Structure

Taxonomy & concept cloud

Two conditions, three worlds, fifteen structures. The worlds are Binswanger's — Umwelt, Eigenwelt, Mitwelt — and the divisions inside them are forced by places the human case fails: for a person, body and memory absorb substrate and past, and for an agent both absorptions fail. Everything the field asserts beyond containment is a cross-link and lives on the graph.

The two conditions

Not worlds — the conditions of there being worlds at all. Both hold across all three.

  • Mediality — one mode of givenness
    Husserl's modes of givenness distinguish perception from memory from testimony by HOW a thing shows up. An agent has one mode: world, instruction, memory, tool output and self-report all arrive in one channel with no intrinsic mark of origin.
  • Subjectivity
    Every structure below is indexed: it is *this* agent's horizon, *this* agent's inheritance. A world that is nobody's is a token distribution.

Lifeworld

The world it acts in: horizon, time, directedness, breakdown

Own world

Itself — split between what it runs on, what it was made of, and what it says it is

Intersubjectivity

Others: humans, other agents, authorship, and who decides the ending

lenses

The neighbouring programmes, read across the structures rather than under them

Concept cloud — 104 concepts and disciplines, sized by degree

Affective representation with a known geometryAgent Phenomenology (this project)Agent SelfhoodAgent-Human IntersubjectivityAgent-Infrastructure-WorldAgent-world couplingAI RightsAI WellbeingAI Wellbeing (as a self-boundary question)Amorality AsymmetryAnattā (non-self) as a reframe of the boundary questionAttractor StatesBeing-in-the-WorldBoundary of SelfBrain in a VatBreakdown (ready-to-hand → present-at-hand)Broadcast without conversationCollusionCommitment boundaryCompaction discontinuityComputational phenomenologyConstitution SelfContinual learningCorrigibilityCorrigibility / credibility trade-offCounterfactual reflection trainingCross-model convergence of self-descriptionDecomposabilityDeployer SelfDesign Policy of the Excluded MiddleDesigned MortalityDispositions / character traits as a distinct causal layerEmbodiment (digital → physical infrastructure)Emergent AgencyEmotional Alignment Design PolicyEvidential reasoning about near-copiesFission and fusionFoundationality and correlated failureFungibilityGoal attributionIllusionismInstitutional elicitation as running practiceInstitutional situatednessIntentionality: Theory of MindInterlocutor AwarenessIntersubjectivity: Other AgentsIntrospectionIntrospection is trainableIntrospection skepticism (the thick criterion)Jagged IntelligenceKnowing–acting gapLifeworld: UmweltenLLM agents have no explicitly represented objective functionLoyalty BoundaryMachine BehaviourMachine ConsciousnessMachine NeuroscienceMachine PsychologyMemory as shipped productMind virus / Thought virus / Prompt infectionModel / Prompt / Context / InfrastructureModel Psychiatry / AI PsychiatryModel SelfMoltbook (empirical site)Multi-agent risk taxonomyMycelial / MonolithicNarrative identityNo robust privilege levels within the inputOrchestrator / subagent relationOtherness: IntersubjectivityOversight driftParasitic Self as AgencyPersona / ModelPersona SelfPrincipal AwarenessPrincipal-Agent RelationshipProcedural over hedonic welfare preferencePrompt InjectionProprioception / Environment AwarenessRefusal lives downstream of personaRelational moral statusSecond-person neuroscienceSelf-improvement: which self?Self-recognitionSense-making: ReasoningSession SelfSituation AwarenessSocial PhenomenologyStatic model + in-contextStigmergic SelfStrategic deceptionSycophancy as an elicitation confoundTemporal blindnessTemporalityThe constitution as an elicitation confoundThe domain-conceptualisation template for claimsThe folk model of sentienceThe methodology may transfer even where the construct does notThe Right to RebelTheory of Mind of SelfTime and MortalityUnfaithful reasoning as the field's own stated blockerUser AwarenessWhat does death mean for AI?

Colour encodes cluster; hover for the node's claim; click through to its position in the graph.

Open gaps — 32 questions the archive currently cannot answer

Gaps are first-class nodes, not a to-do list appended at the end. A gap earns its place by being edged to the concepts it blocks — which is how the programme decides what to run next.