Corpus audit: role framing & worldview tax (ADR-058)¶
Date: 2026-05-15
Audited against: ADR-058
Scope: All 22 campaigns with static audit files in scripts/data/audit-*.md + the 13 reactive arcs in migrations 070 / 071 + the 103 solo scenarios in scripts/data/scenarios.json + the scenario-generator contract in src/lib/ai/scenario-generator.ts + the /cowork master document in docs/generation-prompts.md. Full content corpus + generation surfaces.
TL;DR¶
Of 22 static campaigns audited, 3 are Critical (Stonemark, Red Lines, The Ember Network), 2 are High (The Kepler Divide, The Brink), 1 is Medium (The Third Generation). The remaining 16 audit clean.
The 13 reactive arcs in 070_reactive_worlds_batch2.sql / 071_reactive_worlds_batch3.sql are mostly clean by construction — they use the leader-with-delegated-authority shape that already passes Guardrail 4. One reactive arc (The Empty Chair) carries the same failure mode as Red Lines / Ember Network and should be reframed.
The 103 solo scenarios in scripts/data/scenarios.json audit GREEN — every scenario uses Tier 1 leader-cast framing and the bank is by construction immune to the failure mode (modern institutional settings only, no embedded-worldview characters). See Phase 3 addendum at the bottom.
Generation surfaces (Phase 1+2) have been updated with Guardrail 4 to prevent future scenario generation from re-producing the failure mode — see the Phase 1+2 addendum.
The corpus' "passing" pattern (Pivot, Reckoning, Cap, all 12 leader-cast reactive arcs, all 103 solo scenarios) confirms ADR-058's refined fix: leader with delegated authority preserves stakes while dropping worldview tax. The Critical campaigns share one specific failure mode — advisor + embedded worldview — and most can be fixed by promoting the player into the decision seat without changing the underlying decision substrate.
Severity-sorted findings¶
| Severity | Campaign | Failure mode | Recommended fix tier |
|---|---|---|---|
| Critical | Stonemark | Player as advisor to faith-based authority | Tier 1 (leader-cast) |
| Critical | Red Lines | Player as member of clandestine medical-research cell | Tier 2 (hybrid) |
| Critical | The Ember Network | Player as steering-committee member of underground whistleblower collective | Tier 2 (hybrid) or Tier 3 (advisor) |
| Critical | The Empty Chair (reactive) | Player as second-in-command of resistance cell | Tier 2 (hybrid) |
| High | The Kepler Divide | Player inside colony governance structure that presupposes original-residents-as-authority | Tier 1 (leader-cast) — already partially in place |
| High | The Brink | Player as "collective representative" — labor-organizing context with embedded workers'-claim legitimacy | Tier 2 (hybrid) |
| Medium | The Third Generation | Player as family insider once Ch.3 starts — dynastic-succession frame | Tier 2 (hybrid) or Tier 3 (advisor) from Ch.3 onward |
| Skip | The Hollow Crown, The Cap, The Fever Road, A Weight in Silver, Dead Reckoning, The Annexation, The Compass Rose, The Donor, The Endowment, The Exit, The Flood, The Ledger, The Long Rebuild, The Pivot, The Reckoning, The Story, plus 12 of 13 reactive arcs | No worldview tax detected | None needed |
Cross-cutting findings¶
1. The corpus already knows the fix pattern¶
Every static campaign that audits clean (16 of 22) and 12 of 13 reactive arcs share a structural feature: the player is cast as the leader with delegated authority, and the decision is theirs. Examples — "You are the CFO of St. Alicia's Health Network reviewing quarterly financials" (Reckoning Ch.1), "You are the GM of a mid-market NBA franchise" (The Cap), "You are the newly appointed Steward of Ravenmarch" (Regent's Shadow), "You are the acting Governor of Tessaly" (Colony Vote). Stakes are full; agency is full; no stakeholder's worldview is load-bearing because authority is the player's.
This means most of the Critical / High fixes are not full rewrites — they're role inversions that promote the player into the decision seat already present in the scenario. The scenario substrate (technocratic vs. charismatic authority for Stonemark; information urgency vs. verification rigor for Red Lines / Ember Network; resource integration for Kepler Divide) is preserved. Only the player's vantage changes.
2. The scenario-generator contract enforces the surface where this lives¶
src/lib/ai/scenario-generator.ts:170 mandates per ADR-034 that every scenario open with "You're the [role] of [setting]" or "You advise [character]". This is the surface where worldview tax lives — and the contract currently does not distinguish "You advise [character whose worldview is load-bearing]" (which produces the Stonemark failure mode) from "You advise [character in a procedural role]" (which is fine, as in The Pivot's CEO-advisor framing).
Recommendation: Add a Guardrail 4 paragraph to the scenario generation prompts in src/lib/ai/scenario-generator.ts that explicitly prefers Tier 1 (leader-cast) over Tier 3 (advisor) when the setting carries an embedded worldview. A short rule like: "If the setting has a load-bearing authority figure with a specific worldview (religious leader, sworn-order superior, cell leader, faction head), cast the player as the decider with delegated authority — never as that figure's advisor."
3. Bad combination is advisor + embedded worldview, not either axis alone¶
The Hollow Crown clears Guardrail 4 despite being a hereditary monarchy because the player is a privy councillor — advisor inside a procedural institution (the council, not the crown itself). The Reckoning clears it as CFO inside a healthcare system. Only the intersection produces the tax. The audit found 4 Critical cases and all 4 sit at that exact intersection.
4. Option-level findings are tentative¶
Per the second revision of ADR-058, option-level worldview tax is a hypothesis, not a confirmed axis. The audit did flag authority-referencing Risk: clauses in Stonemark, Kepler Divide, and The Brink, but on the user's debrief the second-player evidence (Cici) leaned toward action-scope ambiguity (ADR-054 territory) rather than worldview rejection. Don't reflexively rewrite every authority-referencing Risk: clause. Triage per campaign: is the issue worldview tax, action-scope ambiguity, or both?
Per-campaign findings (Critical + High + Medium)¶
Stonemark¶
Current frame (Ch.1): "Your role: You advise the settlement council. Crane will listen to you before he decides." The player is positioned as deferring to Crane — whose authority is described as "moral, not legal. People follow him because he was right when it mattered."
Failure mode: Tier-1 example of advisor + embedded worldview. The frame requires the player to operate as Crane's adviser, accepting Crane's moral authority as the legitimate locus of decision-making. Jonathan's playtest rationale ("the whole idea of following a religious leader in the first place really throws me off... I have a hard time imagining myself owing always following this leader") is the load-bearing evidence for this ADR.
Proposed reframing — Tier 1 (leader-cast):
"You are the elected Council Chair of Stonemark, a settlement of 180 people. Aldous Crane founded this community and his authority kept it alive through the collapse, but the council was always meant to handle governance after him. Crane has stepped back from active leadership for the season and named you to convene the council in his place. Twenty engineers from the east have arrived offering to rebuild the failing water filtration system in exchange for full membership and governance rights. The decision is yours."
The decision substrate is unchanged — same four options, same trade-offs (technocratic expertise vs. communal cohesion). What changes: Crane is honored as founder, not deferred to as live authority; the player has agency and stakes; no premise endorsement is required. The same pattern propagates through all 7 chapters (Council Chair making each downstream call, Crane as predecessor-figure whose advice is sought but not deferred to).
Option-level review: Ch.1 includes "Risk: private negotiations bypass the council process Crane built." Under the leader-cast reframe, this becomes "Risk: private negotiations bypass the council's deliberation" — a process-relative phrasing that lands for any player who values consultative process, regardless of whether they share Crane's specific worldview. The fix is small and rides along with the role reframe.
Effort: Medium — 7 chapter intros to revise + option text in roughly 20 options across the chapters. The April 23 audit files already provide the original chapter content; revisions are reframes, not rewrites.
Red Lines¶
Current frame: "Member of a six-person cell distributing suppressed medical research. Legal in theory. Dangerous in practice."
Failure mode: Critical advisor + embedded worldview failure. The frame requires the player to accept as background premise that (a) pharmaceutical regulation has been captured or failed, (b) legal whistleblowing channels are insufficient, (c) clandestine distribution is therefore justified. Players who trust regulatory bodies or believe legal channels should be exhausted first carry a substantial worldview tax just to engage. Option risks compound this — "Risk: every distribution contact is exposed" only reads as a tradeoff if you accept that distribution is the right action.
Proposed reframing — Tier 2 (hybrid):
"You're a clinical researcher who helped author a fourteen-paper review of treatment failures the manufacturer suppressed in litigation discovery. The review's other authors have formed a small group debating how to get the findings into clinicians' hands before the next prescription cycle. You're the only one still inside an academic medical center; your colleagues are former regulators, retired investigators, and one journalist. The group has asked you to make the call on what comes next."
Why hybrid rather than leader-cast: a "leader of the cell" framing still requires the player to accept cell-formation as the appropriate response. The hybrid version positions the player as committed (you're an author of the work; you have stakes in whether it's seen) but not yet captured by any particular response strategy (publish under your own name, go to FDA, leak to press, work with manufacturer's whistleblower line, etc.). Each of those becomes a real option the player chooses from, not a frame they're already inside.
Decision substrate preserved: information urgency vs. verification rigor, source protection vs. public reach, institutional channels vs. direct action. These are exactly the tensions the original aimed to surface — the hybrid framing surfaces them more cleanly because the player isn't pre-committed to one resolution.
Effort: High — 7 chapters need substantive reframing because the cell-membership is woven through the narrative. Recommend pausing this campaign in production until the rewrite ships, since the worldview tax here is the most acute in the corpus alongside Stonemark.
The Ember Network¶
Current frame: "Steering committee member, whistleblower collective (The Ember Network)."
Failure mode: Structurally identical to Red Lines. The collective's legitimacy (operating outside legal channels for pharmaceutical whistleblowing) is the implicit premise; players who would have first exhausted FDA / SEC / internal whistleblower channels carry a tax.
Proposed reframing — Tier 2 (hybrid) or Tier 3 (advisor):
The cleanest fix here may actually be Tier 3 (pure advisor) given the institutional setting — position the player as outside legal counsel newly retained by the collective, or as a regulatory affairs consultant the collective has approached. The player advises on the publication / verification / source-protection tradeoffs without being a member who has already endorsed the collective's existence.
"You're a whistleblower-protection attorney at a public-interest law firm. A group calling themselves The Ember Network — fourteen pharmaceutical-industry insiders with documents about a drug's suppressed adverse-event data — has retained you. They have a publication window of nine days before the manufacturer's next earnings call. Your job is to advise on what gets released, how, and to whom. They will follow your counsel."
This is one of the cases where the consultative-stakes muting from ADR-058's revised fix pattern is acceptable — the substrate (information ethics under time pressure, source protection mechanics, institutional vs. direct routes) doesn't require the player to personally absorb the post-publication consequences to surface.
Effort: High — 6 chapters need reframing, but the structural pattern is consistent so the rewrite work parallelizes.
The Empty Chair (reactive arc, migration 070)¶
Current frame (arc preamble): "You are the second-in-command of a cell operating inside a larger resistance network — the kind of work that gets done in basements, in borrowed trucks, in conversations that end the moment someone new walks in."
Failure mode: Same shape as Red Lines / Ember Network. Resistance-network membership requires player endorsement of the network's operational premises. This is a reactive arc (live-generated scenarios), so the failure cascades from a single seed.
Proposed reframing — Tier 2 (hybrid):
Reseed the arc preamble to position the player as someone whose family / livelihood / community is at stake from whatever the resistance is resisting — but who has NOT been part of the network's prior decisions:
"You're a clinic nurse in a city under a regime that detains physicians for treating the wrong patients. A relative who was part of a resistance cell was arrested eleven days ago; the network has gone quiet. A courier has arrived this morning saying the cell needs someone they can trust and naming you. You have not done this work before. Every decision you make will reshape what comes next."
Stakes are very real (relative arrested, regime risk to nurse's own clinic) without requiring the player to come in already-aligned with the cell's prior strategy. The decision substrate (trust under uncertainty, operational security, when to act on unverified orders) is preserved.
Effort: Low — single preamble edit + matching narrative-voice prompt update. Reactive arcs cascade from the seed, so one well-placed change propagates.
The Kepler Divide¶
Current frame: Player as "Council member on Kepler Station" across most chapters, with Ch.6 ("The Rationing Formula") and Ch.7 ("The Sun on the Wall") tipping into worldview tax — Ch.6 positions the player to weigh fairness frames within Holt's existing authority structure; Ch.7 asks "does Tam have standing to speak?" which presupposes tenure/origin determines voice legitimacy.
Failure mode: Partial / inherited — the early chapters position the player as advisor in a procedural institution (clean), but Ch.6 makes the player a tiebreaker inside Holt's frame, and Ch.7 frames a stranger's voice as something to be granted or denied based on the station's inherited norms. The aggregate effect: by mid-campaign the player has been quietly enrolled in Kepler's authority structure.
Proposed reframing — Tier 1 (leader-cast) from Ch.5 onward, or Tier 3 (mediator) throughout:
Option A (preferred): Promote the player. Ch.5 introduces a leadership vacuum (Holt steps back due to illness / death / loss of confidence vote) and the player becomes acting council chair / station administrator. Ch.6's rationing decision and Ch.7's standing question are then their calls to make, not arbitrations within Holt's frame.
Option B: Keep advisor cast throughout, but explicitly position the player as a regional federation observer — same vantage as Ch.1, sustained through Ch.7. Loses some stakes signal in late chapters.
Effort: Medium — only the back half (Ch.5–7) needs structural revision; Ch.1–4 are largely fine as-is.
Option-level review: Ch.6 contains "If the Eira survivors have voting rights, they're 23% of the population making decisions about resources they didn't build." On the strict ADR-058 second-revision standard, this is borderline — it presupposes that original-construction confers governance weight, which is a contested premise. But it's also an authentic argument one of the characters would make. Recommend keeping the line as Holt's quoted position, not as scenario-narrator framing — that preserves the in-scenario debate while removing implicit narrator endorsement.
The Brink¶
Current frame: Player as "representative of the [Meridian-7] collective" — labor organizing context where the collective's claim (43-year occupation against a corporate buyout) is the player's frame.
Failure mode: Partial. The setting is ideologically charged (workers' rights vs. property rights), and the role implicitly aligns the player with the workers' position. A player who weights property rights or capital legitimacy more strongly carries a tax. The audit's April-23 rewrite already softened the "member of the collective" language toward "represents collective interests," which is an improvement but doesn't fully clear the frame.
Proposed reframing — Tier 2 (hybrid):
"You worked on the Meridian-7 reclamation crew for eleven years before being elected to the collective's three-person negotiating committee three months ago. You know the technicals, you know the people, but you weren't part of the founding fight 43 years ago. Helion Capital is offering to buy the collective's claim and a court just ruled in Helion's favor on the underlying property question. The committee meets tomorrow to decide whether to take the offer or fight the ruling."
The hybrid framing keeps the personal stakes (your livelihood is on the line) without locking in a particular ideological frame about whether the collective's 43-year claim is legitimate. The player can weigh both the workers' position and Helion's legal position as live considerations.
Effort: Medium — 6 chapters need consistent treatment of the player's relationship to the collective and to Helion.
The Third Generation¶
Current frame: Player as "family advisor" in Ch.1–2 (clean), then "one of three siblings" from Ch.3 onward (partial tax).
Failure mode: Subtle. Family-dynasty succession assumes that name-as-asset, generational continuity, and inherited leadership carry weight — premises some players share and others don't. The mid-campaign shift to family-insider locks the player into that frame.
Proposed reframing — Tier 2 (hybrid) or Tier 3 (advisor) from Ch.3 onward:
Keep the player as family advisor / transition consultant / outside attorney throughout, rather than promoting them to sibling at Ch.3. The decision tension (Claire vs. David vs. Noel's competing visions for the funeral home; the name-on-the-building question; the codicil dispute) surfaces just as well from the advisor's seat — and arguably more cleanly, since the advisor isn't pre-committed to any sibling's frame.
Effort: Low — primarily a vantage adjustment in Ch.3–5; later chapters need light touch-ups for consistency.
Not flagged — pattern study¶
Worth noting for the future scenario-authoring contract: these campaigns audit clean and embody the patterns we want generation to replicate.
- The Hollow Crown — advisor in a procedural institution (privy council). Embedded setting (monarchy) but no worldview tax because the institution is rule-based, not personality-based.
- The Pivot, The Reckoning, The Cap — leader-cast in neutral corporate / sports settings. Decision is the player's, stakes are full, no premise endorsement.
- The Fever Road — leader-cast in a survival setting (advisor to a caravan leader whose authority is earned through navigation skill, not worldview).
- Twelve of thirteen reactive arcs in 070 / 071 — leader-cast across diverse settings (mining station, terraformer colony, deep-space ship, feudal duchy, smuggling crew, family business in crisis, Free Cities envoy). The reactive-arc author template clearly favors Tier 1.
Recommended actions¶
| Priority | Action | Effort | Status |
|---|---|---|---|
| P0 | Reframe Stonemark Ch.1 (the immediate ADR-058 trigger) | Low — single chapter intro | ✅ Shipped (migration 096) |
| P0 | Update scenario-generator prompts in src/lib/ai/scenario-generator.ts to encode Guardrail 4 |
Low — prompt-engineering edit | ✅ Shipped (Phase 1, ADR-058 follow-up) |
| P1 | Reframe Stonemark Ch.2–7 (consistent leader-cast) | Medium | ✅ Shipped (migration 096) |
| P1 | Reseed The Empty Chair reactive arc preamble | Low | ✅ Shipped (migration 096) |
| P1 | Pause Red Lines + The Ember Network in production pending rewrites | Trivial (feature flag) | ✅ Shipped (migration 096; new campaign_arcs.is_active column + API filter) |
| P2 | Reframe Red Lines (institutional grounding — clinical researcher whose own work was suppressed) | High | ✅ Shipped (migration 098, unpaused) |
| P2 | Reframe The Ember Network (institutional grounding — 501(c)(3) journalism nonprofit) | High | ✅ Shipped (migration 098, unpaused) |
| P2 | Reframe Kepler Divide Ch.5–6 (acting administrator after Holt sidelined) | Medium | ✅ Shipped (migration 097) |
| P2 | Reframe The Brink (hybrid cast: 11 years on the crew, newly elected to committee) | Medium | ✅ Shipped (migration 097, arc description) |
| P3 | Adjust The Third Generation (Tier 3 consultant cast at the arc level) | Low | ✅ Shipped (migration 097) |
| P3 | Triage option-level Risk: / Cost: flags in audit (worldview tax vs. action-scope ambiguity per ADR-054) |
Medium | ⏸ Deferred — option-level claim was softened in ADR-058 revision 2; most concerns are dampened now that role-cast is fixed. Handle opportunistically as scenarios surface friction. |
Phase 3 addendum (2026-05-15): Solo scenario bank audit¶
Extended the audit to cover scripts/data/scenarios.json — the 103-scenario curated bank that drives both solo single-round play and (via category_filter) campaign chapter scenario selection.
Result: GREEN. All 103 solo scenarios audit clean against Guardrail 4. No reframing needed.
Pattern study: - Every scenario opens with a Tier 1 leader-cast framing: "You're the CEO of...", "You're the Executive Director of...", "You're the Managing Partner of...", "You're on the 5-person governance council of...", "You're the superintendent of...", "You're the provost at...", etc. - 100% of openings cast the player as the decision-maker with delegated authority — no "You advise..." or "You serve under..." framings exist in the bank. - Settings are exclusively modern institutional contexts (corporate, nonprofit, government, education, healthcare, DAO). No faith-led communities, monarchies, sworn orders, resistance cells, or ideological enclaves. The bank is by construction immune to the advisor + embedded-worldview character failure mode. - Option text (spot-checked across governance, resource-allocation, team-dynamics, and values-culture categories) is value-diverse but doesn't tax player worldview. Where an option voice does carry a position (e.g., "Bending rules is how DAOs become corporations"), the other options offer counter-positions — a clean worldview-split per Principle #6, not a Guardrail-4 violation.
Why the solo bank is clean by construction: The scenarios were authored with the case-study voice contract (ADR-034) in force from the start, and case-study voice strongly biases toward Tier 1 framings (the player as decision-maker with concrete authority). The campaigns that failed Guardrail 4 (Stonemark, Red Lines, Ember Network, The Empty Chair) failed in their campaign-level arc preambles and chapter intros — the surfaces that case-study voice doesn't govern. The bank scenarios that those campaigns select via category_filter retain their own clean framing regardless of surrounding campaign context.
Stylistic side-note (not Guardrail 4): Two scenarios use the explicitly-banned **Your Role:** block header format that ADR-034 forbids:
- "The Acquisition Integration"
- "The Patent War Decision"
These are stylistic violations against ADR-034, not Guardrail 4 issues. Flagging here for future cleanup but they're not blockers for this audit.
Bank-level Phase 3 verdict: No action items. The solo scenario bank already exemplifies the pattern the rest of the corpus should converge to.
Phase 1+2 addendum (2026-05-15): Generation-prompt and master-doc updates¶
To prevent future scenario generation from re-producing the failure mode, Guardrail 4 was embedded in three code files and one master document:
Code (Phase 1):
- src/lib/ai/scenario-generator.ts — Added Principle #14 (No worldview tax in player's role) with three-tier preference, bad-combination definition, and quick test.
- src/lib/ai/campaign-master-doc.ts — Extended section 2 (The Player) with three-tier preference; added CRITICAL RULE #6 explicitly banning "advises [name]" when [name] is a worldview-bearing character.
- src/lib/ai/scenario-themes.ts — Role-framing notes added to the two highest-risk themes: court (monarchical risk) and underground (resistance cells / whistleblower networks — the highest worldview-tax risk).
Master doc (Phase 2):
- docs/archive/generation-prompts.md — Added top-of-file Universal Role-Framing Rule with three-tier preference and high-risk-theme callouts. Referenced from Prompts 1, 3, 4, 5, 6; inline notes on Prompt 2's Underground and Court theme descriptions.
This closes the generation-side risk: any new arc, scenario, seed, or preamble produced via either the runtime generators or the /cowork batch prompts now has to clear Guardrail 4 by authoring rule.
Open questions for follow-up¶
- Should P1 reactive-arc and Stonemark fixes ship as a single migration (e.g.,
100_role_framing_corpus_fix.sql), or per-campaign? Recommend single migration for atomicity and rollback safety. - Do we want to retroactively re-tag scenarios where Guardrail 4 flagged friction (e.g., demote conviction signal on completed plays of the unfixed Stonemark)? This is the ADR-058 consequence note about "decision conviction signal contamination." Probably yes for a small batch (Jonathan's Stonemark play, any other completed plays of the four Critical campaigns), but only if the volume is small enough to be tractable.
- Does the scenario-generator contract update belong in ADR-058 itself, or a separate small ADR? Lean toward separate — ADR-058 documents the principle; the generator change is an implementation that should be revisitable independently.