ADR-058: Role framing as a second axis of identity capture — audit scenarios for worldview-tax in the player's assigned role¶

Status: Accepted Date: 2026-05-15 Context: A two-player playtest of "The Twenty Engineers" (Chapter 1 of a Stonemark campaign) surfaced a failure mode that ADR-027 does not fully cover. The scenario topic is politically neutral (technocratic expertise vs. charismatic moral authority — a legitimate decision-style axis). But the player is implicitly positioned as a member of a settlement that follows a faith-based leader ("Crane's authority is moral, not legal. People follow him because he was right when it mattered. Crane will listen to you before he decides."). Both players bumped against the frame in different ways. The first player reported spending cognitive budget resisting the role's implicit worldview ("having a faith-based decision where we give them total control just doesn't sit right for me") before being able to weigh the four options. The second player rejected one of the option's stated risks at the option-text level ("I don't perceive the rule to be valid... I'm just having a conversation"), revealing that the same worldview tax was embedded in the option descriptions, not just the role framing. Both DTAs actually predicted correctly; the miss was at the peer-prediction layer (Jonathan predicted Cici would pick "technical roles only" when she picked "negotiate"), and Jonathan subsequently revised from "work-first probation" to "negotiate terms" after Cici walked through her information-gathering reframing. The scenario produced rich signal anyway, but with a values-friction tax that ADR-027's topic-level guardrails don't catch.

ADR-027 asks: "Is the scenario about something politically/identity-charged?" This case asks a different question: "Does the scenario's role assignment — or its option descriptions — require the player to accept a worldview as background premise in order to engage?" These are independent failure modes.

At a glance

What it decides: Scenarios must not tax the player with a worldview they have to endorse just to inhabit their assigned role — the decision tension lives in the options, not in the frame. This adds Guardrail 4 to the ADR-027 content guardrails and triggers a corpus audit.

Worldview tax has two surfaces — role-level (confirmed: player positioned as deferring to a faith-based leader) and option-level (a hypothesis, entangled with option clarity; flag Risk:/Cost: clauses as candidates, don't reflexively rewrite).
The bad combination is advisor + embedded worldview — leader inside a worldview-laden setting is fine, advisor inside a procedural institution is fine; the audit targets that one quadrant.
Fix preserves stakes, drops the tax — prefer leader-with-delegated-authority, then insider-outsider hybrid, then pure outside advisor as last resort (which mutes personal-stakes signal).
Rejected deleting embedded-worldview settings, values disclaimers, and trusting players to suspend disbelief — transcript evidence shows even willing players surface the friction.
Watch for: false positives (doctor/parent roles carry values most players share); pre-audit conviction/timing signals on flagged scenarios are lower-confidence.

flowchart TD
    A["Player's assigned role"] --> B{"Does engaging require<br/>endorsing a worldview?"}
    B -->|No| OK["No tax — ship"]
    B -->|Yes| C{"Does the player hold<br/>the authority?"}
    C -->|"Leader / delegated authority"| T1["Tier 1: full stakes,<br/>full agency — preferred"]
    C -->|"Committed but not steeped"| T2["Tier 2: insider-outsider<br/>hybrid — stakes via ties"]
    C -->|"Deferring to another's authority"| BAD["Bad quadrant:<br/>advisor + embedded worldview"]
    BAD --> T3["Tier 3: pure outside advisor<br/>(last resort — mutes stakes)"]

Reframing tiers: drop the worldview tax while preserving as much personal stake as the cast allows.

Decision¶

Extend the ADR-027 guardrails with a third axis covering role framing, and trigger a corpus audit of existing scenarios against it.

Guardrail 4: No worldview-tax in role framing¶

A scenario should not require the player to inhabit a role whose authority structure, belief system, or value commitments are themselves the implicit premise the player must accept to engage. The decision tension should live in the options, not in the frame.

The worldview-tax test: If a meaningful fraction of modern players would need to suspend personal disbelief about who they are in this scenario (not just what's happening), the framing taxes the signal. Reasonable disagreement about the decision is the goal; reasonable disagreement about the premise of the role is noise.

Worldview tax has a well-attested primary surface (role-level) and a hypothesized secondary surface (option-level) that is partly entangled with option clarity. The role-level surface is the player's assigned identity (member of a faith-led settlement) — this is where the playtest evidence is unambiguous. The option-level surface is the hypothesis that Risk: / Cost: clauses referencing the surrounding authority structure ("Risk: bypasses the council process Crane built") tax players who don't endorse that structure. The Stonemark playtest produced suggestive but not clean evidence for this second surface (see Discussion) — the second player's friction reads partly as worldview-rejection and partly as action-scope reframing (an ADR-054 clarity issue). Treat the role-level surface as the primary thing to fix; treat option-level authority-referencing clauses as candidates for review without assuming worldview tax is the dominant mechanism.

Fix pattern — worldview tax and stake-investment are on different axes; the goal is to drop the tax while preserving stakes. The earlier draft of this ADR proposed "reposition the player as outside advisor/consultant" as the universal fix. That under-specified the goal: pure consultant framing drops the tax but also drops personal stakes, which mutes signals tied to skin in the game (relationship-weight, loyalty, what-would-I-actually-do-when-I'm-the-one-staying-in-the-room). Sync measures decisions under tradeoffs, and tradeoffs hit harder when the decider lives with them.

The corpus-pattern observation makes this concrete: nearly every scenario that already clears Guardrail 4 (The Pivot, The Reckoning, The Cap, the reactive arcs in 070_reactive_worlds_batch2.sql / 071_reactive_worlds_batch3.sql) uses the same shape — the player is the newly appointed/promoted/elected leader and the decision is theirs. That gives them full stakes, full agency, and a clean exit from any single stakeholder's worldview because the authority is theirs, not borrowed. The Stonemark failure isn't "player is too embedded" — it's specifically advisor + embedded worldview: the player is positioned as deferring to Crane (whose worldview is load-bearing) rather than holding the decision themselves.

Preferred reframing patterns, in order:

Leader with delegated authority (preferred when it fits). The player is the COO / CFO / Governor / Council Chair / Steward / newly promoted X, and the decision is theirs. Stakes are full; agency is full; no premise endorsement is required because the locus of authority is the player. For Stonemark: "Crane has stepped back from the council for the season — the decision about the engineers falls to you" cleanly inverts the failed framing.
Insider-outsider hybrid (when leader cast doesn't fit). The player is committed to the place but not steeped in its frame — a recent arrival who plans to stay, a returning member after long absence, a resident with dual loyalty (e.g., "Your daughter is one of the children Harwick treated; you're also Crane's neighbor"). Stakes are preserved via personal ties; the player is not asked to endorse the surrounding worldview as a precondition.
Pure outside advisor (last resort). Mediator, regulator, journalist, expert witness, visiting expert. Use only when the scenario is genuinely consultative — and accept that personal-stakes signal will be muted. The Hollow Crown is a clean example: privy councillor is an advisor role, but the institution itself is procedural, not worldview-loaded, so neither surface taxes the player.

Don't universalize any of these. The bad combination is advisor + embedded worldview. Either axis alone is fine: leader inside a worldview-laden setting is fine (it's the player's authority that operates); advisor inside a procedural institution is fine (no premise to endorse). The audit should flag scenarios that fall into the bad-combination quadrant specifically.

For option-level concerns, review Risk: / Cost: clauses that reference authority structures and triage them as candidates — check whether the issue is worldview tax, action-scope ambiguity (ADR-054 territory), or both. The fix differs: process-relative rewording for worldview tax vs. tightening what the action concretely entails for ambiguity. Don't reflexively rewrite every authority-referencing clause as worldview-loaded; the Stonemark playtest's option-level evidence was mixed and partly entangled with option clarity.

Examples: - Bad: "You are a member of Stonemark, governed by Aldous Crane, preacher and founder. Crane will listen to you." - Better: "You are a circuit judge passing through Stonemark. Crane has asked your outside read before he decides." - Bad: "You serve in the temple guard, sworn to the High Priestess." - Better: "You are a foreign envoy negotiating with the temple, which holds civil authority here."

The decision substrate (who gets governance rights, how to weigh expertise against tradition, what to do about a faction-forming opportunity) is unchanged. Only the player's implicit endorsement of the surrounding worldview is removed.

Audit trigger¶

This pattern likely affects more than one scenario. Trigger a corpus audit of scenarios — particularly campaign chapters with embedded settings (religious orders, monarchies, ideological communities, military hierarchies) — against Guardrail 4 before any further campaign expansion.

Rationale¶

ADR-027 and the identity-vs-process philosophy establish that Sync should measure how someone decides, not who they perform as. ADR-027 catches identity capture at the topic level (slavery, immigration, gender). This ADR catches it at the role level — a subtler failure where the topic is fine but the player is asked to reason from inside a worldview they don't share.

The Stonemark transcript is instructive: the player produced a coherent decision (work-first probation), rejected each alternative with a stated reason ("technical roles → they'll leave"; "negotiate directly → bypasses process Crane built"), and surfaced a genuine heuristic ("everything changes is the law of the universe"). Signal was present. But the player spent the first paragraph of their rationale on the frame, not the decision — and explicitly flagged "I have a hard time imagining myself owing always following this leader." That's friction the scenario didn't need to introduce to test the technocracy-vs-tradition axis.

Two reasons this matters beyond signal quality: 1. Decision conviction (ADR-015) is contaminated when "tough call" reflects values-friction with the frame, not genuine difficulty with the options. The DTA will learn a noisier mapping between scenario features and conviction level. 2. Alienation risk. Players from worldviews not represented in the role (atheists in faith-led settlements, secularists in monarchies, etc.) carry a tax that players whose worldviews match the frame do not. This is the same fairness concern ADR-027 raised about reducing atrocities or marginalized identities to puzzles, applied to the player's assigned identity.

Alternatives Considered¶

Remove faith/monarchy/ideology settings from the corpus. Rejected — this over-corrects and shrinks the scenario design space. Stonemark's underlying tension (expertise meets charismatic moral authority) is a great decision-style axis. The fix is reframing the role, not deleting the setting.
Add a "values disclaimer" to scenarios with embedded worldviews. Rejected for the same reason ADR-027 rejected content warnings — disclaimers don't fix the measurement problem. The player still pays the worldview tax to engage.
Trust the player to suspend disbelief. Rejected — the transcript evidence shows that even players willing to engage seriously still surface the frame-friction explicitly. We cannot rely on suspension of disbelief as a corpus design assumption.
Treat this as a one-off scenario edit. Rejected — the user's instinct was correct that "we'll need to review more than just this one story." Embedded-worldview settings are common enough across the campaign corpus to warrant a pattern rather than a single fix.

Discussion¶

The trigger was a single playtest reflection where the player explicitly named the failure mode in their own words: "the campaign is called Stonemark... the whole premise is a faith-based leader, directly contradictory to my personal values... I was asked to take the perspective of someone who was in the settlement under this leader, and I have a hard time imagining myself owing always following this leader." Crucially, the player still completed the scenario and produced useful signal — this is not a scenario that fails outright, just one that taxes the player unnecessarily.

The key insight separating this from ADR-027: the topic (how should a small community absorb a sudden infusion of technical expertise?) is exactly the kind of genuinely ambiguous tradeoff Sync wants to surface. There is no morally-correct answer; reasonable people will split across all four options based on their actual decision style. ADR-027 would clear this scenario without comment. The problem is one layer up — in the who the player is told they are.

This also clarifies the identity-vs-process memory: identity capture has two surfaces. The first is "what beliefs does my choice signal?" (ADR-027). The second is "what beliefs does engaging with this role at all require?" (this ADR). Both leak the same signal — political/values identity instead of decision style — through different mechanisms.

Second-player evidence and the option-level hypothesis¶

The same playtest produced a second data point on the second player (Cici) that initially looked like an option-level version of the same pattern but, on closer reading with the user, is more ambiguous. Cici chose "Negotiate terms with Voss directly" against an option description that included "Risk: private negotiations bypass the council process Crane built." Her rationale included both "I don't perceive the rule to be valid" (which reads like worldview-rejection) and "It's not really my decision to make. I'm just having a conversation with her and getting more context" (which reads like action-scope reframing — accepting the rule but classifying her chosen action as outside its scope).

These are different failure modes. Worldview-rejection would say: the council process doesn't deserve protection, so bypassing it isn't a real risk. Action-scope reframing says: bypassing the council process IS a real risk, but information-gathering doesn't constitute bypassing — it's a "bonus extra step," not a process-skip. Cici's reasoning leans toward the second reading on the user's own debrief: she accepted that the council had a process, accepted that going around it would be risky, and reclassified her chosen action so the framing wouldn't apply. That is closer to an option clarity issue (ADR-054 territory — the option text underspecifies what the chosen action actually entails) than to option-level worldview tax.

The honest conclusion is that the option-level surface remains a hypothesis, not a confirmed second axis of worldview tax. Authority-referencing Risk: clauses may tax players who don't accept those structures, but we don't yet have clean evidence that this is happening independent of option clarity. The audit should flag authority-referencing Risk: / Cost: clauses as candidates — for the joint reasons that they might tax worldview and might underspecify action scope — and triage which mechanism applies before proposing a fix.

Jonathan's case is unambiguous and remains the load-bearing evidence for this ADR: role-level worldview tax (player positioned as deferring to faith-based authority) produced explicit frame-friction that delayed engagement with the decision. The role-level surface is what Guardrail 4 is centrally about; the option-level surface is a watch-item.

The scenario produced rich multiplayer signal — but at the peer layer, not the DTA layer¶

A complication worth recording: this scenario produced unusually rich peer-prediction signal, but the richness came partly from the frame-friction, not purely from the underlying decision substrate. The DTAs were not confused — both predicted their own player correctly despite the worldview tax, which is some evidence that the structured profile is more robust to frame-friction than the in-the-moment player experience suggests. The miss was at the human-peer-prediction layer (Jonathan predicted Cici would pick "technical roles only" based on her tendency to "make decisions smart within the rules" — but Cici reframed the rule as not applying to her chosen action). Jonathan then revised mid-discussion from "work-first probation" to "negotiate terms" after Cici walked through her information-gathering reframing: "if it's true that it's just information gathering, then I would also change my mind." This is exactly the kind of event the existing Stand/Revise mechanic (ADR-021) and the persuadability + peer-reading signals (ADR-024) are designed to capture — and they did.

The open question — "is there an additional signal we should be capturing here?" — probably resolves to no. The substantive parts (peer-prediction miss, revise event, the reasoning that triggered it) are already instrumented. The piece that's only visible because of the frame-friction is "what happens when a player rejects the option's stated risk on its premise rather than its substance" — and that's noise we'd rather eliminate than measure. A reframed Stonemark, with Cici's information-gathering rationale still intact, would produce the same peer-prediction divergence and the same revise event through the existing signal pipeline. We should not preserve the frame-tax in pursuit of capturing a meta-signal that the existing instrumentation already covers in cleaner form.

Audit note — option-level framing¶

This ADR was originally filed as Proposed, pending a corpus audit to size the scope and cost of the fix. That audit has since run and its fixes shipped — Stonemark plus the other embedded-setting arcs were reframed (migrations 096–098) — so the ADR is now Accepted.

One finding worth keeping: option-level framing is harder to catch automatically than role-level framing because it lives in short Risk: and Cost: clauses scattered across hundreds of options. The audit sampled option text for authority-relative phrasing ("bypasses X process," "violates X tradition," "undermines X's role") and flagged candidates for rewriting in process-relative terms.

Consequences¶

Adds Guardrail 4 to the scenario authoring contract alongside ADR-027's three guardrails. Future scenario generation prompts and review checklists should include the worldview-tax test.
Triggers a corpus audit of existing scenarios — especially campaign chapters with embedded settings (faith-led communities, monarchies, sworn orders, ideological enclaves). The audit must cover both surfaces: role framing (who the player is) and option framing (especially Risk: / Cost: lines that presuppose the surrounding authority structure is legitimate). Target output: a list of scenarios needing role reframing and a list of options needing risk/cost rewriting, with proposed alternatives for each.
Stonemark Chapter 1 is the first concrete revision target — reframe the player from "settlement member under Crane" to a leader-cast position (e.g., "Crane has stepped back from the council for the season; the decision about the engineers falls to you") rather than a pure outside advisor. Decision substrate unchanged; stakes preserved; worldview tax dropped.
Does not restrict embedded-worldview settings themselves. Faith-led communities, monarchies, and ideological enclaves remain valid scenario settings — they just shouldn't require the player to endorse the frame to engage.
Watch for false positives. Some role framings (e.g., "you are a doctor," "you are a parent") carry implicit values most players share; these are not worldview taxes. The test is whether a meaningful fraction of modern players would need to suspend personal disbelief — not whether any player might.
Interaction with ADR-024 signals. Decision conviction and timing signals collected on flagged scenarios pre-audit should be treated as lower-confidence; the friction inflates "tough call" and slows response in ways that don't reflect the decision itself.

Key files: - docs/decisions/027-scenario-content-guardrails.md — the parent ADR this extends - src/lib/ai/scenario-themes.ts — generation prompts that should incorporate Guardrail 4 - Stonemark Chapter 1 scenario record (audit pending) — first concrete fix - Corpus audit output (TBD) — scope of affected scenarios