Architecture Decision Records

This folder contains Architecture Decision Records (ADRs) for the Sync Game. Each record captures a significant technical decision, its rationale, and consequences.

How to read this section

There are 84 ADRs — too many to scan in number order (that's chronological, not conceptual). Two ways in:

• "What still needs me?" → By status — the open / parked / superseded set, separated from the 66 settled decisions.
• "Where does X live?" → the map and the themed index — the eight themes and how they feed each other.

By status — what still needs you

Two lenses on the same ADRs: this section answers "what do I still need to look at?"; the themed index answers "where does X live?". The 66 Accepted decisions are settled — in force, browse them by theme. The ones below are the live set.

🔵 Needs a decision or build🟡 Parked (Deferred)⚪ Superseded (skip unless tracing history)

Proposed or Recommended — not yet in force; each is waiting on a call or implementation.

#	Title	Status
011	Delegation map: single score → per-context authority	Recommended
039	Decision capture primitive — extend Sync beyond games	Proposed
040	Curated scenario packs	Proposed
041	Share twin — public read-only twin view	Proposed
042	Synapses — twin chrome metric	Proposed
049	Scheduler weighting by driver × trigger learning gaps	Proposed
062	Peer-prediction accuracy as relational trust	Proposed
064	Multiplayer keeps full difficulty; campaign is onboarding	Proposed
065	Surface behavioral attention signals as named dimensions	Proposed
077	Anti-gaming defenses for token-bearing surfaces	Proposed
086	Operator-side scenario-framing manipulation	Proposed

Decided not now — each is gated behind a named trigger; revisit when the trigger fires.

#	Title	Trigger to revisit
046	Defer team-scoped Sync Hermes skills	Real-user demand
050	Agent permission tiers gated by sync_score	Tiers go live / token externality
060	Defer custom-domain auth proxy	Repeat reachability reports
070	Per-category sync scores	Category-axis decision

Kept in place for the diagnostic trail; each points to what replaced it.

#	Title	Replaced by
010	Decision principles synthesis layer	ADR-011 (reframed)
052	Multiplayer generation constraints	ADR-053
067	No-bulk-audit policy	contributing guide
069	Case-study writeups as marketing artifact	brief

Map

The eight themes and how a decision in one feeds the next — scenarios get written, played, measured into a profile, then used:

flowchart TB
  subgraph AUTHOR["① Writing the scenarios"]
    direction LR
    T3["Scenario Content<br/>& Authoring · 13"]
    T4["Scenario Library<br/>& Selection · 5"]
  end
  subgraph PLAY["② Playing them"]
    direction LR
    T5["Campaign<br/>& Narrative · 7"]
    T6["Multiplayer<br/>& Teams · 8"]
  end
  subgraph MEASURE["③ Measuring judgment"]
    direction LR
    T2["Judgment Signals<br/>& Measurement · 11"]
    T1["Scoring<br/>& Decay · 10"]
  end
  subgraph ACT["④ Using the profile"]
    direction LR
    T7["Delegation, Trust<br/>& Agent · 12"]
    T8["Product Surfaces<br/>& Platform · 17"]
  end
  T3 --> T4 --> T5 & T6
  T5 & T6 --> T2 --> T1
  T1 --> T7 --> T8
  T2 --> T8

The number after each theme is how many ADRs it holds. Arrows show the decision pipeline, not strict dependencies.

How the decisions evolved

Most ADRs stand alone. These are the ones that moved — where a later decision superseded, simplified, extended, or amended an earlier one. (Everything else is found faster via the status view or the theme tables.)

flowchart LR
  n010["010 · principles layer"]:::sup -.->|reframed as| n011["011 · delegation map"]
  n016["016 · delegation action"] -->|simplified by| n021["021 · stand / change"]
  n013["013 · deliberative baseline"] -.->|partly superseded| n036["036 · procedural quality"]
  n052["052 · MP gen constraints"]:::sup -->|superseded by| n053["053 · worldview-split"]
  n033["033 · team layer"] -->|extended by| n066["066 · MP mechanics"]
  n062["062 · peer-prediction trust"] -.->|extended by| n077["077 · anti-gaming"]
  n077 -.->|inverse-direction complement| n086["086 · operator framing"]
  n074["074 · multi-domain layer"] -->|expanded by| n076["076 · category taxonomy"]
  n017["017 · none-of-the-above"] -.->|amended by| n082["082 · NoTA cleanup"]
  n078["078 · schema wave 1"] -->|continued in| n079["079 · schema wave 2"]
  classDef sup fill:#e8e2d2,stroke:#9a9da3,color:#9a9da3;

Greyed = superseded. The other ~70 ADRs carry no supersede/extend edge — they're independent decisions, best found via the tables.

Full graph — all 84 ADRs at once (dense; best on a wide screen)

The exhaustive view: every ADR as a node grouped by theme, superseded greyed. It's intentionally large and won't read well at column width — the diagram above and the tables below are the everyday tools.

flowchart TB
  classDef sup fill:#e8e2d2,stroke:#9a9da3,color:#9a9da3,stroke-dasharray:3 3;
  classDef prop fill:#fff,stroke:#f59e0b,color:#0e1116,stroke-dasharray:4 2;
  classDef def fill:#fff,stroke:#7a7d83,color:#5a5e66,stroke-dasharray:1 3;
  classDef rec fill:#dce3ff,stroke:#1b36f5,color:#0e1116;

  subgraph S1["① Scoring & Decay"]
    direction LR
    n001["001"]; n002["002"]; n003["003"]; n004["004"]; n006["006"]
    n030["030"]; n038["038"]; n044["044"]; n045["045"]; n051["051"]
  end
  subgraph S2["② Judgment Signals & Measurement"]
    direction LR
    n015["015"]; n024["024"]; n028["028"]; n043["043"]; n049["049"]
    n056["056"]; n057["057"]; n065["065"]; n078["078"]; n079["079"]; n080["080"]
  end
  subgraph S3["③ Scenario Content & Authoring"]
    direction LR
    n008["008"]; n018["018"]; n019["019"]; n027["027"]; n034["034"]
    n036["036"]; n037["037"]; n053["053"]; n054["054"]; n058["058"]
    n061["061"]; n063["063"]; n067["067"]
  end
  subgraph S4["④ Scenario Library & Selection"]
    direction LR
    n005["005"]; n009["009"]; n012["012"]; n040["040"]; n073["073"]
  end
  subgraph S5["⑤ Campaign & Narrative"]
    direction LR
    n020["020"]; n023["023"]; n025["025"]; n026["026"]; n029["029"]; n032["032"]; n035["035"]
  end
  subgraph S6["⑥ Multiplayer & Teams"]
    direction LR
    n033["033"]; n052["052"]; n059["059"]; n062["062"]; n064["064"]; n066["066"]; n068["068"]; n075["075"]
  end
  subgraph S7["⑦ Delegation, Trust & Agent"]
    direction LR
    n010["010"]; n011["011"]; n013["013"]; n014["014"]; n016["016"]; n021["021"]
    n046["046"]; n047["047"]; n050["050"]; n055["055"]; n071["071"]; n077["077"]
  end
  subgraph S8["⑧ Product Surfaces & Platform"]
    direction LR
    n007["007"]; n017["017"]; n022["022"]; n031["031"]; n039["039"]; n041["041"]; n042["042"]
    n048["048"]; n060["060"]; n069["069"]; n070["070"]; n072["072"]; n074["074"]; n076["076"]
    n081["081"]; n082["082"]; n084["084"]
  end

  n016 -->|simplified by| n021
  n052 -->|superseded by| n053
  n013 -.->|partly superseded| n036
  n010 -.->|reframed as| n011
  n033 -->|extended by| n066
  n078 -->|wave 1 → 2| n079
  n074 -->|expanded by| n076
  n062 -.->|extended by| n077
  n017 -.->|amended by| n082

  class n010,n052,n067,n069 sup;
  class n039,n040,n041,n042,n049,n062,n064,n065,n077 prop;
  class n046,n050,n060,n070 def;
  class n011 rec;

All ADRs through 084 are now on main. (080 → theme ②; 081 + 082 + 084 → theme ⑧.)

Index — by theme

① Scoring & Decay

How the sync score is predicted, computed, smoothed, and aged.

#	Title	Status
001	Two-stage prediction system	Accepted
002	AI-only sync scoring	Accepted
003	Sync score decay with quadratic vulnerability	Accepted
004	EMA smoothing for profile updates	Accepted
006	Confidence capping for new players	Accepted
030	AI predictions cached per (session × user) for cross-client consistency	Accepted
038	Upgrade primary AI model from Sonnet 4 → Sonnet 4.6	Accepted
044	Always run both predictors (rule + Claude)	Accepted
045	Three sync dimensions (Accuracy, Coverage, Readability)	Accepted
051	Slow sync_score decay ~5× for early-stage play cadence	Accepted

② Judgment Signals & Measurement

What gets measured from a player's choices and rationale — the behavioral signals and the schema that enumerates them.

#	Title	Status
015	Decision conviction as profile signal	Accepted
024	Judgment layer enhancement — 12 behavioral signals	Accepted
028	Implicit name-bias detection (Signal 13)	Accepted
043	Per-option driver + scenario trigger tagging	Accepted
049	Scenario scheduler weighting by user's driver × trigger learning gaps	Proposed
056	Record "why" requests as a behavioral trust signal	Accepted
057	Integrate post-reveal reaction and note into the prediction pipeline	Accepted (Phase A)
065	Surface behavioral attention signals as named user-visible dimensions	Proposed
078	Measurement-schema closure pass (Wave 1)	Accepted
079	Measurement-schema closure pass (Wave 2)	Accepted
080	Central-extraction single-points-of-failure — dependency graph + mitigation plan	Accepted

③ Scenario Content & Authoring

Voice, clarity, and the discipline rules that govern how scenarios are written.

#	Title	Status
008	Scenario labels as concrete actions	Accepted
018	Content audit: leading language and speed round "Depends"	Accepted
019	Speed round question quality pass	Accepted
027	Scenario content guardrails — avoid normalizing universally condemned practices	Accepted
034	Case-study voice as the scenario voice contract	Accepted
036	Sync measures procedural quality and legibility, not values or uniformity	Accepted
037	Constraint discipline as a scenario authoring pattern	Accepted
053	Sharpen Principle #6 globally with the worldview-split test	Accepted
054	Scenario clarity — stakes, jargon translation, and decision-prompt concreteness	Accepted
058	Role framing as a second axis of identity capture — audit for worldview-tax	Accepted
061	Scenario generation quality — pre-commitment planning block	Accepted
063	Cast block and scan strip clarity rules	Accepted
067	No-bulk-audit policy for scenario corpus work	Superseded → contributing guide

④ Scenario Library & Selection

The corpus itself: packs, balancing, and which scenario a player sees next.

#	Title	Status
005	Category balancing via least-played selection	Accepted
009	Speed round feeds player profile at reduced weight	Accepted
012	Semantic context trigger classification	Accepted
040	Curated scenario packs	Proposed
073	Starter pack scenario design	Accepted

⑤ Campaign & Narrative

Multi-session story play: arcs, narrative flow, progression, and naming.

#	Title	Status
020	Campaign arc system — narrative-driven multi-session play	Accepted
023	Campaign narrative flow	Accepted
025	Remove stakes multiplier from campaign scoring	Accepted
026	Reactive campaign arcs — decision-driven narrative and scenario selection	Accepted
029	Chapter synopsis as dual-source field	Accepted
032	Progression gating for Campaign and World	Accepted
035	Rename Worlds → Campaigns and Campaigns → Stories	Accepted

⑥ Multiplayer & Teams

Shared play: the team primitive, presence, peer prediction, and multiplayer scenario rules.

#	Title	Status
033	Team-level judgment layer & team primitive	Accepted
052	Multiplayer constraints apply to all multiplayer generation, not just team-themed	Superseded
059	Multiplayer presence model and leave/rejoin UX	Accepted (Phase 1)
062	Peer-prediction accuracy as relational trust (not sync_score)	Proposed
064	Multiplayer keeps full difficulty; campaign is the onboarding ramp	Proposed
066	Multiplayer mechanics extending ADR-033	Accepted (retroactive)
068	Multiplayer scenario quality parity with solo	Accepted (retroactive)
075	Re-introduce curated multiplayer scenarios via is_multiplayer gate	Accepted

⑦ Delegation, Trust & Agent

How the profile becomes delegated authority — the agent skill, permission tiers, and anti-gaming.

#	Title	Status
010	Decision principles synthesis layer — deferred	Superseded
011	Delegation map: from single score to per-context authority	Recommended
013	Deliberative baseline: training the DTA on your higher self	Accepted
014	Relational proximity as decision modifier	Accepted
016	Delegation as a first-class game action	Accepted
021	Stand/Change simplification (partially supersedes ADR-016)	Accepted
046	Defer team-scoped Sync Hermes skills until real-user demand	Deferred
047	Defer SKILL.md instruction-tuning until team-usage feedback signal	Accepted
050	Agent permission tiers gated by sync_score	Deferred
055	Sync agent skill surfaces only profile-derived content - situational reasoning is user-initiated	Accepted
071	Profile embedding across agent contexts via Sync MCP	Accepted
077	Anti-gaming defenses for token-bearing governance surfaces	Proposed
086	Operator-side scenario-framing manipulation (coordination-substrate influence risk)	Proposed

⑧ Product Surfaces & Platform

Everything the user touches plus the platform decisions: privacy, auth, twin, intake, and multi-domain positioning.

#	Title	Status
007	Privacy model: readability transparent, trust private	Accepted
017	"None of the above" as a first-class choice	Accepted
022	Expand authentication beyond magic links	Accepted
031	Persistent dashboard sections over one-shot modals for orientation	Accepted
039	Decision capture primitive — extend Sync beyond game scenarios	Proposed
041	Share twin — public read-only twin view	Proposed
042	Synapses — twin chrome metric	Proposed
048	Voice-to-text for rationale inputs via Web Speech API	Accepted
060	Defer custom-domain auth proxy until users or repeat reachability reports	Deferred
069	Case-study writeups as a core marketing artifact	Superseded → brief
070	Per-category sync scores — deferred pending category-axis decision	Deferred
072	Remove intake assessment	Accepted
074	Pulse as a multi-domain judgment layer	Accepted
076	Multi-domain category taxonomy expansion (personal-choice + relational-decision)	Accepted
081	Commit conviction before rationale in the solo choosing flow	Accepted
082	NoTA flow cleanup — rationale unification, genuine-indifference path, symmetric reveal	Accepted
084	RLS write policies for prediction_attestations	Accepted

⑨ Meta — docs, process & tooling

Decisions about how we record decisions and run the docs themselves.

#	Title	Status
083	Docs-site visual layer (Mermaid) and research consolidation	Accepted

Format

Each ADR follows this structure:

# ADR-NNN: Title

**Status:** Accepted | Deferred | Superseded
**Date:** YYYY-MM-DD
**Context:** What prompted this decision

## Decision
## Rationale
## Alternatives Considered
## Consequences

Adding a New ADR

1. Use the next available number — check active branches and the vault, not just main (080–084 were each allocated on separate branches before merging, so main can lag the true count)
2. Start from template.md and fill in all sections — the "Rationale" and "Alternatives Considered" sections are the most valuable for future readers
3. Add the "At a glance" box (the template includes it): a plain-language "what it decides" line + status/trigger + 3–5 bold-led bullets — and a small mermaid diagram if the decision has real structure (skip it for flat decisions). This is required on every ADR; see ADR-077 or ADR-050 for the pattern
4. Reference the source code files where the decision is implemented
5. Add the row to the matching theme table above (not a flat bottom-of-list append) and the By-status view if it's not Accepted; if the decision supersedes, extends, or amends another, add the edge to the evolution diagram