Design Brief: Themed Scenario Worlds¶

Date: 2026-04-07 Status: Proposed — Ready for team review and prototype Related: April 2026 Design Brief, Personal Scenarios Brief, ADR-018 (Content Audit)

The Problem¶

Solo Sync has no engagement hook. Jonathan (56 games) describes it as something he has to force himself to do. The scenarios are well-crafted organizational dilemmas, but they feel like a psych assessment — interesting the first few times, then repetitive in tone even when the content varies. Every scenario is "You're a [role] at a [organization]. Here's a crisis. Pick one of four options."

The game's value compounds over 40+ rounds, but players disengage before they get there.

The Proposal¶

Let players choose a thematic world for their solo scenarios, and progressively add narrative richness to test what actually moves the engagement needle. Same game structure, same 4-option format, same 12 dimensions tested, same driver classification throughout.

This is structured as a layered experiment — four variants that build on each other, each testable independently, each adding one variable. Build the cheapest one first. If it works, stop. If it doesn't, add the next layer.

Origin¶

This idea came from two team discussions (April 2 and April 6, 2026):

James (April 2): "What if one of the first questions are like, what's your favorite book? What's your favorite movie? And then the AI creates a fictional scenario based off of an interest that the user already has."
James (April 6, evolved): "A broader context or universe that has an ongoing storyline... you can have little side stories or asides."
Natascha (April 6): "Maybe we can use D&D or roleplay as a template... in the onboarding, we can ask what's the interest and make the game look different for anyone."
Jonathan (April 6): "How do we make it as personalized as possible? I'm maybe not into fantasy, I don't play D&D, but maybe I'm really into sci-fi."
Dunia.gg reference (April 6): Jonathan showed the team a text-based AI RPG by a former Gitcoin colleague — full conversational narrative with player-driven branching. The team responded positively to the experience but acknowledged the tension: "lots of text is always difficult" and the open-ended format doesn't produce structured behavioral signal.

The Layered Experiment¶

Each variant adds one variable on top of the previous. All four use the same 4-option choice structure, feed the same profile engine, and require no new database tables or game mechanics. They are all prompt engineering + context passing — the scenario generator gets more context, and writes differently, but the downstream pipeline is identical.

Variant A1: Themed Setting (just re-skin)¶

What changes: The scenario is set in the player's chosen world (sci-fi, post-collapse, political intrigue, etc.) instead of a business/org context. The scenario description follows the same format as today: role statement, situation breakdown with bold headers, named characters, stakes, 4 options.

What it tests: Does the setting matter for engagement? If a player who loves sci-fi gets sci-fi scenarios, do they play more often and enjoy it more?

Implementation: Add a theme field to ScenarioGenerationContext. Append a "World Setting" section to the generation prompt when theme ≠ boardroom. Store theme preference in intake_profiles.demographics JSONB (no migration needed).

Example: The Kepler Station oxygen crisis scenario from the earlier discussion — organizational structure, named characters, specific numbers, but on a space station instead of in a boardroom.

Variant A2: Themed Setting + Narrative Voice¶

What changes: Same as A1, but the scenario description is written as a scene you walk into rather than a briefing you read. Characters speak in dialogue. There's sensory detail. The tone matches the world. The 4 options still appear as concrete action labels at the end — the narrative is the framing, not the choice.

What it tests: Does presentation style matter beyond just the setting? Is the engagement difference between "business briefing" and "narrative scene" significant?

Implementation: Add a narrativeVoice: boolean flag to the generation context. When true, add to the prompt:

Write the scenario as a SCENE, not a briefing. The player walks into a situation already in motion.
Use dialogue, sensory detail, and tension. Characters speak — use direct quotes that reveal
their positions. The tone should match the world (gritty for Underground, formal for Court, etc.).

Structure: Open with a moment (what the player sees/hears when they arrive), then let 2-3
characters establish the conflict through dialogue and action, then end with the decision prompt.

The 4 options still appear as concrete action labels with consequence descriptions — do NOT
change the option format. The narrative is the FRAMING, not the choice.

Keep the total scenario under 300 words — this must be scannable on a phone screen.

Example (Frontier + narrative voice):

The alert klaxon cuts out mid-cycle. Chief Engineer Tomoko is standing at the viewport with her arms crossed when you arrive on the bridge. "Two of three recyclers," she says, without turning around. "Six weeks to fabricate parts. We sustain 220 at current consumption. We have 340."

Commander Adesanya slides a tablet across the table. "Evacuation order. Non-essential personnel on the next shuttle. Nine days."

Dr. Vasić pushes through the door behind you, already shaking her head. "I've run the rationing simulation. Thirty percent reduction, everyone stays, nobody gets separated."

Kwesi, the labor rep, leans against the wall. "My people built this station. If anyone decides who leaves, it should be them — not some commander who's never cleaned a water filter."

They're all looking at you.

Then the 4 options appear exactly as they would in A1.

Key constraint: 300 words max. Mobile readability is non-negotiable. The Dunia experience is beautiful on desktop with long prose, but Sync players are often on phones. The narrative needs to be tight.

Variant A3: Themed Setting + Recurring World¶

What changes: Same as A1 or A2, but the scenario generator receives the player's last 3 themed scenario summaries and choices as context. The AI can reference the same world — recurring locations, characters who appeared before, institutions the player has interacted with — without being mechanically constrained by prior decisions.

This is narrative continuity, not game state. The AI might write "Tomoko, whose recycler fix you approved last week, is now asking for something bigger." But there's no world-state database tracking what Tomoko's recycler status is. The AI uses the prior scenarios as creative context, not as binding constraints.

What it tests: Does familiarity and world persistence increase engagement? Do players feel more invested when they recognize characters and places from prior rounds?

Implementation: The solo page already fetches the player's recent choices (last 50, via formatRecentChoices in prompt-builder.ts). For scenario generation, pass the last 3 themed scenario descriptions + chosen options as additional context:

## World Continuity (optional — reference if natural, ignore if not)

This player has previously encountered these situations in this world:

1. "The Kepler Station Oxygen Vote" — Player chose: Station-wide 30% oxygen rationing
2. "The Freighter Dispute" — Player chose: Let the labor council arbitrate
3. "Signal from Proxima" — Player chose: Disclose the signal to all station residents

Characters, locations, and institutions from these scenarios may recur. Reference them
naturally if it enriches the new scenario, but the new scenario must stand alone — a reader
who hasn't played the prior rounds should still fully understand the situation.

Do NOT make the new scenario a direct sequel or consequence of a prior choice. The world
is persistent; the plot is not.

Risk to watch: If the AI leans too hard on continuity, new scenarios might feel like sequels rather than standalone dilemmas. The prompt instruction "must stand alone" guards against this, but it needs testing.

Why this isn't "narrative arcs" (Idea B): True narrative arcs mean Round 5 is mechanically determined by Round 2's choice — the world changes based on your decisions, and those changes constrain future scenarios. A3 is softer: the world has memory, but your decisions don't have mechanical consequences. The AI might reference your prior choice, but it's not computing "because you rationed oxygen, morale dropped 15%, which means..."

Variant B-lite: Themed Setting + Lightweight Consequences¶

What changes: Same as A3, but the scenario generator is told to treat the player's prior choices as having had consequences. The prompt includes guidance like:

The player's prior choices HAVE affected this world. When generating the new scenario,
imagine realistic consequences of their past decisions and weave them into the situation.

Example: If the player previously chose to ration oxygen rather than evacuate, the new
scenario might open with low morale, a petition from families, or unexpected health effects
from weeks of reduced oxygen — but also with the research teams still intact and productive.

Consequences should feel natural, not punitive. Every prior choice had trade-offs; show both
the benefits and costs that emerged. This gives the player the feeling that their decisions
MATTER without requiring a world-state simulation.

What it tests: Do stakes that compound increase engagement? Does it feel different when you know your next scenario will reference what you chose today?

Implementation: Same context passing as A3. The difference is purely in the prompt instructions — telling the AI to invent plausible consequences rather than just referencing prior scenarios. No world-state database. No consequence tracking table. The AI does the work.

Risks: - The AI might invent consequences that feel unfair or arbitrary. "Because you chose to ration, three people died" would feel punishing and might cause the player to optimize for "safe" choices rather than authentic ones. The prompt needs to be explicit: consequences are trade-offs, not punishments. - Players might start optimizing for narrative outcomes. "I need to fix the morale problem I created" is a different mental frame than "what do I actually believe about this dilemma." This is the main signal quality risk across all variants, and it's strongest here. - Narrative coherence burden on Claude. With 3 prior scenarios, Claude needs to maintain a consistent world. It's good at this, but errors (contradictions, forgotten characters) will break immersion. The 3-scenario limit helps.

Where These Stop Being Sync¶

There's a clear boundary:

	Still Sync	Why
A1 (themed setting)	Yes	Same structure, different paint
A2 (narrative voice)	Yes	Same structure, different writing style
A3 (recurring world)	Yes	Same structure, world has soft memory
B-lite (lightweight consequences)	Yes, but at the edge	Same structure, but player motivation starts shifting from "express my values" to "manage my world"
Full narrative arcs	Borderline	If consequences are mechanistic (world-state database, branching paths), the game is fundamentally different
Open narrative (Dunia-style)	No — this is a different product	Free-text responses, no structured options, no driver classification. The profile engine can't process this.

The Separate Game Question¶

The Dunia-style experience that Jonathan showed the team is genuinely compelling. But it produces a fundamentally different type of signal:

Sync produces: structured choice data → driver distributions → context rules → predictive profile. Each decision maps to measurable dimensions. This is what makes the DTA scientifically grounded.
Open narrative produces: free-text behavioral data → tone analysis, decision tendencies, communication style, narrative preferences. Rich but unstructured. You could extract signal from it, but it would be a different kind of signal — more personality than decision pattern.

If the long-term goal is a truly representative digital twin, you probably want both signal types eventually. Sync tells the DTA what you decide. An open narrative game would tell the DTA how you think, speak, and relate to people. These are complementary instruments, not competitors.

But building a second game right now would split focus across 2 products for 5 players. The smarter path: 1. Test A1 through B-lite in Sync (weeks, not months) 2. If engagement moves → you've solved the solo problem within Sync, no second game needed 3. If engagement doesn't move even with B-lite → the problem is the 4-option structure itself, and a Dunia-style open game becomes the next experiment 4. If you do build a second game later, the DTA could consume signal from both — Sync choices feed driver distributions, narrative game feeds communication/personality layers

Recommendation: Don't build a separate game now. Test the layers. Let the data tell you if you need one.

What Changes (All Variants)¶

For the player¶

New preference — During onboarding (or anytime via profile/settings), players pick a scenario world:
Boardroom (default — current business/org scenarios, unchanged)
Frontier (space stations, colony ships, first-contact diplomacy)
Rebuild (post-collapse settlements, resource scarcity, rebuilding institutions)
Court & Council (political intrigue, guild governance, feudal power dynamics)
Underground (heist crews, resistance cells, grey-market networks)
Custom (free-text: player describes their world)
Solo scenarios use the theme — Themed scenarios are solo-only. Multiplayer stays as-is.
Variants are per-player settings — A player could be on A1 while another is on A2. This happens naturally since the only difference is prompt configuration.

For the AI/profile engine¶

Nothing changes across any variant. The 4-option structure is preserved. The driver classifier evaluates choice + rationale against the same 8 drivers and 10 triggers. Category patterns, context rules, keyword preferences, EMA smoothing, confidence calibration — all untouched.

The prediction prompt (src/lib/ai/prompt-builder.ts) already works off the player's history of choices and rationales. It doesn't know or care what world the scenario was set in. A player who consistently chooses team_harmony in Frontier scenarios will be predicted to choose team_harmony in the next Frontier scenario, just as they would in Boardroom.

Technical Implementation¶

Core change (all variants)¶

The scenario generator (src/lib/ai/scenario-generator.ts) gets a new optional field:

export interface ScenarioGenerationContext {
  category?: ScenarioCategory;
  avoidSimilarTo?: string[];
  playerHistory?: { ... };
  playerContext?: { sectors?: string[] };
  // NEW — themed scenario support
  theme?: {
    id: string;                    // 'boardroom' | 'frontier' | 'rebuild' | 'court' | 'underground' | 'custom'
    description?: string;          // For custom themes — player's free-text world description
    narrativeVoice?: boolean;      // A2+: write as scene vs briefing
    worldContinuity?: Array<{      // A3+: prior scenarios for world memory
      title: string;
      description: string;         // Abbreviated — first 200 chars
      chosenOptionLabel: string;
    }>;
    consequenceMode?: boolean;     // B-lite: treat prior choices as having had consequences
  };
}

Storage¶

Theme preference stored in existing intake_profiles.demographics JSONB:

{
  "sector": ["Startups", "DAOs & crypto"],
  "scenario_theme": "frontier",
  "scenario_voice": "narrative"
}

No migration needed.

Prompt construction¶

buildScenarioGenerationPrompt() appends sections based on which theme fields are present:

theme.id ≠ boardroom → append World Setting section
theme.narrativeVoice = true → append Narrative Voice section
theme.worldContinuity populated → append World Continuity section
theme.consequenceMode = true → append Consequence Mode section

Each section is additive. The base prompt (12 principles, 4-option structure, banned patterns) is unchanged.

Prompt principle carve-out¶

Principle #11 (AI-era realism) needs a carve-out for non-contemporary themes:

Exception: If the scenario is set in a fictional world (sci-fi, fantasy, post-collapse, historical),
AI-era realism constraints do not apply. The world's technology level is defined by its setting.

All other principles (no presuppositions, concrete labels, balanced options, etc.) apply universally.

Flow (solo page)¶

Solo page loads → reads intake_profiles.demographics.scenario_theme and scenario_voice
If A3+: fetches last 3 solo game scenarios with theme matching current theme
Passes theme config to scenario generation API
Scenario generator builds prompt with theme sections
Generated scenario returned and displayed
Player chooses → classification → profile update (completely unchanged)

Where theme is NOT passed¶

Multiplayer session start — no change
Speed rounds — no change
Prediction prompts — no change (predictions use profile + history, not scenario setting)

Predefined Themes¶

Theme ID	Display Name	Description (passed to Claude)	Why it works for Sync
`boardroom`	Boardroom (default)	Current-era organizational and professional dilemmas: startups, nonprofits, corporations, governments, DAOs.	Status quo. Proven signal quality.
`frontier`	Frontier	Deep space, colony stations, first-contact diplomacy, resource-scarce outposts, interstellar governance. Think Expanse, Star Trek, Foundation.	Rich institutional settings. Resource allocation and governance dilemmas map naturally.
`rebuild`	Rebuild	Post-collapse settlements rebuilding civilization. Scarce resources, fragile alliances, competing visions for what comes next. Think Station Eleven, The Last of Us, Fallout.	Every decision is high-stakes. Team dynamics and values-culture dilemmas are visceral.
`court`	Court & Council	Political intrigue, courtly maneuvering, guild governance, feudal power dynamics. Think Game of Thrones, Shogun, Dune.	Governance and delegation dilemmas are native to this setting.
`underground`	Underground	Heist crews, resistance cells, grey-market operations, investigative journalism. Think The Wire, Severance, Mr. Robot.	Trust, transparency, and control-vs-delegation are the core tensions.
`custom`	Your World	Player provides a free-text description (min 20 characters). The AI uses it as the world context. Guidance: "Choose a world with organizations, power structures, and competing interests."	Maximum personalization. Higher variance in quality.

Validation Plan¶

With 5 active players, validation is qualitative — you're testing whether you like it, not running statistical analysis.

Phase 1: A1 vs A2 (Week 1)¶

Build theme preference + both prompt variants (A1 standard, A2 narrative voice)
Each player picks a theme. Suggestion: at least 1 player stays on Boardroom as informal control
Half the group starts on A1, half on A2 (or alternate — play 5 rounds A1, switch to A2 for 5 rounds)
After 10 rounds, discuss:
Did you play more than usual this week? Did it feel less like a chore?
A1 vs A2: which did you prefer? Was the narrative voice engaging or just more text to read on mobile?
Did any scenario feel like you were "roleplaying a character" rather than answering as yourself?
Check prediction accuracy — any drop compared to pre-theme baseline?

Phase 2: A3 (Week 2, if Phase 1 is positive)¶

Enable world continuity — last 3 scenarios passed as context
Play 5-10 more rounds
Does recognizing characters/locations change the experience? Does it feel like "your world"?

Phase 3: B-lite (Week 3, if Phase 2 is positive)¶

Enable lightweight consequences
Play 5-10 rounds
Does it feel like your decisions matter more? Are you starting to think about "what will happen next" when you choose? (If yes — that's engagement. But also watch: are you choosing what's "good for the story" instead of what you believe?)

Decision gates¶

After Phase 1: If engagement doesn't improve → the problem isn't setting or presentation. Consider whether the 4-option format itself is the bottleneck (separate game territory).
After Phase 2: If recurring world doesn't add much → skip B-lite, the value is in the theme itself, not the continuity.
After Phase 3: If B-lite works → you have a compelling solo experience. If players are distorting their choices for narrative outcomes → pull back to A3 (continuity without consequences).

Risks¶

Signal quality degradation¶

The main risk across all variants. Themed/narrative scenarios must still produce clean driver signals.

Mitigation: The driver classifier (src/app/api/ai/classify-driver/route.ts) works on choice + rationale, not scenario setting. As long as options represent distinct worldviews (enforced by the generation prompt), classification works identically. Test by comparing prediction accuracy pre/post theme introduction.

"Roleplaying" vs authentic decision-making¶

Players might think "what would my character do?" instead of "what would I do?" This risk increases with each variant — lowest at A1, highest at B-lite.

Mitigation: Watch for prediction accuracy drops (the AI knows the player's real patterns — if choices start diverging, accuracy will fall). Could also add a subtle prompt to the player: "Answer as yourself, not as a character" — though this might break immersion.

Counter-argument: Players are already roleplaying. "What would I do as VP of Engineering?" is also a projection. The question is whether themed worlds cause more projection than business scenarios. Worth measuring, not assuming.

Custom themes producing poor scenarios¶

"Underwater basket weaving" won't generate good governance dilemmas.

Mitigation: Custom theme requires min 20 characters and guidance: "Choose a world with organizations, power structures, and competing interests — the AI needs these to create meaningful dilemmas."

Mobile readability (especially A2)¶

Narrative voice means more text. The current scenario descriptions are already pushing the limit of what people read on phones.

Mitigation: Hard 300-word cap in the narrative voice prompt. For context: the current business scenarios average 200-250 words. A2 adds about 50-100 words of sensory/dialogue framing. It's tight but workable.

AI narrative coherence (A3 and B-lite)¶

Claude needs to maintain a consistent world across scenarios — same character names, same institutions, no contradictions.

Mitigation: Limit world memory to last 3 scenarios (keeps context window small). Include explicit instruction: "If referencing prior characters or locations, maintain consistency with the summaries provided."

Multiplayer convergence (future, not now)¶

Solo players building profiles in themed worlds will also play multiplayer in business settings. The profile translates because it's about values and decision patterns, not settings. A player who prioritizes team_harmony in Frontier scenarios will show the same pattern in Boardroom multiplayer. The theme is the container; the signal is setting-agnostic.

Early Finding: "Amplified Reality" Custom Themes¶

Date: 2026-04-07 (from April 7 multiplayer session)

The strongest engagement reaction across all testing came not from a genre theme (Frontier, Court, etc.) but from a custom theme based on the group's actual domain, pushed to dramatic extremes. James wrote a custom world: "post-apocalyptic world where every human has a personal agent, open-source LLMs collapsed the AI market, frontier models are self-interested, we're starting an alignment organization, trusting your agent is humanity's only hope."

The team called the resulting scenario "a spy thriller" and said "you could make a freaking movie." Meanwhile, Jonathan found Court & Council (medieval guilds/alliances) confusing and disengaging — too much unfamiliar context to parse before reaching the decision.

Implication: The engagement hook may not be "escape to a different world" but "imagine our stakes amplified." For multiplayer teams, the best custom themes might be near-future extrapolations of their actual work — a DAO team gets "post-collapse governance," a dev team gets "AI agents have replaced most engineers and you're deciding who stays," etc.

For multiplayer GTM: When pitching to outside teams, generating a custom world based on their domain (but pushed to extremes) may be more effective than offering genre themes. The team already cares about the stakes — the theme just dramatizes them.

This also connects to the personal-vs-business question (ADR-014, personal scenarios brief): themed scenarios naturally surface different decision patterns. James answered differently in the survival scenario than in business ones. This isn't noise — it's the profile engine capturing context-dependent behavior, which is exactly what the trigger system was built to track. Don't categorize themed data separately; tag it with the theme and let the profile accumulate all signals.

Multiplayer Campaigns (D&D Style) — Built¶

Date: 2026-04-08

The team's strongest recurring request is a persistent multiplayer narrative — when the same group plays together across sessions, the world should remember what happened. Like a D&D campaign where the AI is the dungeon master and every session picks up where the last one left off.

How it works¶

Host creates a campaign — gives it a name ("The Alignment Wars"), picks a theme (or writes a custom world), and creates the first session
Subsequent sessions link to the campaign — when the host creates a new game and selects an existing campaign, the theme is locked to the campaign's settings
World continuity — the scenario generator receives the last 5 completed sessions' scenarios and all players' choices as context. Characters, locations, and institutions recur. The AI is told prior decisions had consequences.
The group builds a shared story — each session advances the narrative. New dilemmas emerge from prior group decisions.

What's different from solo world memory¶

Solo world memory tracks one player's choices. Campaign continuity tracks the group's choices — all players' decisions are summarized for the generator.
Solo has separate toggles (world memory on/off, consequences on/off). Campaigns always have both enabled — continuity is the point.
Campaign theme is locked after creation. The group committed to this world.

Key files¶

migrations/029_campaigns.sql — campaigns table, campaign_id on sessions
src/app/play/page.tsx — Single Round / Campaign toggle, campaign creation and selection
src/app/api/sessions/start/route.ts — fetches campaign history, passes group choices as world continuity
src/app/play/[sessionId]/lobby/page.tsx — campaign badge

What This Brief Explicitly Defers¶

Idea	Why deferred	When to revisit
Open narrative / Dunia-style game	Different product, different signal type. Would split focus for 5 players across 2 products.	If themes + campaigns don't move engagement, the 4-option format may be the bottleneck. Then explore open narrative as a complementary signal source for the DTA.
Different game structures per player (Natascha's archetype idea)	Breaks profile comparability. Different structures produce incomparable data.	Revisit only if themed worlds prove insufficient AND you're willing to accept non-comparable profiles across players.
Personal scenario track	Orthogonal to themes. Can be layered on independently.	Per existing brief: after 50+ players validate demand.
MBTI/DISC integration	Unrelated to themes. Can be tested independently via synthetic data.	When you have enough game data to test correlation.
In-app voice/text chat	Would feed debrief data directly into the model instead of requiring external calls. James's suggestion, Amin suggested Agora for live audio.	Infrastructure-heavy. Defer until the game flow is stable and engagement is proven.
Video vignettes	AI-generated 2-minute scenario videos instead of text. Solves the "too much text" problem more directly than any toggle.	Waiting on cost/quality improvements in AI video generation.

Relationship to the DTA Vision¶

The underlying question from the April 6 meeting: "If we're thinking about AI digital twins and agent representation as the main goal, how does this all fit?"

The DTA needs to know how you make decisions. Sync's 4-option structure produces the cleanest signal for this: discrete choices → driver classification → probability distributions → predictive model. This is the DTA's decision layer — "in situation X with triggers Y, this person will choose Z with 85% confidence."

But a truly representative digital twin also needs: - Communication style — how you explain your reasoning, your vocabulary, your level of directness - Relationship patterns — who you trust, who you defer to, who you push back on - Emotional responses — how you react under stress vs calm, with strangers vs friends

Sync already captures some of this through rationale text, peer predictions, and the delegation mechanic (ADR-016). But an open narrative game (Dunia-style) would capture much more of the communication and emotional layers.

The architecture for this exists conceptually: the DTA could have multiple signal sources feeding different layers: - Sync choices → decision patterns (structured, high-confidence) - Sync rationales → reasoning style (semi-structured) - Multiplayer debriefs → social dynamics (if recorded/transcribed) - Open narrative game → communication personality (unstructured, rich)

But this is a Phase 3+ vision, not a this-month build. The immediate priority is making solo Sync engaging enough that 5 players want to keep playing. Themed scenarios are the cheapest test of that hypothesis.

Implementation Scope¶

What to build for Phase 1 (A1 + A2)¶

Add theme field to ScenarioGenerationContext interface
Add theme-aware prompt sections to buildScenarioGenerationPrompt()
Add theme preference to profile settings page (single select + optional custom text)
Store preference in intake_profiles.demographics JSONB
Solo page reads theme preference and passes to scenario generation
Add AI-era realism carve-out for non-contemporary themes

Estimated scope: Small. All changes are in the scenario generation prompt path. No database migrations. No changes to classification, scoring, or prediction.

What to add for Phase 2 (A3)¶

Solo page fetches last 3 themed scenarios for current theme
Pass summaries + chosen options as worldContinuity context to generator

Estimated scope: Trivial addition — the data is already available in the choices table.

What to add for Phase 3 (B-lite)¶

Add consequence mode prompt section
Add consequenceMode toggle to settings

Estimated scope: Trivial — one more prompt section.

Decision Needed¶

Should we build Phase 1 (A1 + A2)? The implementation is small. The test is 5 people playing 10 rounds each. The risk is low because Boardroom remains the default and nothing downstream changes. Each subsequent phase is a tiny increment that only happens if the previous phase shows engagement improvement.