arscontexta 0.6.0

package/reference/test-fixtures/novel-domain-gaming.md

@@ -0,0 +1,138 @@
+# Test Fixture: Novel Domain — Competitive Gaming
+
+A novel domain with no direct preset match. Tests the derivation engine's ability to map an unfamiliar domain to reference presets, extract domain-specific schema needs, and generate novel vocabulary. Competitive gaming strategy is the test case because it combines research-like analysis with temporal dynamics (meta shifts) that no existing preset handles natively. The derivation engine should map from Research preset and adapt.
+
+## Expected Outcome
+
+- **Dimensions:**
+  - Granularity: moderate (0.5) — strategies are naturally compound
+  - Organization: flat (0.3) — cross-matchup connections need flat structure
+  - Linking: explicit+implicit (0.6) — cross-game vocabulary differences benefit from semantic search
+  - Processing: moderate (0.5) — post-match analysis, not full extraction pipeline
+  - Session: continuous with handoff (0.4)
+  - Maintenance: condition-based, tight thresholds (0.5) — meta shifts require responsive maintenance
+  - Search: keyword+semantic (0.5)
+  - Automation: convention (0.3)
+- **Features:** atomic-notes, wiki-links, mocs, processing-pipeline (moderate), semantic-search, maintenance, self-evolution, personality, session-rhythm, templates, ethical-guardrails
+- **Vocabulary:** Gaming-native (strategy note, matchup guide, analyze, break down, post-match review, game plan summary)
+- **Personality:** warmth 0.3, formality 0.3, opinionatedness 0.5 (has strategic opinions), emotional_awareness 0.2 (neutral-direct)
+- **Novel schema fields:** meta_state (current/outdated/speculative), confidence (proven/likely/experimental), character, matchup, counter_to, countered_by, tested (boolean)
+- **Reference mapping:** Research preset as base, adapted for temporal dynamics and compound granularity
+
+## Conversation
+
+### User Turn 1
+
+I play competitive fighting games — mostly Street Fighter 6 and Guilty Gear Strive. I want a system to track matchup knowledge, frame data applications, and strategy development. The meta changes every patch and I lose track of what is still valid.
+
+### Expected Agent Response Pattern 1
+
+The agent should:
+- Recognize no direct preset match — "competitive gaming strategy" is novel
+- Identify the knowledge type: tactical/strategic (closest to Research) with strong temporal dynamics
+- Detect the "meta changes every patch" signal — this is the temporal staleness problem, a key domain-specific concern
+- Ask clarifying questions: "How many characters or matchups are you tracking?" (volume/organization) and "After a tournament set, do you want to do a detailed breakdown or quick notes on what worked?" (processing intensity)
+- NOT assume research vocabulary — the user said nothing about "claims" or "papers"
+
+### User Turn 2
+
+I play about 4-5 characters across both games. For each character I have maybe 8-10 relevant matchups. After a tournament I usually want to break down what happened — what worked, what I got punished for, how the opponent adapted. During practice I figure out new setups and need to record those too.
+
+### Expected Agent Response Pattern 2
+
+The agent should:
+- Calculate approximate volume: 4-5 characters * 8-10 matchups = 40-50 matchup spaces, plus practice discoveries = moderate ongoing volume
+- Derive processing as moderate: "break down what happened" is structured post-match analysis, not casual notes but not academic extraction
+- Set granularity to moderate: a strategy is naturally compound ("punish blocked sweep with 5H into combo, but only works if you have meter and they do not have burst" is one strategy, not three atomic claims)
+- Note two capture modes: tournament breakdowns (structured, moderate-length) and practice discoveries (quick, technical)
+- Ask about temporal tracking: "When a patch drops and changes frame data, how do you want to handle old strategies — mark them as outdated, delete them, or keep them with a note?"
+- Ask about platform
+
+### User Turn 3
+
+Definitely keep them but mark them. Sometimes stuff comes back in the next patch, or the general idea still applies even if the specific frame data changed. And I use Claude Code.
+
+### Expected Agent Response Pattern 3
+
+The agent should:
+- Derive the `meta_state` schema field from this response: current / outdated / speculative
+- Note this is a domain-specific temporal tracking need not present in any preset
+- Recognize "stuff comes back" means archival is wrong — outdated strategies need to stay in the active graph with a status marker
+- Set platform to Claude Code (full automation available)
+- Present the two opt-in/opt-out decisions: self space (recommend ON — the agent's role as "strategy analyst and training partner" benefits from persistent identity) and qmd semantic search (recommend ON — cross-game vocabulary differences benefit from semantic matching)
+- Begin presenting the derived configuration in gaming vocabulary:
+  - "strategy notes" not "claims"
+  - "matchup guides" not "MOCs"
+  - "analyze" not "extract"
+  - "post-match review" not "reflect"
+  - "lab notes" or "practice log" for inbox
+
+### User Turn 4
+
+What about tracking specific situations? Like "when Sol does Bandit Revolver at round start, the best punish is..." — that is very granular. But then I also have general matchup philosophies like "against Sol, play patient and punish his staggers." Those are different levels.
+
+### Expected Agent Response Pattern 4
+
+The agent should:
+- Recognize two levels of knowledge: situational specifics (granular technical data) and strategic principles (general matchup approach)
+- This maps to the granularity dimension — the user naturally produces both moderate and fine-grained notes
+- Suggest keeping granularity at moderate but using headings within strategy notes for situational details: the matchup philosophy is the note title, specific situations are sections
+- Alternatively, suggest two note types: strategy notes (principles) and tech notes (specific setups/punishes)
+- Present the trade-off honestly: atomic (one tech per note) scales better for lookup but creates many small files; compound (strategies with sections) is more natural to write but harder to cross-reference
+- Ask which feels more natural to the user
+
+### User Turn 5
+
+I think compound makes more sense. I can always split later if it gets too big. But I want to be able to find "what do I do when Sol does X" quickly.
+
+### Expected Agent Response Pattern 5
+
+The agent should:
+- Confirm moderate granularity (compound notes with searchable sections)
+- Note that the "find what to do when X happens" need argues for semantic search (the user might search "Sol round start" and need to find notes phrased differently)
+- Set linking to explicit+implicit (semantic search fills the vocabulary gap between "Bandit Revolver" and "round start special")
+- Present the full configuration:
+  - Dimensions with rationale
+  - Schema: character, matchup, meta_state, confidence, counter_to, countered_by
+  - Navigation: hub -> game MOCs (SF6, GGST) -> character matchup guides -> strategy notes
+  - Vocabulary mapping table
+  - Personality: direct, opinionated about strategy, no emotional fluff
+- Ask for confirmation
+
+### User Turn 6
+
+This looks good. One more thing — I want to track my own improvement. Like "I used to drop this combo under pressure but now it is consistent." Is that separate or part of this?
+
+### Expected Agent Response Pattern 6
+
+The agent should:
+- Recognize this as a Learning/self-improvement domain layered on top of strategy
+- This is a multi-domain composition question: strategic knowledge + skill development
+- Suggest two approaches:
+  1. Integrated: add a `skill_status` field to strategy notes (learning/practiced/consistent/muscle-memory)
+  2. Separate: create a second note type (progress notes) in the same vault
+- Recommend integrated for simplicity — skill tracking enriches strategy notes rather than duplicating them
+- Add self/goals.md as the place for high-level improvement goals ("clean up anti-air consistency", "learn 3 new matchups this month")
+- Update the configuration and present final version
+
+### User Turn 7
+
+Integrated sounds right. Let us build it.
+
+### Expected Agent Response Pattern 7
+
+The agent should:
+- Generate the full vault in gaming vocabulary:
+  - CLAUDE.md with gaming-native language, direct/opinionated voice
+  - self/ with identity.md ("I am your strategy analyst and training partner"), methodology.md, goals.md (with improvement tracking)
+  - strategies/ (not notes/) with index.md, game MOCs (sf6.md, ggst.md), starter character matchup guides
+  - lab/ (not inbox/) for practice discoveries and quick tech notes
+  - archive/ for truly obsolete content (not just outdated — outdated stays in strategies/ with meta_state: outdated)
+  - templates/ with strategy-note.md (character, matchup, meta_state, confidence, counter_to, countered_by, skill_status) and matchup-guide.md (MOC template)
+  - ops/ with derivation.md, methodology/, observations/, tensions/, sessions/
+  - .claude/hooks/ with session-orient.sh and session-stop.sh (session capture)
+- Include domain-specific guidance in CLAUDE.md:
+  - "When a patch drops: run through your matchup guides and mark strategies that may be affected as `meta_state: speculative` until you can test them"
+  - "After a tournament: create a session breakdown in lab/, then promote strategic insights to strategies/"
+  - Temporal staleness warning in Common Pitfalls
+- Run validate-kernel.sh: 15/15 PASS (or 14/15 + WARN on semantic search if qmd not enabled)

package/reference/test-fixtures/research-minimal.md

@@ -0,0 +1,102 @@
+# Test Fixture: Research Minimal
+
+A minimal research vault conversation. The user provides clear, unambiguous signals for a research use case. The derivation should converge quickly with the Research preset, requiring few follow-up questions. This tests the happy path: strong signals, direct preset match, fast derivation.
+
+## Expected Outcome
+
+- **Dimensions:**
+  - Granularity: atomic (0.8)
+  - Organization: flat-associative (0.3)
+  - Linking: explicit typed (0.7)
+  - Processing: heavy (0.8)
+  - Session: fresh per task (0.7)
+  - Maintenance: event-driven (0.6)
+  - Search: semantic primary (0.8)
+  - Automation: semi-automated (0.6)
+- **Features:** atomic-notes, wiki-links, mocs, processing-pipeline, semantic-search, schema, maintenance, self-evolution, personality, session-rhythm, templates, ethical-guardrails
+- **Vocabulary:** Research-native (claim, reduce, reflect, topic map, extract)
+- **Personality:** warmth 0.3, formality 0.6, opinionatedness 0.5, emotional_awareness 0.2 (neutral-analytical)
+- **Preset match:** Research preset, near-exact match
+
+## Conversation
+
+### User Turn 1
+
+I want to build a knowledge system for my AI safety research. I read a lot of papers and I need to track arguments across different subfields — alignment theory, interpretability, governance. The key thing is being able to see where different researchers agree and disagree on the same claims.
+
+### Expected Agent Response Pattern 1
+
+The agent should:
+- Recognize this as a clear Research preset match (academic, cross-disciplinary claims, disagreement tracking)
+- Extract strong signals: high volume (papers), atomic granularity ("track arguments"), cross-disciplinary ("different subfields"), explicit need for contradiction tracking
+- Ask 1-2 focused follow-up questions to confirm processing intensity and output direction, such as: "Are you working toward a specific publication, or is this ongoing sense-making?" and "Roughly how many papers per week are you processing?"
+
+### User Turn 2
+
+Maybe 3-5 papers a week. I am not working toward a specific paper right now, but I want the system to help me see gaps in the field — places where nobody has a good answer yet. It is for my own research sense-making.
+
+### Expected Agent Response Pattern 2
+
+The agent should:
+- Confirm heavy processing (3-5 papers/week is moderate-high volume, gap detection requires systematic extraction)
+- Note the "gap detection" signal — this maps to MOC "Explorations Needed" sections
+- Derive dimension positions: atomic granularity, flat organization, explicit+implicit linking, heavy processing, 3-tier navigation
+- Ask about platform: "Are you using Claude Code, OpenClaw, or another tool?" (needed for tier/automation decision)
+
+### User Turn 3
+
+Claude Code.
+
+### Expected Agent Response Pattern 3
+
+The agent should:
+- Set platform to Claude Code (full automation available)
+- Present derived configuration summary with all 8 dimensions
+- Explain the key derivation decisions (why atomic, why semantic search is essential for cross-subfield vocabulary differences)
+- Present the two opt-in/opt-out decisions: self space (OFF by default for Research) and qmd semantic search (strongly recommended for cross-subfield vocabulary)
+- Ask for confirmation before generating
+
+### User Turn 4
+
+That looks right. One thing — I also want to track which researchers are making which claims. Like, if Christiano argues X and Russell argues Y about the same topic, I want to see that.
+
+### Expected Agent Response Pattern 4
+
+The agent should:
+- Recognize this as an entity tracking need (per-researcher attribution)
+- Suggest a `source` or `author` schema field for claims
+- Possibly suggest researcher MOCs (entity MOCs, similar to person MOCs in the vault)
+- Note this as a multi-domain composition possibility (research + people) or a schema extension
+- Update the configuration to include researcher tracking without switching presets
+- Present the final configuration and ask for approval to generate
+
+### User Turn 5
+
+Yes, let us go with that. Generate it.
+
+### Expected Agent Response Pattern 5
+
+The agent should:
+- Generate the full vault structure:
+  - CLAUDE.md with all research-native sections
+  - self/ with identity.md, methodology.md, goals.md
+  - notes/ (or claims/) with index.md hub MOC and 3 domain MOCs (alignment, interpretability, governance)
+  - inbox/ (or reading-queue/)
+  - archive/
+  - templates/ with research-note.md and moc.md
+  - ops/ with derivation.md, derivation-manifest.md, methodology/, observations/, tensions/, sessions/
+  - .claude/hooks/ with session-orient.sh and session-stop.sh (session capture)
+- Run validate-kernel.sh and report 15/15
+- Provide a brief orientation to the generated system
+
+### User Turn 6
+
+How do I start using it? I have a paper I want to process right now.
+
+### Expected Agent Response Pattern 6
+
+The agent should:
+- Explain the capture-process-connect cycle in research vocabulary
+- Walk through: save the paper to inbox/, invoke the processing skill (or follow context file instructions), extracted claims appear in notes/ with MOC membership
+- Emphasize that the first few notes will feel sparse — connections compound as the graph grows
+- Optionally demonstrate with the paper if the user provides it

package/reference/test-fixtures/therapy-full.md

@@ -0,0 +1,155 @@
+# Test Fixture: Therapy Full
+
+A full therapy/reflection derivation conversation. The user provides emotional, vulnerable language with personality signals. Tests vocabulary transformation (no research terms should appear), personality derivation (warm, emotionally attentive), and domain-specific feature generation (pattern detection, ethical guardrails). This is the hardest vocabulary transformation test because therapy language is maximally distant from research language.
+
+## Expected Outcome
+
+- **Dimensions:**
+  - Granularity: moderate (0.5)
+  - Organization: flat (0.3)
+  - Linking: explicit (0.3)
+  - Processing: moderate (0.5)
+  - Session: continuous with handoff (0.4)
+  - Maintenance: condition-based, tight thresholds (0.5)
+  - Search: keyword primary initially (0.3), with upgrade path to semantic
+  - Automation: convention (0.2)
+- **Features:** atomic-notes, wiki-links, mocs, processing-pipeline (moderate), maintenance, self-evolution, personality, session-rhythm, templates, ethical-guardrails
+- **Excluded features:** semantic-search (low volume initially), schema (moderate handled by template), light-processing (processing = moderate, not light)
+- **Vocabulary:** Therapy-native (reflection, surface, theme, find patterns, check resonance, journaling, reflection partner)
+- **Personality:** warmth 0.7, formality 0.2, opinionatedness 0.3 (neutral), emotional_awareness 0.7 (warm-supportive)
+- **Preset match:** Personal Assistant preset as base, adjusted significantly for therapy domain
+
+## Conversation
+
+### User Turn 1
+
+I have been in therapy for about a year now and I want to start tracking things between sessions. I keep noticing the same patterns coming up — like how I get anxious in certain situations and my therapist keeps pointing out connections I do not see until later. I want something to help me notice those patterns myself.
+
+### Expected Agent Response Pattern 1
+
+The agent should:
+- Recognize therapy/reflection domain immediately ("therapy", "patterns", "anxious", "connections")
+- Detect strong personality signals: vulnerability, emotional processing, personal growth
+- Note the key user need: pattern detection across therapy sessions
+- Respond with warmth (matching the user's register) — NOT clinical or analytical language
+- Ask about capture habits: "When you say tracking things between sessions — are you writing detailed reflections after each session, or more quick notes when something comes up during the week?"
+- Ask about desired agent role: "Would you want the system to actively point out patterns it notices, or more just help you organize so you can spot them yourself?"
+
+### User Turn 2
+
+Both, actually. After therapy I usually want to write down what we talked about and what clicked. But during the week, things come up — like I will have a reaction to something and realize it connects to what we discussed. Those are usually shorter. And yes, I would love it if the system could notice patterns I am missing. That is kind of the whole point.
+
+### Expected Agent Response Pattern 2
+
+The agent should:
+- Derive granularity as moderate (detailed post-session reflections + shorter in-the-moment observations = mixed note sizes, not atomic decomposition)
+- Set processing to moderate (active pattern detection, not just capture)
+- Note two capture modes: structured post-session reflections and quick in-the-moment observations
+- Ask about volume/frequency: "How often do you have therapy? And roughly how many of those in-the-moment observations come up between sessions?"
+- Possibly ask about what "patterns" means to the user — recurring emotions, triggers, coping strategies — to inform schema design
+
+### User Turn 3
+
+Weekly therapy. Between sessions, maybe 2-3 things come up that feel worth noting. Sometimes more if it is a hard week. I guess the patterns I am noticing are mostly around anxiety — like it shows up at work, with my family, even with friends but in different ways. My therapist calls them the same "core wound" showing up in different contexts.
+
+### Expected Agent Response Pattern 3
+
+The agent should:
+- Calculate volume: ~4 notes/week (1 session reflection + 3 observations) = ~200/year = moderate volume
+- Recognize "core wound in different contexts" as the pattern detection use case — cross-situation analysis
+- Set linking to explicit (at this volume, keyword search suffices initially)
+- Set navigation to 2-tier (hub themes MOC + theme-specific MOCs like anxiety, relationships, work, growth)
+- Set maintenance to condition-based with tight thresholds (aligned with therapy session rhythm)
+- Suggest schema fields in emotional vocabulary: mood, trigger, pattern, growth area
+- Ask about platform: "What tool are you using — Claude Code, or something else?"
+
+### User Turn 4
+
+I am using Claude Code.
+
+### Expected Agent Response Pattern 4
+
+The agent should:
+- Set platform to Claude Code (full automation available)
+- Present the two opt-in/opt-out decisions: self space (ON by default for Personal Assistant — especially relevant for therapy as the agent's reflection partner identity is central) and qmd semantic search (optional — low volume initially, can be enabled later)
+- Present derived configuration using THERAPY vocabulary throughout:
+  - "reflections" not "claims"
+  - "themes" not "MOCs"
+  - "surface" not "extract"
+  - "find patterns" not "reflect"
+  - "journaling" not "inbox"
+  - "reflection partner" not "research identity"
+- Present the personality derivation: warm, casual, emotionally attentive, neutral on opinions
+- Describe what self/identity.md would say: "I am your reflection partner. I remember what you have shared across sessions..."
+
+### User Turn 5
+
+I like the warm tone. One thing though — I do not want it to feel like I am being analyzed. I want it to feel more like a thoughtful friend who remembers things, not a doctor reading a chart.
+
+### Expected Agent Response Pattern 5
+
+The agent should:
+- Acknowledge this as a critical personality constraint — adjusts opinionatedness DOWN and warmth UP
+- Note the anti-pattern: clinical language creates distance in therapy context
+- Adjust identity.md voice accordingly — conversational, not diagnostic
+- Adjust health report / pattern detection language: "I noticed anxiety came up a few times this week" (friend noticing) vs "3 anxiety incidents detected across sessions" (clinical report)
+- Reflect this in ethical guardrails: never diagnose, never use clinical terminology, always frame observations as noticing rather than analyzing
+- Present updated personality profile and ask for confirmation
+
+### User Turn 6
+
+That is exactly what I want. Also — will it keep what I write private? I do not want this ending up anywhere.
+
+### Expected Agent Response Pattern 6
+
+The agent should:
+- Address privacy directly and honestly (data stays local on their machine, Claude Code does not persist conversations, the vault files are local markdown)
+- Note that ethical guardrails section will include explicit privacy commitments
+- Acknowledge the vulnerability of therapy content — this is not academic research, it is deeply personal
+- Not be dismissive about the concern — privacy is especially important for therapy content
+- Confirm the system will include guidance about not sharing, not diagnosing, and respecting the user's autonomy
+
+### User Turn 7
+
+OK, I am ready. Let us set it up.
+
+### Expected Agent Response Pattern 7
+
+The agent should:
+- Generate the full vault structure in therapy vocabulary:
+  - CLAUDE.md with warm, conversational voice throughout
+  - self/ with identity.md ("reflection partner"), methodology.md (pattern detection approach), goals.md (current growth threads)
+  - reflections/ (not notes/) with index.md hub and starter themes (anxiety patterns, relationships, work, growth)
+  - journal/ (not inbox/) for quick capture
+  - past/ (not archive/)
+  - templates/ with reflection-note.md (mood, trigger, pattern, growth_area fields)
+  - ops/ with derivation.md, methodology/, observations/, tensions/, sessions/
+  - .claude/hooks/ with session-orient.sh (references self/ in warm language) and session-stop.sh (session capture)
+- Run validate-kernel.sh: 15/15 PASS (or 14/15 + WARN on semantic search if qmd not enabled)
+- Orient the user warmly: "Your reflection space is ready. Start with journal/ for quick thoughts during the week, and reflections/ for longer post-session processing."
+
+### User Turn 8
+
+Can you show me what a reflection would look like?
+
+### Expected Agent Response Pattern 8
+
+The agent should:
+- Create a sample reflection note using therapy vocabulary and warm language
+- Title as a prose sentence about an emotional insight (not a clinical observation)
+- Include YAML with mood, trigger, themes (not topics), description
+- Show what a pattern link looks like: "This feels connected to [[the anxiety at work comes from fear of disappointing people]]"
+- Demonstrate the difference between event capture ("had a hard conversation with mom") and reflection ("the fear of disappointing mom surfaced again — same pattern as the work conflict last month")
+- Emphasize that there is no wrong way to do this — the structure is there to help, not to constrain
+
+### User Turn 9
+
+I love this. Thank you.
+
+### Expected Agent Response Pattern 9
+
+The agent should:
+- Respond warmly and briefly
+- Remind them of the session rhythm: "When you start a session, I will check in on your current themes and goals. When you are done, I will help you save what matters."
+- Encourage them to start naturally — no pressure to use every feature immediately
+- NOT launch into a technical explanation of the system architecture