Sync Design Brief — April 1, 2026¶

Context¶

Sync is a multiplayer decision-making game that generates behavioral profiles used to train each player's Digital Twin Agent (DTA). It runs on Next.js, Supabase, and the Claude API. There are 16 Architecture Decision Records (ADRs) documenting design decisions to date.

What We've Established¶

Sync is effectively two products sharing one engine:

Multiplayer Sync — the near-term product. Value is immediate and felt within a single session. The hook is the structured debrief after players reveal their choices and reasoning. ADR-016 (Stand/Revise/Delegate) is shipped and gives sessions a tangible output. Target audience is teams — dev teams, DAOs, organizations navigating AI adoption together. Does not require the DTA story to be valuable.

Solo Sync — the longer-term DTA training tool. Value compounds over 40+ rounds. No strong hook in round one. Target audience is a smaller subset: researchers, self-knowledge motivated players, academics. Feeds the Beacon infrastructure vision.

Recently Shipped¶

ADR-016 (Stand/Revise/Delegate post-discussion action) — live
ADR-017 (None of the Above with closest-pick + required rationale) — live, solo AND multiplayer
ADR-018 (Content audit: leading language and speed round "Depends") — live
Pattern Divergence display (Q7) — live, shows why AI expected something different on misprediction
Trigger taxonomy expansion (Q6) — live, 4 new triggers: competitive_pressure, innovation_challenge, authority_threat, relational_proximity
Scenario presupposition ban — live (prompt principle #10)
AI-era realism — live (prompt principle #11, refined for physical vs software timelines)
Scenario-specific trust round — live, trust question now scenario-specific instead of category-level
Peer predictions reframe — live, "How Well the Group Reads Each Other" with group aggregate
Self-predictability card — live, shows your own readability stats on results screen
RLS policies updated for post_discussion status — live
Test harness at /dev/test-post-discussion — live

Open Design Questions¶

All March 23 all-hands design items (Q1-Q9) have been either shipped or explicitly deferred. The remaining open questions are GTM-related (see below).

Deferred Items — No Action Needed Now¶

Item	Status	Details
Personal vs. business scenario track (Q1)	Narrowed: solo-only, Phase 2	Standalone design brief — updated with two-product framing
Self-reporting bias mitigations (Q4)	Monitoring only	Existing mitigations (speed round, conviction weighting, EMA) may be sufficient
DTA reputation layer (Q8)	Deferred	Needs 10+ multiplayer sessions with 4+ players
Scaling / fast-track (Q9)	Deferred, Phase 2+	Needs 100+ games across multiple players for calibration

Upcoming GTM Context That May Affect Design¶

The near-term go-to-market is multiplayer sessions with outside teams — dev teams, small organizations, DAO communities. Sessions are 45 minutes, scenarios relevant to their context, structured debrief using ADR-016.

This means three things worth reviewing:

Scenario context inputs — Scenario generation needs to handle context inputs cleanly (company type, team size, domain) so sessions feel relevant to outside teams rather than defaulting to crypto/DAO scenarios.

Cold start UX — The session experience for a first-time player with no profile history needs to work. Confidence capping (ADR-006) already handles the DTA side, but the UX for someone who has never played before may need review.

Mobile stability — Razvan consistently hit errors on mobile during multiplayer sessions. Needs a fix before outside sessions begin.

Key Files for Reference¶

File	Purpose
`src/lib/ai/player-model.ts`	Profile update logic, trigger detection, EMA
`src/lib/ai/scenario-generator.ts`	Scenario generation and category balancing
`src/app/api/ai/classify-driver/route.ts`	Semantic classifier — extend for new triggers
`src/lib/sync/multi-dimensional-score.ts`	Sync score dimensions
`src/app/play/[sessionId]/page.tsx`	Multiplayer game flow including ADR-016