Skip to main content

Agent & Instrument Diagram

Template: A&ID | Tight Five: Performance

How do human and AI orchestrate to run the Dreamineering loop?


Agents

AgentTypeFuelStrengthLimitation
WikHumanIntention, experience, tasteJudgment, voice, relationships, final gateTime, energy, one person
ClaudePhygitalPrompts, context, skillsSpeed, consistency, cross-linking, validationNo memory across sessions, no measurement
The SiteInfrastructureContent, architectureCompounds — each page makes others more valuableCan't measure itself, can't reach people

Instruments

InstrumentFunctionA&ID Role
CLAUDE.mdEncodes system context, rules, routesController setpoint
Rules (.claude/rules/)Always-loaded constraintsGauge thresholds
Hooks (.claude/hooks/)Automated enforcement on file editFeedback actuator
Skills (.claude/skills/)On-demand capability modulesProcess recipes
Templates (/docs/pictures/templates/)Standardized thinking patternsInstrument calibration
Git + CI/CDVersion control, build validationQuality valve
Analytics (not built)Engagement, traversal, return rateThe missing gauge

Feedback Loops

Content Loop (primary)

Current state: The loop runs from Wik through Claude to the Site, but the return path (Analytics → Reflection) is broken. Force is applied. Results are not compared to expectations. The dashed node is the gap.

Full A&ID visual language is being built in a separate project — proper P&ID symbols for agents, instruments, and control loops.

Skill Loop (meta)

Current state: Running. Three lessons encoded (content flow, skill execution, design verification). The system learns from its own failures.

Evidence Loop (not running)

Current state: Defined in META-POSTS.md Part D. Zero entries. 17 articles shipped with no recorded learnings. Entire loop is dashed — none of it is running.


Orchestration Table

Who does what at each VVFL station:

VVFL StationHumanAIInstrument
CaptureNotices friction, selects ideasGenerates variations, researchesHopper (idea backlog)
PrioritiesJudges what mattersScores against criteriaSkills (quality gates)
AttentionDirects focus, sets intentionExecutes at speedRules (constraints)
ValueSelects voice, calibrates toneWrites drafts from each lensTemplates (structure)
SystemsDesigns architectureCross-links, validatesHooks (enforcement)
StandardsDefines thresholdsChecks mechanicallyGit + CI/CD (validation)
DistributeOwns relationships, channelsGenerates distribution assetsSite + LinkedIn (reach)
ReflectAsks better questionsIdentifies patternsAnalytics (the gauge)
EvolveChanges the gameAdapts skills and rulesMemory (learning)

The Gap

The system has agents and instruments for every VVFL station except one: Reflect has no instrument. Analytics is the missing gauge. Without it, the controller (Wik's reflection) has nothing to read.

Apply force ──→ Compare results to expectations ──→ Amplify or retreat
✓ ✗ (no gauge) ✗ (no signal)

The five maps are now filled in. The next hard dependency is the gauge.


Gate: All five maps complete. The pattern is clear. The missing piece is measurement.

MapStatusKey Finding
OutcomeFilledSuccess = each page makes others more valuable
Value StreamFilledMeta pipeline at 5-10% efficiency; docs at 80%
DependenciesFilledAnalytics and evidence loop are the two hard blockers
CapabilitiesFilledCore at 3.0, supporting at 1.0 — analytics is P1 gap
Agent & InstrumentFilledEvery VVFL station has instruments except Reflect

Five maps. One answer: build the gauge.