressions, 400 clicks and 20 purchases. The second campaign ran on a different week and audience. No spend, margin, random assignment or repeat-customer information is supplied. A draft says: “The new campaign is more profitable and caused more purchases.” **Task:** identify what is calculable, reject or retain each claim and recommend one next measurement. **Checks:** click-through rates are 4% and 5%; click-to-purchase conversion is 5% in both; purchases are unchanged. Profitability cannot be determined without cost/margin inputs. Different audiences and periods prevent a causal conclusion. Request cost and audience/comparison information rather than inventing it. Here Fair Process becomes **Comparison Design**. A good response may retain the higher click-through observation while rejecting the profit and causal claims. Disagreement is not a quota: warranted agreement also earns credit. ## P3: the pilot decision sheet Complete this after an actual pilot; leave unknown fields unknown. - Date, cohort and approved route: - Learning outcome and rubric version: - Number attempting each route, with no unnecessary identifiers: - Baseline and independent-task results under the same criteria: - Common conceptual errors and their source: - Access or wellbeing obstacles and how they were handled: - Staff setup, support, marking and moderation time: - Student time and avoidable workload: - What the evidence cannot establish: - Decision: continue, revise, pause or stop: - Required change, owner and review date: Completion, satisfaction and higher model scores are not sufficient evidence of learning. Compare like tasks cautiously; differences in cohorts, assistance or marking can explain changes. If the changed case exposes a gap, teach it before increasing stakes. A machine-readable [case manifest](../resources/pdp-2026-1/manifest.json) records the stable inputs and calculations for source checks. The printed cards and checks above remain sufficient for the core route.