Skip to content

Research

Working research that informs design decisions but isn't itself a decision or specification. ADRs and briefs cite these documents — they're load-bearing context, not history. (Truly dead iteration history lives in ../archive/.)

Organized into subfolders: foundations/ (the stable models) · experiments/ · audits/ · evaluations/ · briefs/ (briefs + open questions).

Start here

Two clusters do most of the conceptual heavy-lifting, and each has a plain-language entry point with a diagram — read these before the canonical docs:

How research feeds the record

flowchart LR
  subgraph FOUND["foundations/"]
    MS["Measurement Schema<br/>(+ Reader's Guide)"]
    CUBE["Scenario Dimension<br/>Cube"]
  end
  subgraph WORK["experiments · audits · evaluations"]
    EXP["Experiments<br/>money-subscore · new-domain<br/>conversation-capture · cross-domain"]
    AUD["Audits<br/>conversation-decisions · corpus"]
    EVAL["Evaluations<br/>sync-effect · inline-eval"]
  end
  subgraph OUT["The record"]
    ADR["ADRs"]
    BRIEF["briefs/"]
    PAPER["Light-paper<br/>positioning"]
  end
  MS --> ADR
  CUBE --> ADR
  EXP --> ADR
  AUD --> ADR & BRIEF
  EVAL --> ADR
  MS --> PAPER
  CUBE --> BRIEF

Arrows point from research to the artifacts that cite it. Research is upstream of the record, never the reverse.

Contents

foundations/ — the stable models

Doc What it is
measurement-schema-readers-guide.md Start here for measurement. Plain-language, 5-layer diagram, full glossary.
measurement-schema.md Canonical primitive enumeration (70 primitives across 5 layers). Stable.
📊 measurement-schema.html Interactive schematic — clickable primitives, search, filters, pipeline overview. Hand-built for the team.
scenario-dimension-cube.md Canonical cube: 12 structural axes + 2 content axes that shape every scenario. Stable.
📊 scenario-dimension-cube.html Interactive schematic — clickable axes, search, layer filters, inspect drawer. Hand-built for the team.
🎛️ cube-console.html Interactive console — the keeper. Shape a decision with the cube levers; the scenario rewrites (changed words flash). Switch the surface (Career/Health/Emergency/Sport…) and the options re-word while the drivers + signals hold; pull a lever and the signals move. Precomputed, no LLM.
🧭 explore.html Hub — front door to these explorers, grouped References (schema, cube) vs Explore (the console).

experiments/

Doc What it is
experiment-money-subscore.md Closed at N=1 — surfaced the alignment-vs-opposition design constraint. Retracts the Reading A tilt.
experiment-new-domain-signals.md Scoped: do career/health/relationship scenarios yield signals beyond governance? Reasoning-style vs aspiration-elicitation.
experiment-conversation-decision-capture.md Scoped: can post-conversation extraction produce decision data at low enough friction to be routine? (ADR-039.)
cross-domain-consistency-experiment.md · cross-domain-experiment-scenarios.md Does judgment hold across domains? Design + the scenario set (stratify by C2).

audits/

Doc What it is
conversation-decisions-audit.md 88-item, 6-month audit of Claude Code history. Source for ADR-066–071.
scenario-corpus-audit-2026-05-30.md Corpus-shape audit (trigger imbalance, etc.).

evaluations/

Doc What it is
sync-effect-comparison-log.md With-Sync vs no-Sync side-by-side log. Direct evidence for ADR-047.
inline-eval-coachj-historical-decisions.md Directional with-/without-Sync comparison against CoachJ's real decisions.

briefs/ — briefs & open questions

Doc What it is
judgment-vs-normative-competence.md Open question: is judgment-as-process too narrow for delegated-action use cases?
multi-domain-category-taxonomy-brief.md Category-axis design feeding ADR-076.
oq1-oq2-nota-handling-brief.md "None of the above" handling — the two coupled open questions.
multiplayer-curated-starter.md Curated multiplayer starter corpus + rigor fixes.
openclaw-hermes-research.md Deep research on how people use personal AI agents. Source for ADR-070.
adr-024-classify-driver-backfill.md Backfill notes for the classify-driver pipeline.

When to add to this folder

Working research that (a) informs one or more ADRs / briefs, (b) stays relevant beyond the conversation that produced it, and (c) is not itself a decision or specification. Drop it in the matching subfolder (foundations/, experiments/, audits/, evaluations/, briefs/) and add a row above.

If it's just a historical record of what was tried and abandoned — or iteration notes whose conclusions have been promoted to a canonical doc — it belongs in ../archive/ instead.