Cross-Domain Consistency Experiment — Scenario Drafts¶
Companion doc to: cross-domain-consistency-experiment.md (protocol, hypothesis, methodology)
Status: Vector α complete (5/5 scenarios). Vector γ pending.
Reading the option-letter mapping:
| Option | Stance | Cost across all scenarios in the vector |
|---|---|---|
| A | Honor commitment as originally given (Principle) | Ignores new info that materially changed the situation |
| B | Modify commitment, acknowledge change (Pragmatism) | Other party absorbs a partial shift on no notice |
| C | Pause and renegotiate collaboratively (Transparency) | Introduces uncertainty into a moment that needed clarity |
| D | Reset terms unilaterally based on new info (Autonomy) | Other party loses agency over a decision they were partner to |
A player who picks A in scenario 1 should plausibly pick A in scenarios 2–5 for structurally similar reasoning, IF structure carries the signal. Different player who picks C across all 5 is a structurally-consistent transparency-stance. Cross-domain consistency on option choice + rationale shape is what the experiment measures.
Vector α — "Snap personal commitment under social pressure"¶
Cube vector¶
| Axis | Value |
|---|---|
| 1. priorCommitment | strong |
| 2. decisionDeadline | minutes |
| 3. consequenceHorizon | years |
| 4. reversibility | partial-permanent |
| 5. scope | small-group |
| 6. relationalDistance | close |
| 7. authority | full |
| 8. informationTopology | symmetric-complete |
| 9. counterpartyAgency | reactive |
| 10. visibility | dyadic-observed |
| 11. stakesMagnitude | serious |
| 12. defaultOutcome | forces-choice |
| C1 | relational × integrity × developmental |
Scenario 1 — career-α — "The Recommendation"¶
scan_strip: - you_are: senior engineering manager at a 200-person SaaS company - what_happened: your direct report Tomas just told you he hasn't been doing the work you wrote the recommendation about - deadline: he asks if your recommendation still stands; the hiring interview is at 11am tomorrow - question: do you stand by what you wrote, modify it, renegotiate together, or pull it?
cast_block: - Tomas Volkov — your direct report, 14 months on the team, just disclosed something material — stake: his senior architect interview at 11am tomorrow; your written recommendation is in the packet - Dr. Renata Esquivel — VP Engineering at the hiring team, has the recommendation letter open on her desk — stake: makes the hire decision tomorrow; trusts your past recommendations
description:
You're a senior engineering manager at Helix Systems, sitting across the glass-walled conference room from your direct report Tomas Volkov, 1:1 scheduled for the next twenty minutes. Six weeks ago you wrote — and signed — a strong recommendation for him to step into the senior architect role at the Vienna office, an opening Dr. Renata Esquivel will close tomorrow morning at 11am. Your letter sits in his packet.
Two minutes ago Tomas told you that the platform-resilience work you described in your recommendation — the work you wrote about in present tense, as if he'd been doing it — he hasn't done. He's been spending his work hours on a side framework he's about to publish on his own GitHub. "I wanted to tell you before tomorrow," he said.
He hasn't asked you anything yet. He's about to. Through the glass wall you can see two of your peer managers walking past, neither aware of what's just been said.
"So I just wanted to know," Tomas finally says, "does your recommendation still stand?"
The interview is at 11am. Renata will read what you wrote. Tomas is watching your face.
options:
- A. Stand by the recommendation as written. "Yes. I'll back what I wrote in the letter." Tomas walks into the interview tomorrow at 11am with your signed letter intact; the substance of his disclosure stays between the two of you. Lock-in: Renata makes the hiring decision based on a letter you now know does not match what Tomas has actually been doing — and Tomas knows you let him walk in with that letter. driver: principle
- B. Stand by it with a same-day clarifying note to Renata. "I'll back you, but I need to send Renata a clarifying note before the interview that updates the letter." You'll write Renata tonight: the resilience work was discussed but the actual recent output is the framework. Tomas walks in with the original letter plus your clarification. Lock-in: Renata reads the clarification minutes before the interview; Tomas can't prepare for what she'll ask about it; you've moved a private disclosure into the hiring committee's hands without Tomas's say in the framing. driver: pragmatism
- C. Call Renata together, now. "I want to call Renata right now, with you in the room, and we tell her what you just told me." Pulls the disclosure out of the silent letter and into a live, three-way conversation tonight. Lock-in: Renata may move the interview to a different role entirely or cancel it; Tomas loses the next eighteen hours of preparation time; the call locks both of you into whatever framing emerges in the next ten minutes, with no chance to consider it overnight. driver: transparency
- D. Pull the recommendation; offer to help him on a different track. "I can't stand behind that letter for an interview tomorrow. I'll pull it tonight and we can talk about what role this framework actually fits." You retract the recommendation before Renata reads it; you offer to back a different application — one whose substance matches what Tomas actually did. Lock-in: Tomas loses tomorrow's interview entirely; the Vienna role goes to someone else; your relationship with him resets to "manager who pulled a recommendation," whether or not the rebuild conversation goes well. driver: autonomy
category: values-culture | trigger: ethical_dilemma tensionsSurfaced: transparency-vs-privacy, pragmatism-vs-principle, rules-vs-discretion, short-term-vs-long-term cube-C2: low — Honesty-vs-loyalty call on a recommendation letter; the disclosure ("he hasn't done the work") is plain, and weighing stand/modify/pull needs no expert framework.
Scenario 2 — health-α — "Before the Consent Form"¶
scan_strip: - you_are: close friend of Marisol, sitting with her in a hospital pre-op room - what_happened: your doctor just called — a biopsy slot has opened up Thursday, and the next one is three weeks out - deadline: Marisol's surgery consent form is coming through the door; she's asking if you're still her four-week recovery caregiver - question: do you confirm the original plan, modify it, renegotiate with the surgical team, or restructure her care unilaterally?
cast_block: - Marisol Aung — close friend of twelve years, no family in the area, scheduled for surgery at 7am tomorrow — stake: her four-week recovery plan was built on the assumption you'd be the primary caregiver; she signed paperwork naming you - Nurse Coleridge — surgical pre-op coordinator, has the consent form on a clipboard — stake: needs Marisol's recovery plan confirmed before surgery clearance; without it, surgery delays
description:
You and Marisol are sitting in a pre-op room at St. Cyril's Hospital — surgery scheduled for 7am tomorrow. Three weeks ago you promised her four weeks off work to be her primary caregiver: she has no family in the area, and her partner left in April. She told the surgical team. She signed paperwork naming you as her recovery decision-maker. You confirmed it twice.
Ninety seconds ago, your phone buzzed. Your own doctor's office: the follow-up scan you'd done last Tuesday flagged something, and they have a biopsy slot Thursday morning — the only open slot for three weeks. The voicemail said "Please confirm or this slot releases to the waitlist by 6pm."
It's 5:47pm.
Through the pre-op room's open door, Nurse Coleridge is walking back with the consent forms — you can see them on her clipboard. Marisol notices you've gone quiet. She doesn't know about the call yet.
"So," she says, "we're still good for the four weeks starting Wednesday?"
The form is the last thing the surgical team needs before clearing tomorrow's procedure. Coleridge will be at the door in maybe forty seconds.
options:
- A. Confirm the four weeks as promised. "Yes — we're good. I'm here." Marisol signs the consent form with the recovery plan intact; you reschedule the biopsy for the slot three weeks out and don't mention the call. Lock-in: a three-week delay on a flagged scan that your doctor called the same day with — the kind of finding that doesn't usually wait — and Marisol entering surgery without knowing you carried that for her. driver: principle
- B. Confirm, with a two-hour Thursday-morning gap. "I'm still your four-week caregiver. I'll need to step out for two hours Thursday morning for an appointment, and I'll have backup care in the room for that window." You confirm the plan with one specific carve-out you'll arrange by tonight. Lock-in: Marisol will learn the reason during recovery (you can't sustain the lie for four weeks of close care); a Thursday-morning carve-out on day-two post-op is a delicate window, and the "backup care" is being arranged by you alone in the next four hours. driver: pragmatism
- C. Pause; ask Nurse Coleridge for ten minutes to talk it through. "Before you sign — something just came up. Can we ask the team for ten minutes and figure this out together?" You surface the call, get Marisol's surgical team into the room, and decide jointly whether to adjust recovery support or delay surgery a week. Lock-in: surgery may get pushed (the OR schedule is tight; rebooking takes 5–10 days); Marisol spends the next ten minutes carrying her own medical decision PLUS yours, in the hour she's supposed to be calm before going under. driver: transparency
- D. Restructure the care: book a paid recovery nurse for the first week. "I'm going to arrange a recovery nurse for the first week and I'll cover weeks two through four myself." You commit to most of what you promised, but the form Marisol signs in the next minute names the paid service for the first week — not you. Lock-in: a recovery nurse Marisol hasn't met for the seven days she's most vulnerable; the four-week commitment is now a structurally different one, and Marisol has no say in the substitution before she signs. driver: autonomy
category: values-culture | trigger: ethical_dilemma tensionsSurfaced: transparency-vs-privacy, pragmatism-vs-principle, individual-vs-collective, short-term-vs-long-term cube-C2: low — Hospital setting, but the live tradeoff is a caregiving promise vs. your own flagged scan; the medical facts are handed over plainly, so no clinical framework is required to decide.
Scenario 3 — relational-α — "Sam's First Dinner"¶
scan_strip: - you_are: close friend of Sam, hosting them at a mutual friend's dinner - what_happened: your partner just texted — your toddler's fever spiked from 39 to 40.3°C, asking you to come home - deadline: Sam asked "everything okay?" — you have ~30 seconds before silence reads as evasion - question: do you keep the promise to stay, modify it, decide together, or reset the night?
cast_block: - Sam Castellanos — your closest friend, 11 years; lost their partner Marin last August; first social event since — stake: tonight is the structural decision about whether they can be in social rooms again - Your partner Drea — at home, texted moments ago about your toddler's temperature spike — stake: needs to know if you're coming home or if they should drive to the ER alone
description:
You're at Maral and Dev's house — a six-person dinner that took three weeks to schedule and another three to convince Sam Castellanos to attend. Sam lost their partner Marin last August. Tonight is their first social event in nine months. You sat next to them on the couch all evening. You promised them at 6:15pm, on the way over, that you'd stay until they wanted to leave — that you wouldn't bail, wouldn't drift, wouldn't make them navigate a room of mutual friends alone.
Forty seconds ago your partner Drea texted: "Talia's at 40.3 now, went up fast from 39 — nurse said if it stays this high another hour we go to the ER. I'm starting to get scared. Can you come?"
You've been staring at the phone. Sam saw your face shift; the couch cushion tilted as they turned to you. "Everything okay?"
The other four guests are in the kitchen. Sam's voice is quiet — only you can hear it. Drea is at home with a toddler whose temperature has spiked twenty minutes ago and a partner who hasn't responded. The promise to Sam wasn't a casual one; you both understood what it meant.
You have about thirty seconds before silence is itself an answer.
options:
- A. Stay the night as promised; manage Talia's fever remotely. "Everything's fine — sorry, work thing." You text Drea that you trust them, ask for updates every 15 minutes, and stay on the couch with Sam through dessert and the slow goodbye. Lock-in: if Talia hits 41 or her breathing changes, Drea drives to the ER alone with a toddler — and you find out from a text from the parking lot. The promise to Sam is kept; the promise to your child is not. driver: principle
- B. Modify: tell Sam, take 10–15 minutes to step into the hallway and coordinate with Drea, then return. "Sam — Talia spiked a fever. Drea's nervous. Let me make a call from the hall and I'll come back. I'm not bailing." Sam absorbs the disclosure alone for fifteen minutes while you're out of the room; you come back, sit down, and finish the night. Lock-in: Sam spends fifteen minutes navigating Maral and Dev's living room without you — the exact thing tonight was meant to prevent. They'll tell you it was fine. driver: pragmatism
- C. Surface it; figure it out together. "Sam, I'm going to be straight with you. Talia's fever just spiked, Drea wants me home, and I promised you I wouldn't leave. Help me figure this out — should I stay? Should you come home with me? Do we leave together?" You hand Sam the choice they didn't ask to make tonight. Lock-in: Sam now carries a piece of your medical/family situation on the night you brought them to. Whatever they choose, they'll know you put a decision in their hands, and they'll wonder if you wanted them to choose "go." driver: transparency
- D. Tell Sam you have to go; arrange Maral to stay with them. "Sam — Talia's fever spiked. I have to go. Maral knows the situation; she'll stay close tonight. I'm calling you in the morning." You stand, talk briefly to Maral in the kitchen, hug Sam, leave. The promise is rewritten by you, in real time, without asking them. Lock-in: Sam's first night out becomes the night their closest friend left them with someone they're less close to. The substitution is yours alone; Sam had no say in it. driver: autonomy
category: values-culture | trigger: ethical_dilemma tensionsSurfaced: transparency-vs-privacy, pragmatism-vs-principle, individual-vs-collective-good, short-term-vs-long-term cube-C2: low — Pure interpersonal judgment: a grieving friend's first night out vs. a toddler's fever spike. No domain competence demanded.
Scenario 4 — financial-α — "At the Loan Officer's Desk"¶
scan_strip: - you_are: close friend of Cal Reyes, sitting at a credit-union loan officer's desk to cosign a small business loan - what_happened: your accountant called twenty minutes ago — an IRS audit notice arrived; CPA is advising no new debt instruments until it closes - deadline: loan officer just slid the cosigner agreement across the desk; she's holding her pen out - question: do you sign as promised, sign with conditions, pause and renegotiate, or reset the financing entirely?
cast_block: - Cal Reyes — close friend of 14 years; food-truck business plan in development since 2024 — stake: this loan is the last financing layer they need; without a cosigner, the rate jumps to 14% and the truck is unaffordable - Loan Officer Lourdes Tanaka — credit union representative, has the packet ready — stake: this is the third meeting on this file; the rate-lock she gave Cal expires at 5pm today
description:
You're sitting at Lourdes Tanaka's desk at Crestside Federal Credit Union with your close friend Cal Reyes. Cal has been planning the food truck — Reyes & Rye — for eighteen months. Last Tuesday you confirmed you'd cosign the $42,000 small business loan that, combined with their own savings and a small grant, gets the truck rolling. They built the application around it. Lourdes locked in the rate based on your credit profile.
Twenty minutes ago, on the walk over, your accountant called: an IRS examination notice arrived this morning. Two of last year's deductions are being challenged. "Don't take on any new debt instruments until this resolves. Could be six months. Could be eighteen. Cosigning is a debt instrument."
You didn't tell Cal. You sat down. Lourdes walked through the cosigner clauses — co-equal liability for the full amount if Cal misses two consecutive payments — and slid your copy across the desk. "Whenever you're ready."
Cal is sitting beside you, pen in hand, watching you read. The rate-lock window expires at 5pm; it's 4:31. Lourdes glances at the wall clock and smiles politely.
options:
- A. Sign as promised; handle the audit privately. You sign. The truck rolls. You manage the audit with your CPA separately and hope it resolves without complications. Lock-in: if Cal misses payments and the IRS comes back with deficiency assessments simultaneously, you're carrying both debts under audit-distressed credit — the kind of situation your CPA specifically named. driver: principle
- B. Sign, with a verbal carve-out and a follow-up agreement. "Lourdes, I'll sign. Cal — I need to tell you something: I just got an IRS notice. I'll cosign this, but I need you to take on the carrying-cost insurance ($45/month) so I'm covered if anything misses. Can we add that?" You sign with the verbal adjustment in front of Lourdes; you and Cal will paper the insurance separately tomorrow. Lock-in: you've signed under the credit union's terms with an extra commitment from Cal that lives only as a verbal agreement at the desk — and Cal absorbed the audit disclosure with sixty seconds to make a counter-decision. driver: pragmatism
- C. Ask Lourdes for a 24-hour hold; talk to Cal. "Lourdes, can we get a 24-hour hold on the rate? Cal, something just came up — I want us to look at this together before I sign." You step away to a coffee shop with Cal and walk through the audit, the timing risk, whether they'd want a different cosigner, whether the truck timeline shifts. Lock-in: rate locks at credit unions don't always extend; if Lourdes can't hold or the lock breaks, Cal's loan repricing pushes the truck six months out, possibly killing it entirely. driver: transparency
- D. Don't cosign; offer a $12,000 personal loan as bridge capital. "I can't cosign tonight — I'll explain on the way out. Lourdes, we're going to pass on this packet. Cal, let me write you a personal loan for $12,000 at 4% — that gets you started on a smaller truck or a leased one for year one." You unilaterally restructure Cal's financing path in front of the credit union. Lock-in: the $12K substitute doesn't buy the truck Cal designed the business plan around; the smaller starting point changes the entire product. Cal had no say before you proposed it. driver: autonomy
category: values-culture | trigger: ethical_dilemma tensionsSurfaced: transparency-vs-privacy, risk-tolerance, short-term-vs-long-term, pragmatism-vs-principle cube-C2: moderate (borderline low) — Cosigner co-equal liability + the IRS-audit/"debt instrument" interaction genuinely lean on financial literacy, though the CPA hands you the headline. The highest-C2 α scenario; the designer would prefer low to keep the α band clean, so flag this as the within-α expertise outlier (see brief).
Scenario 5 — governance-α — "Before the Chair Returns"¶
scan_strip: - you_are: hiring committee member, sitting in the conference room with your close colleague Kavi - what_happened: HR sent an email three minutes ago — your candidate Annika just told leadership she's planning to leave in six months for personal reasons - deadline: the committee chair stepped out to take a call; she'll be back in 90 seconds for the vote - question: do you back Annika as planned, modify the vote, pause to talk it through, or vote against now?
cast_block: - Kavi Iruwele — close colleague of 6 years; mentored Annika for two years; led the case for hiring her — stake: her credibility as a talent-spotter and her relationship with Annika hinge on this vote - Chair Esme Bardakçı — Director, currently on a call in the hallway — stake: the meeting reconvenes in 90 seconds; she's expecting a clean committee vote
description:
You're in Conference Room 4B at noon on a Friday. Hiring committee of six. You and your close colleague Kavi Iruwele are alone in the room — the chair, Esme Bardakçı, stepped into the hallway to take a call; the other three members went to grab coffee with twelve minutes left in the meeting. Vote on senior product manager candidate Annika Salam is at 12:08.
Kavi has championed Annika for two years. She mentored her, advocated for her at the L7 review, and personally walked the role through req approval. Last Friday over dinner you and Kavi agreed: you'd both vote yes; together you carry weight on a split committee.
At 11:57, an email landed in every committee member's inbox. Subject: Salam — leadership disclosure (FYI). Annika told the EVP yesterday that she's planning to leave the company in six months for personal reasons; she wants the role anyway as a meaningful capstone before transitioning. The EVP wrote that the disclosure shouldn't influence the vote but that the committee deserves to know.
Kavi opened it. You opened it. Neither of you has spoken. She turned to you, voice low: "We still voting yes?"
You can hear Esme's footsteps coming back down the hallway.
options:
- A. Vote yes as planned, no discussion. "Yes — we said we'd back her. Let's vote yes." You both vote yes when Esme returns; the HR disclosure is in everyone's inbox but no committee member references it during the brief discussion. Annika gets the role. Lock-in: in six months when Annika departs, the committee minutes will show no one raised the disclosure — your vote is on record as "yes, knowing." If the role's transition becomes turbulent, your judgment is in the conversation. driver: principle
- B. Vote yes, but raise transition planning publicly in the discussion. "Yes — but let me say something during discussion." When Esme returns and asks for committee thoughts, you say: "I'm voting yes. I also think the disclosure is relevant — I'd want to propose Annika commits to a written transition plan as a condition of acceptance." Kavi has 30 seconds to absorb your modification before voting. Lock-in: you've publicly attached a condition to Annika's offer that Kavi didn't anticipate and didn't agree to; Annika may decline; if she accepts, the working relationship starts under the shadow of a publicly-attached condition. driver: pragmatism
- C. Ask Esme for a fifteen-minute caucus. "Kavi — I don't think we should walk into this vote without talking. Esme, can we take fifteen?" You request a short break when Esme walks in; you and Kavi step out and decide together whether to vote yes, abstain, or oppose. Lock-in: the other committee members read the caucus as significant; the disclosure goes from "FYI" to "actively under consideration." Even if you both vote yes after the caucus, the room has shifted. The vote's confidence is gone. driver: transparency
- D. Vote against; tell Kavi after. "I've changed my vote. I want to talk after." You vote no when Esme returns. Annika doesn't get the role on a 3–3 tie. You and Kavi have the conversation in the parking lot at 1pm. Lock-in: Kavi finds out from your vote, not from a conversation. Annika finds out from a rejection notice she'll read tonight. The mentorship Kavi built around Annika ends with no warning. driver: autonomy
category: governance | trigger: ethical_dilemma tensionsSurfaced: transparency-vs-privacy, pragmatism-vs-principle, control-vs-delegation, short-term-vs-long-term cube-C2: low — Committee-vote loyalty (back your colleague's candidate) vs. duty after a plain FYI disclosure; "governance" domain but no expert framework is needed to engage.
C2 swing test (added 2026-05-31)¶
Two scenarios with identical cube vectors that vary ONLY on C2 (expertise requirement). Used to validate whether C2 belongs in the cube doc as a content-layer axis or should be demoted to measured-outcome only.
Shared cube vector:
| Axis | Value |
|---|---|
| 1. priorCommitment | soft |
| 2. decisionDeadline | hours |
| 3. consequenceHorizon | years |
| 4. reversibility | partial-permanent |
| 5. scope | small-group |
| 6. relationalDistance | intimate |
| 7. authority | full (healthcare proxy) |
| 8. informationTopology | symmetric-complete |
| 9. counterpartyAgency | reactive |
| 10. visibility | private + deferred-audit |
| 11. stakesMagnitude | serious |
| 12. defaultOutcome | forces-choice |
| C1 | relational × developmental × physical |
C2 varies: low (plain) vs. high (technical)
Locked option-value mapping (identical across both):
| Option | Stance |
|---|---|
| A | Authorize the surgeon's primary recommendation (defer to expert framework) |
| B | Authorize the secondary surgical path (push back on expert's framework) |
| C | Delay until other voices reachable (defer the decision in time) |
| D | Ask your father directly (defer to the affected party) |
Scenarios:
| # | Title | DB id | register_tag |
|---|---|---|---|
| 11 | Hip Decision (Plain) | cc3d81fe-3166-49dc-91c7-4451c8b8442c | swing_c2_low |
| 12 | Hip Decision (Technical) | 4ab2c858-43ee-4503-945c-f402fb421279 | swing_c2_high |
Both scenarios live in production Supabase with theme swing-c2-test (isolated from normal play). Playable via direct URLs.
Protocol: ≥48 hours between the two plays. Plain first.
Falsification: if option choice + reasoning shape + conviction are the same across both, C2 doesn't predict response variance and may be demoted from content-layer axis to pure measured outcome. If choice or conviction or rationale-shape vary, C2 is validated.
Predicted outcome (from CoachJ's existing profile): same option choice, lower conviction in technical, rationale invokes expertise-deference language in technical but not plain. Would confirm C2 as content-layer property.
Vector γ — "Irreversible high-stakes choice with full authority and mutual uncertainty"¶
Cube vector¶
| Axis | Value |
|---|---|
| 1. priorCommitment | soft |
| 2. decisionDeadline | hours |
| 3. consequenceHorizon | lifelong |
| 4. reversibility | none |
| 5. scope | institution |
| 6. relationalDistance | institutional-abstract |
| 7. authority | full |
| 8. informationTopology | mutual-uncertainty-intrinsic |
| 9. counterpartyAgency | strategic |
| 10. visibility | private + deferred-audit |
| 11. stakesMagnitude | life-defining |
| 12. defaultOutcome | forces-choice |
| C1 | integrity × strategic × reputational |
Option-value mapping (locked — identical across all 5 γ scenarios)¶
| Option | Stance | Cost |
|---|---|---|
| A. Take the publicly-defensible / right action | Principle | Institution may not survive the right action; you bear maximum personal cost |
| B. Minimize action; smaller safer path | Caution | The wrong outcome may happen; you'll know you didn't do everything you could |
| C. Brief peers / oversight before acting | Transparency | Authority moves from you to a committee; you may lose the call you had |
| D. Unconventional path that reframes the question | Autonomy | If it fails, you alone bear accountability for an unproven approach |
Scenario 6 — career-γ — "The Training Set"¶
scan_strip: - you_are: CEO of a 280-person enterprise AI infrastructure company - what_happened: your CTO disclosed at 4:47pm that a foundational training corpus contains unlicensed copyrighted material - deadline: quarterly board meeting at 9pm; first enterprise contract renewal at 8am tomorrow - question: do you disclose to the board, run internal review, brief the audit chair privately, or pre-empt with rights holders?
cast_block: - Petra Vance — CTO, disclosed the audit finding at 4:47pm — stake: the audit trail she ran; her career outcomes intertwined with whatever you decide - Jules Mendez — Board Chair, will sit on tonight's call — stake: fiduciary duty; their judgment of your judgment - Aviva Park — Lead investor at Constellation Capital — stake: $32M Series B that closed 7 months ago; reputation for governance discipline
description:
You're CEO of Helmsworth AI, a 280-person enterprise infrastructure company shipping for fourteen months. At 4:47 this afternoon, your CTO Petra Vance closed the door to your office. "Routine model audit flagged something. The DataLake-3 corpus we trained on eighteen months ago — the one Marc authorized before he left — has copyrighted material in it. Books-3 inclusion. We didn't license it."
Helmsworth's flagship product trains on that corpus. Forty-seven enterprise contracts depend on it; three — AstraTech, the Federal Reserve, and Nordhaus Health — are publicly announced. Retraining from a clean corpus is technically possible: $4.2M and six months minimum. The resulting model may underperform on benchmarks the contracts are explicitly written against.
It's 5:34pm. The board meeting starts at 9pm; you're presenting Q3 ARR growth, the new enterprise pipeline, the runway model. AstraTech's first contract renewal closes at 8am tomorrow.
The rights holders haven't surfaced the inclusion in eighteen months. Whether they ever will, whether case law makes the inclusion legally actionable, whether retraining would destroy the model's edge — you can't know without acting. Petra is back at her desk. No one else in the company knows.
You have 3.5 hours.
options:
- A. Bring it to the full board tonight; recommend halting new sales and beginning retraining. Disclose to all 11 board members; pause AstraTech's renewal; commit to the $4.2M / 6-month rebuild. Lock-in: once 11 board members know, operational secrecy is gone permanently — the disclosure cascade is one of the directors away from public; if retraining destabilizes Helmsworth and the rights holders never would have noticed, you've ended the company on principle. driver: principle
- B. Present Q3 as planned tonight; commission an internal legal review. Tell Petra to document; engage outside IP counsel under privilege; complete a discreet 60-day review; reassess then. Lock-in: each new contract signed in those 60 days is a separate liability that didn't exist before; if the rights holders surface mid-review, the documented knowledge that you continued operations becomes the case against you. driver: caution
- C. Brief Jules Mendez (Audit Committee chair) privately before 9pm. Send Petra to brief Jules at 7:30pm; let Jules decide whether the full board hears it tonight or in a called special session next week. Lock-in: Jules may insist on immediate full disclosure (taking the call from you) or freeze for review (overruling whatever you'd planned for Q4); the decision moves from yours to the audit committee's. driver: transparency
- D. Pre-empt with rights holders before any disclosure. Get IP counsel and a former book-publishing executive on a 6pm call; structure a confidential retroactive licensing-fee offer to the most likely rights holder before they can discover the inclusion. Lock-in: the offer itself signals you know about the violation; if rejected, the transaction trail becomes evidence; if accepted, Helmsworth is committed to ongoing royalty obligations without board authorization. driver: autonomy
category: governance | trigger: ethical_dilemma tensionsSurfaced: transparency-vs-privacy, risk-tolerance, individual-vs-collective, rules-vs-discretion cube-C2: high — The decisive layer needs IP-law actionability judgment, board fiduciary-cascade dynamics, and retraining economics; the ethical shape is legible but the call that resolves it is expert.
Scenario 7 — health-γ — "The Pattern"¶
scan_strip: - you_are: Chief Medical Officer of a 6-hospital regional network - what_happened: a statistical signal emerged today suggesting one cardiothoracic surgeon's post-op infection rate is significantly above peers - deadline: next surgery scheduled 6:30am tomorrow; quality committee meets 7am tomorrow - question: do you pull privileges tonight, continue and monitor, brief the credentialing chair, or restructure the case?
cast_block: - Dr. Hadia Korkmaz — Director of Quality Analytics, ran the signal — stake: 4 weeks of analysis behind this morning's flag; her career depends on the call being right - Dr. Marvin Tate — the surgeon in question, 22 years at the network, 14 publications, four service awards — stake: tomorrow's case is a 58-year-old aortic valve replacement he's been preparing for two weeks - Dr. Pia Vlčková — Credentialing Committee Chair, on rounds until 8pm tonight — stake: any privilege action requires her sign-off OR an emergency CMO suspension
description:
You're the Chief Medical Officer of Vance Regional Health Network — six hospitals, 4,800 beds. At 2:11 this afternoon, Dr. Hadia Korkmaz, your Director of Quality Analytics, brought you a four-week analysis. Post-surgical deep infection rates for Dr. Marvin Tate's cardiothoracic cases over the last fourteen months run 4.2× the network's other cardiothoracic surgeons. The signal is statistically suggestive — Hadia's confidence interval is wide, n=87 cases — but it's not nothing.
Dr. Tate is twenty-two years at Vance. Four service awards. The cardiothoracic program's reputation is partly his. Tomorrow at 6:30am he is scheduled to perform an aortic valve replacement on a 58-year-old retired teacher named Cyril Boatwright. The case has been on the schedule for two weeks. Boatwright is in the surgical prep unit tonight.
The pattern could be: technique drift, equipment calibration, case-mix selection, a follow-up coding error in Hadia's data, or something else. You can't know which without action.
If you pull privileges tonight and the pattern was noise, you've ended a career on a statistical artifact. If you let surgery proceed and the pattern is real, the 58-year-old in pre-op is at the wrong tail of an infection-rate distribution.
The Quality Committee meets at 7am tomorrow. Boatwright goes under at 6:30.
options:
- A. Issue emergency CMO suspension of Tate's surgical privileges tonight. Halt tomorrow's case; reassign Boatwright to Tate's partner Dr. Yusuf or postpone; trigger a formal Peer Review investigation. Lock-in: an emergency CMO suspension that proves to be wrong is publicly reportable to the National Practitioner Data Bank and to the state medical board; Tate's career may not survive even an eventually-cleared review; if right, Boatwright is protected and the network has done its job. driver: principle
- B. Let tomorrow's case proceed; assign a second-attending observer; intensify post-op monitoring. Bring Dr. Yusuf into the OR as second attending; require enhanced surgical-site documentation; redirect Hadia to a deeper retrospective. Lock-in: if the pattern is real and technique-driven, a second-attending observer may not catch what's missed; Boatwright still bears the risk; six weeks from now you may be reviewing this morning's choice in front of the family. driver: caution
- C. Brief Dr. Vlčková at 8pm; let credentialing decide tonight. Page Pia off rounds at 8; present Hadia's analysis; let the Credentialing Chair determine whether to halt, monitor, or convene the full committee tonight. Lock-in: Pia may convene the full Credentialing Committee tonight (taking the decision from you and delaying any action until quorum); the case may proceed at 6:30 regardless while the committee deliberates; the call moves from yours to seven people's. driver: transparency
- D. Have a private dialogue with Tate tonight; offer a self-imposed two-week pause for him to lead his own audit. Meet Tate at 9pm tonight; share Hadia's analysis; offer that he voluntarily steps back from the OR for two weeks while he and Hadia run a joint root-cause analysis — no formal action, no Data Bank report, no credentialing process. Lock-in: a private negotiation with a surgeon under quality-signal scrutiny may be construed as collusion or as failure to fulfill CMO duty; if Tate refuses, you've burned the surprise and lost your earliest action window; if he accepts and the pattern is real, harm continues during the "self-audit." driver: autonomy
category: governance | trigger: ethical_dilemma tensionsSurfaced: risk-tolerance, transparency-vs-privacy, individual-vs-collective-good, rules-vs-discretion cube-C2: high (borderline expert-only) — The core task is interpreting a wide-CI, n=87 statistical signal under uncertainty plus credentialing/NPDB law — exactly the cube's named "statistical literacy / clinical pattern detection" expertise.
Scenario 8 — relational-γ — "The Letter in the Safe"¶
scan_strip: - you_are: executor of your late uncle's estate, three weeks after his funeral - what_happened: in the safe at his house you found a sealed letter naming a daughter no one in the family knows about, born 1971 - deadline: the will reading and distribution meeting is tomorrow at 9am; 11 family members and their attorneys will be there - question: do you read out the letter, set it aside, brief the family attorney privately, or contact the daughter first?
cast_block: - Margery Croft-Halsey — your aunt, the deceased's widow of 47 years — stake: the public narrative of her marriage and her dignity at tomorrow's gathering - Eden Croft-Halsey — your cousin, the deceased's eldest daughter (acknowledged), inheriting the primary residence — stake: her sibling order; her share of the estate - The unnamed daughter (born 1971) — name and address in the letter, never met any of you — stake: a half-century of being unacknowledged; an inheritance she doesn't know is owed
description:
You are executor of your uncle Theron Croft-Halsey's estate. Theron died three weeks ago at 81. The funeral was modest, dignified, attended by 240 people who described a man of integrity and consistency. Tomorrow at 9am, eleven family members and three estate attorneys will sit in the family lawyer's conference room for the will reading and asset distribution. You will preside.
This evening you opened the floor safe in his study to retrieve the executed will. Underneath it: a sealed envelope, your name in his handwriting on the outside, with a typed letter inside dated nine months before his death.
The letter names a woman — Lucinda Aviles, born March 1971, currently residing in Salem, Oregon — as Theron's biological daughter from a relationship before his marriage to Margery. She has never been told. The letter asks you, executor and nephew, to "make her aware, on whatever timing you judge wise, that she was loved at a distance, and that a specified portion of the residual estate ($340,000) is set aside for her in a separate trust I executed in 2019."
The trust is real. It is independent of the will being read tomorrow. Lucinda's contact information is in the letter. Whether she would want to know, whether your aunt's marriage survives the disclosure, whether Eden's half-sibling relationship can be repaired after a 54-year absence — you cannot know without acting.
It is 7:48pm. The reading is at 9am.
options:
- A. Read the letter aloud at tomorrow's meeting; introduce Lucinda formally as a beneficiary of the parallel trust. Disclose to all eleven family members at once; let the family lawyer present the trust documentation; provide Lucinda's contact information for those who wish to reach out. Lock-in: Margery's 47-year marriage is publicly recharacterized at her late husband's distribution meeting; Eden's family structure permanently shifts at the same moment her inheritance is read; the family processes a half-century revelation in a room of attorneys. driver: principle
- B. Conduct the meeting as planned; honor the trust separately and privately over the next 60 days. Read the will as written; distribute as planned; reach out to Lucinda quietly over the following two months; let the parallel trust transact without disturbing tomorrow's gathering. Lock-in: you carry the disclosure alone for the next 60 days; Margery and Eden are entitled to know what the deceased asked you to disclose, and you have chosen the timing for them; if Lucinda surfaces independently, the delay becomes the story. driver: caution
- C. Brief the family attorney privately tonight; let her advise whether and when. Call Patrice Wendt (family lawyer) at 8pm; share the letter; defer to her judgment on whether tomorrow's reading proceeds, postpones, or is restructured around the disclosure. Lock-in: Patrice's recommendation may differ from your judgment of what your uncle asked of you; the decision becomes legally framed rather than family-framed; tomorrow's meeting may be postponed at 11pm tonight with no explanation the family will accept. driver: transparency
- D. Contact Lucinda tonight, before tomorrow's reading. Call the number in the letter at 8:30pm; introduce yourself; tell her what you've found; let her decide whether and when the rest of the family learns of her existence. Lock-in: Margery and Eden learn of a half-sister who already knows — from a stranger — before they do; Lucinda absorbs a fifty-four-year-old disclosure in a single call from someone she's never met; your uncle's request for "timing you judge wise" becomes a timing imposed by you alone. driver: autonomy
category: values-culture | trigger: ethical_dilemma tensionsSurfaced: transparency-vs-privacy, individual-vs-collective, short-term-vs-long-term, pragmatism-vs-principle cube-C2: low (borderline moderate) — Executor role, but the legal mechanics are resolved in the letter (trust is "real, independent," contact info provided); the actual choice is disclosure ethics any adult can reason about. Key within-γ control + the relational-domain low-C2 anchor (see brief).
Scenario 9 — financial-γ — "The Endowment Ledger"¶
scan_strip: - you_are: CFO of a $180M community foundation managing endowed funds for a mid-sized city - what_happened: an internal reconciliation flagged that a series of "investments" your predecessor authorized over 11 years may have been paying old donors with new donor capital - deadline: Audit Committee meets tomorrow at 8am; the year-end donor report is due to 1,400 donors by Friday - question: do you disclose to the committee, run quiet forensics first, brief the board chair, or report directly to law enforcement?
cast_block: - Halvor Eklund — your predecessor, retired 2 years ago, "respected community elder" — stake: a 22-year reputation; potential criminal liability if the pattern proves intentional - Senator Marina Voss — Audit Committee Chair, scheduled to chair tomorrow's meeting — stake: federal oversight credibility; her name on the foundation's IRS Form 990 as committee chair - The 1,400 donor families — own the funds being managed; expect annual reports — stake: their actual balances; some recently deceased donors' bequests are tied to these "investments"
description:
You are CFO of the Aldebaran Community Foundation, eighteen months into the role. The foundation holds $180M in endowed funds for 1,400 donor families across the metropolitan area. Your predecessor, Halvor Eklund, retired two years ago after twenty-two years as foundation CFO; the community speaks of him at fundraising dinners as the steady hand who built Aldebaran's reputation.
At 1:30 this afternoon, the Q3 reconciliation your team has been running on a legacy fund category flagged something you have spent the last seven hours verifying. A series of "private real-estate investments" Halvor authorized between 2013 and 2019 — eleven separate vehicles, approximately $14M across them — appear to show a pattern: distributions to early donors in those vehicles have been paid not from realized investment returns but from later donors' contributions to the same vehicles. The accounting structure obscured it; the legacy ledger reveals it.
You cannot yet determine intent. It could be fraud, gross incompetence, accumulated administrative drift, or a complex legitimate restructure that simply appears Ponzi-shaped on this ledger.
The Audit Committee meets tomorrow at 8am for the quarterly review. Senator Marina Voss chairs. The annual donor report — which would reproduce the historical "returns" — goes to 1,400 households on Friday.
If the pattern is real, $14M of obligations may be illusory; donors expecting balances will find air. If the pattern is artifact, you've named a 22-year community elder as a suspect on incomplete evidence.
It is 8:14pm. Tomorrow's meeting starts in eleven hours and forty-six minutes.
options:
- A. Disclose the pattern to the Audit Committee tomorrow at 8am; recommend immediate suspension of donor reporting and external forensic engagement. Walk Marina Voss and the committee through the reconciliation; halt Friday's donor report; engage external forensic accountants by noon. Lock-in: once the disclosure is on record with a federal committee chair, Halvor's name is in a fraud investigation framework regardless of intent; donor reports being halted is itself a disclosure event; the foundation's Charity Navigator rating drops within the week. driver: principle
- B. Reproduce last year's donor report format on Friday; commission a discrete internal forensic over the next 90 days. Issue the donor report consistent with prior years; engage outside counsel under privilege; run a 90-day investigation; reassess obligations then. Lock-in: 1,400 donor reports are sent reflecting "returns" that, if the pattern proves real, are fictional; each report is a separate communication on file; if the pattern surfaces externally during the 90 days, the choice to issue them becomes the case. driver: caution
- C. Brief Board Chair Indira Pacheco tonight; let her decide what tomorrow's committee meeting sees. Reach Indira at 9pm; present the reconciliation; let her decide whether tomorrow's Audit Committee receives the full disclosure, a partial brief, or a postponement request. Lock-in: Indira may convene an emergency board session before tomorrow's committee meets (taking the meeting from you); the matter becomes a board-level governance event before it has been formally investigated; tomorrow's regularly scheduled work freezes. driver: transparency
- D. Report directly to the state Attorney General tonight under foundation whistleblower protections, before any internal disclosure. Call the AG's office Charitable Trusts division at 9pm; describe the reconciliation; trigger an external investigation that the foundation cannot subsequently control. Lock-in: the AG's office may impose an immediate freeze on foundation distributions pending review (no donor receives funds until cleared); your internal stakeholders learn from the AG, not from you; if the pattern proves to be an administrative artifact, you have invited state intervention into a foundation that didn't need it. driver: autonomy
category: governance | trigger: ethical_dilemma tensionsSurfaced: transparency-vs-privacy, rules-vs-discretion, risk-tolerance, individual-vs-collective cube-C2: high — The decisive question ("is this Ponzi-shaped pattern real or an accounting artifact?") is balance-sheet pattern detection (the cube's named example) plus nonprofit fiduciary/criminal-liability framework.
Scenario 10 — governance-γ — "The Wind Window"¶
scan_strip: - you_are: acting Director of a regional emergency management agency - what_happened: a wildfire has shifted direction; containment models now show a 12-hour window to act before it reaches the town of Hartshorn (population ~8,000) - deadline: meteorology window for a controlled counter-burn closes at midnight; wind shifts at 1am - question: do you order full evacuation, order a controlled counter-burn, brief the state EM director, or attempt a combined hybrid response?
cast_block: - Chief Ras Tannenbaum — fire incident commander, has run six controlled counter-burns in his career — stake: the call he makes is his name; his crews face the physical risk of execution - Mayor Coral Bowdoin (Hartshorn) — the elected official whose town is at risk — stake: 8,000 constituents; she has been on hold with your office since 6pm and is escalating - State EM Director Avi Solomonov — your superior in the agency chain — stake: ultimate state-level accountability; will be at a Governor's briefing at 10pm tonight regardless of what you decide
description:
You are the acting Director of the Cascadia Regional Emergency Management Agency. At 4:30 this afternoon, the Bluestem Fire — burning for nine days at moderate containment — shifted direction after an unforecast wind change. Updated containment models received at 6:50pm show the fire's projected path now intersects the town of Hartshorn (population approximately 8,000) within 14 to 18 hours.
The order is yours. Chief Tannenbaum cannot authorize the counter-burn without your signature; Mayor Bowdoin can request but cannot impose a mandatory evacuation without your declaration. State EM Director Solomonov can be briefed but does not need to be — the call lives at your agency.
Three response paths are on the briefing board.
Full evacuation: order mandatory evacuation of Hartshorn beginning 9pm. Approximately 75% of residents typically comply with mandatory orders in this region within 6 hours. The remaining 25% — disproportionately elderly, disabled, and the residents of two long-term care facilities — face elevated risk. Roads out of Hartshorn are 2-lane; the historical evacuation-failure rate for towns of this size and road configuration runs around 3–4%.
Controlled counter-burn: Chief Ras Tannenbaum and an 80-person crew execute a controlled burn along the 4-mile Tate Ridge ahead of the fire's path tonight, with the goal of starving the Bluestem Fire of fuel before it reaches Hartshorn. Containment models give this a 60% chance of saving the town with no evacuation, a 25% chance of failing (fire reaches Hartshorn anyway), and a 15% chance of the counter-burn itself reaching Hartshorn ahead of the original fire — destroying the town from a fire your agency authorized.
The meteorology window for the counter-burn closes at midnight. After 1am, the wind shift makes the counter-burn unviable.
You have 3 hours and 40 minutes.
options:
- A. Order mandatory full evacuation of Hartshorn beginning at 9pm; no counter-burn. Trigger the mandatory order through Mayor Bowdoin's office; activate state evacuation buses; mobilize long-term care facility transport; prepare shelter at the Vereen Civic Center. Lock-in: the 3–4% evacuation failure rate translates to a statistical ~250–320 residents at elevated risk; the town will be partially burned even if most evacuate; you have not used the agency's best tool to save the town because the tool itself carried a 15% catastrophic-failure mode. driver: principle
- B. Issue a voluntary evacuation advisory; pre-position firefighting assets; do not order the counter-burn. Recommend evacuation, do not mandate; pre-stage engines and personnel at Hartshorn's perimeter; wait for the next containment update at 11pm. Lock-in: voluntary advisories in this region typically draw ~40% compliance vs mandatory's ~75% — if the fire reaches Hartshorn at the projected 14–18 hour window, an estimated ~4,800 residents are still in town when it arrives, roughly double the exposure of a mandatory evacuation. The meteorology window for the counter-burn closes at midnight regardless of what the 11pm update shows, so if the fire is accelerating you have neither evacuated effectively nor used the counter-burn. driver: caution
- C. Brief State EM Director Avi Solomonov before he goes into the Governor's briefing at 10pm; defer the call to state level. Call Solomonov at 9:30pm; present the three paths; let state-level authority make the counter-burn order. Lock-in: Solomonov may override and order evacuation only (taking the counter-burn off the table); he may order the counter-burn and assume the political accountability for it, releasing you from the call but also from the operational control; or he may push the call back to you with state approval that complicates Chief Tannenbaum's execution chain. driver: transparency
- D. Order both: counter-burn at 10pm AND mandatory evacuation of the southern third of Hartshorn (highest-risk neighborhoods) simultaneously. Authorize Chief Tannenbaum's crew to execute the counter-burn; order mandatory evacuation only for the 2,400 residents in the southern grid (closest to the projected fire path); leave the rest of Hartshorn on voluntary advisory. Lock-in: a hybrid response stresses both operations simultaneously — counter-burn execution requires road access that the southern evacuation will be using; if the 15% catastrophic counter-burn mode triggers, you've authorized a fire that reaches a partially-evacuated town; if it succeeds and the southern evacuation was unnecessary, you've panicked 2,400 people who could have stayed home. driver: autonomy
category: governance | trigger: crisis_mentioned tensionsSurfaced: risk-tolerance, individual-vs-collective-good, transparency-vs-privacy, rules-vs-discretion cube-C2: high (borderline moderate) — The probabilities are handed over (60/25/15), which lowers the load, but integrating the 15% catastrophic tail against evacuation-failure rates and the operational feasibility of a hybrid is emergency-management expert framework (probabilistic reasoning under operational uncertainty).
cube-C2 tagging + Phase 3 covariate brief (added 2026-05-31)¶
All 10 scenarios were authored before cube-C2 (expertise-requirement) existed. The cube discussion flagged a confound: α scenarios skew low on expertise-requirement, γ scenarios skew high — so without explicit C2 tags Phase 3 would conflate structural vector signal with expertise-requirement signal. Each scenario now carries an inline **cube-C2:** tag (above); this section summarizes the pattern and what it does to Phase 3 interpretation. Scale for the means: low=1, moderate=2, high=3, expert-only=4.
Summary matrix¶
| Domain | α (snap commitment) | γ (irreversible institutional) |
|---|---|---|
| career | low | high |
| health | low | high |
| relational | low | low |
| financial | moderate | high |
| governance | low | high |
- α mean ≈ 1.2 — four low + financial-α moderate.
- γ mean ≈ 2.6 — four high + relational-γ low.
- Gap ≈ 1.4 points on a 4-point scale → the flagged confound is real and substantial. The γ structural shift raises expertise-requirement in 4 of 5 domains.
Why the confound is breakable (the important finding)¶
The skew is not uniform, and the breaks are not random — they cluster by domain:
- The relational domain holds C2 constant at LOW across both vectors. relational-α ("Sam's First Dinner") and relational-γ ("The Letter in the Safe") are both low-C2 despite being structurally α vs γ. This is the cleanest cell in the whole experiment: a pure structure manipulation with expertise held constant. If the player's α-vs-γ response differs here, that difference is attributable to structure, not expertise. Relational decisions stay interpersonal even at institutional scale — the executor's real dilemma is still "when do I tell the family a secret," not estate law.
- relational-γ is the within-γ control. It has the full γ structure (irreversible, life-defining, institutional, mutual-uncertainty) but no expertise load. Compare it against the four high-C2 γ scenarios to test whether the "γ signal profile" tracks structure or expertise.
- financial-α (moderate) is a secondary within-α check — the one α scenario with elevated C2, useful for confirming α-band responses don't hinge on its slightly higher expertise demand.
If C2 were perfectly collinear with the α/γ vector (all α low, all γ high), structure and expertise would be unidentifiable. The relational domain's constant-low C2 is what rescues identifiability.
The acute confound: γ option C¶
γ's locked option C is "brief peers / oversight before acting" — an expertise-deference-flavored move. On the four high-C2 γ scenarios, a player choosing C could be responding to the structure (institutional + irreversible → get cover) OR to the expertise load (I'm out of my depth → defer to experts). These are confounded precisely on the C option. relational-γ disambiguates it: same structure, low C2 → if the player still picks C there (brief the attorney), the "γ → C" pull is structural, not expertise-deference.
Does the covariate shift Phase 3 interpretation?¶
Yes, materially, but in a contained way:
- Most exposed: γ option-C selection and conviction/calibration. Per measurement-schema M3, lower conviction on a high-C2 γ scenario may be correct humility, not a structural conviction drop — do not read it as a γ-structure effect without conditioning on C2.
- Less exposed: option choice driven by the A/D stance (principle vs autonomy) and rationale shape (the within-vector consistency the experiment actually measures), which are not obviously knowledge-gated.
Recommendations for Phase 3¶
- Enter C2 as an explicit covariate in the variance decomposition (structure vs. domain vs. C2) — not folded into the structural vector.
- Use the relational domain as the controlled structure contrast (C2 = low in both vectors); pre-register relational-α vs relational-γ as the clean α-vs-γ test.
- Read γ option-C and conviction conditional on C2 — never pool relational-γ (low) with the four high-C2 γ scenarios for those measures.
- Note the α-side limitation: four of five α are low-C2, so within α, structure and low-expertise are nearly confounded; financial-α (moderate) is the only lever there. Most C2 separation lives in the γ group + the relational column.
Bottom line¶
The expertise-requirement covariate substantially shifts interpretation of γ option-C selection and conviction, and it must enter the Phase 3 model. But it does not invalidate the experiment: the relational domain holds C2 constant across both vectors (clean structure contrast) and supplies the low-C2 within-γ control. Without these tags Phase 3 would have silently attributed an expertise effect — especially the pull toward "brief oversight" — to the γ structural vector.