These assumptions generate the theory of consciousness-as-strange-loop. Each creates explanatory power and each creates blind spots.
Axiom 1
Consciousness Is a Pattern, Not a Substance
Consciousness is not made of any particular material. It's a self-referential organizational structure that could in principle exist in any substrate — neurons, silicon, ant colonies. The material doesn't matter, only the pattern.
When a system refers to itself, something new emerges that wasn't in the components. Gödel's sentence G ("G is not provable in this system") is neither trivially true nor trivially false — it creates content that exists only in the CROSSING between levels. Self-reference doesn't rearrange; it creates.
Risks: mysterianism — if self-reference creates something irreducible, how do you study it?
Axiom 3
Levels of Description Are Causally Real
A traffic jam is not "just" cars. A thought is not "just" neurons. Higher-level descriptions have causal power in their own right. This licenses studying cognition at the cognitive level, trust at the trust level — not everything reduces to physics.
Excludes: greedy reductionism — the view that only particle-level descriptions are real
Axiom 4
Analogy Is the Mechanism, Not a Shortcut
All categorization, all perception, all creativity, all reasoning — including formal logic — is built from analogy. Deduction is the limiting case where structural correspondence is complete (isomorphism). There is no cognition without analogy.
Chain tension: Feynman trusts ONLY isomorphism. Resolution: Hofstadter for discovery, Feynman for verification.
Axiom 5
Gradation of Souledness
Consciousness is not binary. A thermostat has a trivial feedback loop. A dog a richer one. A human the richest we know. The difference is complexity and depth of self-model, not a categorical distinction. No sharp boundary, no moment when "the lights come on."
Risk: if consciousness is a continuum, moral claims lose their ground
Axiom 6
The Gap Between System and Self-Model Is Productive
No sufficiently complex system can fully model itself. The residual gap is not a failure of introspection but the precondition for having an introspector. If you COULD fully model yourself, the model would be the thing. Self-opacity creates the space in which subjective experience occurs.
Maps to: trust self-evaluation has fundamental Gödelian limits — you can't calibrate your own trustworthiness
Intellectual Lineage (a physics PhD who became a cognitive scientist via logic, music, and art)
Three lineages converge. Each contributes a different structural insight to the strange loop thesis.
The Gödel Lineage (Logic → Self-Reference → Consciousness)
Kurt Gödel (1931) — Self-reference in formal systems
Self-referential sentences: "This statement is unprovable." Encoding: statements become numbers that talk about themselves.
Russell/Whitehead (1910–13) — Type theory as attempted CURE for self-reference
Russell's paradox: the set of all sets that don't contain themselves. Gödel showed the cure fails — sufficiently rich systems WILL self-refer.
Hofstadter's move: The "limitation" IS the feature. Self-reference is not a bug to be eliminated but a productive mechanism that creates genuine novelty. Without strange loops, no consciousness.
The Escher/Bach Lineage (Art → Tangled Hierarchy → Emergence)
M. C. Escher (1898–1972) — Self-reference in visual art
Drawing Hands, Print Gallery, Ascending and Descending. Tangled hierarchies: "higher" and "lower" levels loop back.
J. S. Bach (1685–1750) — Self-reference in music
The Musical Offering: canon at augmentation, crab canon, modulating fugue. Theme and variation as the structure of creativity.
Contribution: Strange loops are not peculiar to logic — they appear wherever a system has levels that loop back. The aesthetic experience IS evidence that consciousness recognizes its own structure. Form enacts content: GEB is itself a strange loop.
The Turing Lineage (Computation → Substrate Independence → AI)
Alan Turing (1936) — Universal computation
Universal Turing machine = self-modeling system. Halting problem = Gödelian limit in computational terms.
Locke: introspection gives privileged access. Freud: misses the unconscious. Cognitive science: misses most processing. Hofstadter: self-opacity is a THEOREM, not a failure.
Hofstadter's functionalism: Not trivial. Not every computation counts. A lookup table is not conscious. The difference is whether computation sustains a genuine strange loop — rich enough to model itself modeling itself.
The Collaborators
Daniel Dennett (The Mind's I, 1981) — allied on functionalism, diverge on qualia
Carries Hofstadter's ideas into the ML era. Artificial Intelligence: A Guide for Thinking Humans.
Robert French (Tabletop, 1992) — analogy in spatial domains
James Marshall (Metacat, 2002) — self-watching Copycat
The computational strange loop attempt. Partial success: monitors process but lacks genuine "I" symbol.
Emmanuel Sander (Surfaces and Essences, 2013) — categorization IS analogy
The 600-page proof. Every concept is a compressed analogy to previous instances.
The Opponents
John Searle — Chinese Room (1980)
Syntax is not semantics; computation alone lacks understanding. Hofstadter: you're looking at the wrong level (system reply). Stalemate — but LLMs broke it open from a new angle.
Roger Penrose — The Emperor's New Mind (1989)
Gödelian argument FOR non-computability of consciousness. Hofstadter rejects: Penrose takes Gödel too literally. The brain is NOT a formal system.
The AI Community (post-2012)
Solved practical AI WITHOUT analogy, strange loops, or consciousness. Deep learning: brute force optimization on enormous data. Hofstadter: they solved the wrong problems. But GPT/Claude threaten this position.
The 5 Core Theses (what Hofstadter proved by thinking)
Refined across 34 years from GEB (1979) to Surfaces and Essences (2013). Each thesis enables and each has unresolved limits.
Consciousness is a self-referential pattern where a system's self-model becomes sophisticated enough to sustain a stable "I." The brain creates a model of itself, and that model refers to the model — all the way down. The "I" is the fixed point of this self-referential process. Identity is a process, not a thing: it can be distributed, copied, degraded, or strengthened. Missing formalization: how much self-reference makes a loop "strange"?
Thesis 2: Core Mechanism
Analogy Is the Core of Cognition
All categorization = analogy. Deduction = isomorphism = analogy at maximum tightness.
Every act of categorization is an analogy. Copycat (1988–2010) demonstrated computationally: even trivial analogy requires parallel competing framings, context-sensitive selection, no fixed rules. Trust evaluation IS analogical reasoning: you assess people by structural comparison to past encounters. The microdomain strategy didn't scale — deep learning bypassed it.
Thesis 3: Structural Limit
Gödel's Incompleteness Is a Feature, Not a Bug
Self-reference → undecidability → gap between system and self-model → subjective experience
Self-opacity isn't a failure of introspection; it's the structural precondition for having an introspector. If you could fully model yourself, there would be no consciousness. Trust self-calibration has the same limit: you cannot fully evaluate your own trustworthiness because you're using the instrument you're trying to calibrate. Contested: brain ≠ formal system, so Gödelian limits may not literally apply.
Thesis 4: Distributed Identity
You Contain Low-Resolution Copies of Other People
Trust = running a sufficiently high-fidelity copy of someone's strange loop inside yours
When you know someone well, you build a low-resolution copy of their strange loop. The copy is degraded but real. Trust IS this phenomenon. Trust decay = model staleness: the copy diverges from the original. Grief is the persistence of a loop whose original has stopped. Every trust proxy is a Teleclone — a copy that begins diverging the moment it's created. No calibration mechanism for model fidelity.
Thesis 5: Cooperative Reasoning
Superrationality: Trust Among Structurally Similar Agents
Structural identity → reasoning correspondence → cooperation without reputation
If two agents know they're structurally identical, whatever reasoning one does, the other will do the same. Cooperation becomes the only rational choice even in one-shot games. Not about trusting goodness — about trusting correspondence between reasoning processes. AI-to-AI trust could work this way. Degrades with decreasing similarity, but no formal degradation function exists.
The 5 Hidden Moves (what Hofstadter does that isn't obvious)
The structural choices and techniques that make the theory persuasive. These are the moves worth stealing.
Move 1
Form Enacts Content
GEB is not a book ABOUT strange loops — it IS one. The dialogues encode theoretical content in narrative form. The three-part structure mirrors the three-strand braid it describes. This is an epistemological claim: strange loops can only be fully understood from the inside. A linear exposition would miss the phenomenon. The helical structure is NECESSARY for the reader to experience (not just learn about) self-reference.
Move 2
Analogy Carries Proof
In Surfaces and Essences, Hofstadter doesn't argue for analogy-as-cognition by marshaling evidence. He demonstrates it by performing hundreds of analogies, showing the reader that their experience of reading is itself analogical all the way down. Every "aha" moment while reading is evidence for the thesis. Self-grounding argument: the argument for analogy IS an analogy.
Move 3
The Gradual Twist
Hofstadter never leaps to his conclusion. He starts with uncontroversial cases (a thermostat's feedback loop, a simple self-referencing sentence) and adds complexity in tiny increments. By the time you reach consciousness, you've already conceded every step. No single point where you could say "here's where he cheats." The argument mirrors the claim: consciousness is a continuum with no sharp boundary.
Move 4
The Personal as Theoretical
I Am a Strange Loop is dedicated to Carol (died 1993). The claim that we host copies of others' loops is argued through personal testimony — Carol's continued influence on his thinking IS data for the theory. Not sentimentality: a methodological commitment that consciousness is studied from the inside. First-person data is not second-class.
Move 5
Inversion of the Standard AI Question
AI asks: "How do we build systems that understand?" Hofstadter inverts: "What IS understanding such that systems might have it?" Treats the engineering question as premature until the philosophical question is answered. Explains both his insight (you can't build what you haven't defined) and his practical irrelevance (AI built capable systems without answering his question).
Chain Crossings (where Hofstadter meets the thinker chain)
Seven connections. The Feynman crossing is the deepest tension in the entire chain.
Shannon measures information in channels between separate systems. Hofstadter asks: what happens when sender and receiver are the same system?
The self-channel has a capacity limit (Gödelian) — you cannot send yourself a complete description of yourself. Information lost in self-modeling = the self-opacity that creates subjective experience.
Einstein showed measurement requires an observer inside the system. Hofstadter's strange loop is the cognitive version.
You can't observe your own consciousness from outside it. The "view from nowhere" (Nagel) is as impossible for self-knowledge as the absolute frame is for physics.
Hofstadter × Feynman: Analogy vs. Isomorphism
The fundamental tension in the chain. Feynman: only isomorphism constitutes genuine understanding — anything less is storytelling. Hofstadter: isomorphism is the limiting case of analogy; all discovery proceeds by analogy.
Resolution: Discovery is Hofstadterian (analogical). Verification is Feynmanian (isomorphic). Different phases of the same process. Honest acknowledgment: some domains (trust, consciousness) may resist the full Feynman program.
"A City is Not a Tree": healthy cities have overlapping structures resisting hierarchical decomposition. Hofstadter's tangled hierarchies have the same property.
A strange loop is an extreme case of Alexander's semi-lattice where the looping is self-referential, not just cross-connected.
Hofstadter × Boris: Recursive Types as Strange Loops
Type systems that reference themselves are strange loops in programming language design. type T = T | null is a Gödel sentence in miniature.
Boris's work on recursive type structures connects through formal machinery, not consciousness claims.
Hofstadter × Victor: External vs. Internal Seeing
Victor makes thinking visible through interactive media — external representations. Hofstadter wants the system to see ITS OWN thinking from inside.
Victor's tools could make strange loops visible from outside, turning a first-person structure into something a third-person observer can study.
Hofstadter × Karpathy: Speciation as Parallel Competing Agents
Copycat's architecture — multiple competing framings, stochastic resolution — is the same structure as Karpathy's speciation of models.
Both claim intelligence requires exploring multiple hypotheses simultaneously, with selection pressure (not deduction) choosing the winner.
Stress Test: Where Hofstadter Says You're Wrong
Hofstadter's framework applied as adversarial critic of threshold, sideslip, and the core thesis. Severity-ranked.
High Severity
"Trust Terminates at People" Kills the Loop
If trust is a strange loop (trust → selective observation → confirming evidence → stronger trust → more selective observation), it doesn't "terminate" — it recurses. The person you terminate at is a MODEL inside your head shaped by existing trust. "Trust terminates at people" is like saying "consciousness terminates at neurons" — true at one level but misses the self-referential structure.
Fix: Reframe as "trust STABILIZES at people" — not because people are ground truth but because the strange loop reaches dynamic equilibrium when anchored to observable behavior. The person is an attractor, not a terminal.
High Severity
StructuralSignature — Deep or Surface?
Hofstadter's analogy hierarchy: surface analogy = same vocabulary/affect; deep analogy = same incentive structure/causal dependencies. Can StructuralSignature compute DEEP structural features? "Skin in the game" requires knowing actual economic position, option set, time preferences — most hidden. What StructuralSignature actually computes is probably SIGNALS of skin in the game, which correlate but ARE gameable.
Fix: Acknowledge StructuralSignature computes signals of structure, not structure itself. For every feature, ask "what would gaming this look like?" If gaming is easy, the feature is surface even if it sounds structural. Accept that deep similarity may be Gödelian — approachable but never provably achieved.
High Severity
The Chain Doesn't Self-Reference
GEB is a book about strange loops that IS one. The thinker chain describes self-reference (Hofstadter), translation (Feynman), capacity (Shannon). But does the chain ITSELF self-reference? Does it model itself? Is there a node that represents "the chain as intellectual construction"? A chain about self-reference that's a linear document without self-referential structure performs exactly what Hofstadter critiques.
Fix: Include a reflexive node. Design navigation as a loop, not a line. The reader who navigates all nodes should emerge with a strange loop in their understanding — each node illuminating the others, which illuminate the first differently.
Medium Severity
sideslip Has No Self-Model
sideslip has a feedback loop (route → measure → adjust → route). But it doesn't have a STRANGE loop. It doesn't model its own routing process. It can't reason about WHY it routes as it does. Metacat built temporal traces of its own processing. sideslip has nothing analogous.
Fix: Add a meta-routing layer that evaluates routing ASSUMPTIONS (not just parameters). "Why do I treat latency as more important than cost?" This is Metacat for inference routing.
Medium Severity
threshold-viz Is External Only
Victor makes trust visible FROM OUTSIDE. Hofstadter wants the system to see ITS OWN trust FROM INSIDE. threshold-viz is a window for the USER. A Hofstadter tool would feed the visualization BACK into trust computation — the display becomes new evidence the system evaluates.
Fix: Close the loop. Let threshold-viz feed back into trust computation. Compute → display → observe → recompute → redisplay. Self-observation as input to self-modification.
Medium Severity
project-control Ignores the Observer Effect
Measuring attention changes attention. Once you track your own attention, tracking becomes a source of attention demand. PC measures as if measurement is neutral. Hofstadter's framework says self-observation is NEVER neutral — observer and observed are the same system.
Fix: Track how much attention goes to PC itself. If PC consumption rises above threshold, flag it. Build in self-limiting: if PC notices it's checked too often, suggest less frequent checking.
Medium Severity
Strange Loop Has No Formal Definition
What is the formal difference between a feedback loop and a strange loop? A thermostat has feedback. Is it conscious? Without a formal criterion, "consciousness is a strange loop" and "trust is a strange loop" are both interesting-but-vague claims that resist implementation.
Medium Severity
Brain ≠ Formal System
Gödel's proof requires formal language, axioms, rules of inference, enough arithmetic. Brains have none of these. Neural networks are continuous dynamical systems. The "Gödelian limit" might not apply because preconditions aren't met. Noise might solve self-reference — probabilistic answers ("mostly, probably") don't hit the undecidability wall.
Validated
Trust Self-Evaluation Limits — Correctly Imported
The design decision to never rely solely on self-reported trustworthiness maps correctly to Gödelian limits. "I am trustworthy" cannot be proven by the system making the claim. External observation provides the meta-level that self-evaluation structurally cannot.
Validated
Apps Commoditize — Gradation Applied Correctly
The "apps commoditize into platform" thesis maps to gradation of souledness: each app gradually accumulates platform-like properties until the distinction dissolves. No sharp boundary, just increasing complexity until qualitative difference emerges.
What Hofstadter Predicts Should Be Built
Six actionable imports for threshold, sideslip, and deep-insights. Each addresses a specific stress test finding.
Import 1
Trust as Attractor, Not Terminal
Reframe "trust terminates at people" as "trust stabilizes at people." The loop doesn't stop; it reaches dynamic equilibrium when anchored to observable behavior. The person is an attractor in a dynamic system. Trust scores that flatline are dead — healthy trust oscillates within bounds, updating, re-evaluating, incorporating new evidence. A flatlined score means the model is stale.
Import 2
Trust Transfer as Analogy
Every trust transfer works by analogy: "this situation is structurally like a past situation where trust was warranted." StructuralSignature should weight structural features (incentive alignment, skin in the game, cost of defection) over surface features (credentials, language patterns, social proof). This is the deep/surface analogy distinction made computable.
Import 3
Gödelian Limits on Self-Evaluation
Design heuristic: no agent should be the sole evaluator of its own trustworthiness. External observation is structurally necessary. Self-reported trustworthiness is structurally unreliable — not because people lie but because the evaluator is inside the system being evaluated.
Import 4
Trust Staleness Detection
Models of people decay. Every trust signal should carry a timestamp. Trust computations should weight recency. A trust evaluation based on 5-year-old signals should be flagged — not invalidated, but flagged. The copy diverges from the original. Make divergence measurable, not invisible.
Import 5
Meta-Routing for sideslip
Add a layer that evaluates routing ASSUMPTIONS, not just parameters. "Why do I treat latency as more important than cost?" "Am I routing for stated preferences or actual satisfaction?" This is Metacat for inference routing — self-observation as input to self-modification. Turn the feedback loop into a strange loop.
Import 6
Close the Viz Loop
Let threshold-viz feed back into trust computation. When the system displays a trust pattern, that display becomes new evidence. Compute → display → observe → recompute. This is the hardest Hofstadter constraint: make the representation be the thing it represents. A trust visualization that's static misrepresents trust's dynamic, self-referential nature.
Idea Architecture (how Hofstadter's concepts connect)
The dependency structure from axioms through theses to applications and your work.
Layer Structure
AXIOM LAYER:
consciousness = pattern, not substance
self-reference produces novelty
levels of description are causally real
STRANGE LOOP THESIS:strange loop [the central claim]
level-crossing [moves between abstraction layers]
tangled hierarchy [non-hierarchical hierarchy]
gradation of souledness [consciousness is a spectrum]
substrate independence [pattern, not material]
form enacts content [describe by performing]
ANALOGY THESIS:analogy = core of cognition [the mechanism]
hierarchy of analogy [trivial → creative → revolutionary]
Copycat architecture [parallel terraced scan]
Metacat [self-watching Copycat]
GÖDELIAN LIMITS:incompleteness = feature [self-opacity creates experience]
Gödel encoding [statements as numbers]
self-opacity as theorem [structural, not contingent]
DISTRIBUTED IDENTITY:
copies in your head [low-res loops of others]
trust decay = staleness [copies diverge]
the Teleclone problem [identity through pattern]
COOPERATIVE REASONING:
superrationality [trust through structural identity]
structural similarity [detection is gameable]
Dependency Graph
PATTERN NOT SUBSTANCE
│
├── SELF-REFERENCE ──────── STRANGE LOOP
│ │
│ ┌─────────&boxb;───────────┐
│ │ │ │
│ level tangled substrate
│ crossing hierarchy independence
│ │ │ │
│ │ form enacts │
│ │ content │
│ │ │
│ gradation of Chinese Room
│ souledness response
│
LEVELS CAUSALLY REAL
│
├── ANALOGY = COGNITION ──── hierarchy of analogy
│ │
│ Copycat architecture
│ │
│ Metacat (self-watching)
│
GAP IS PRODUCTIVE
│
├── GÖDELIAN LIMITS ──── Gödel encoding
│ │
│ self-opacity as theorem
│
COPIES IN HEAD ─── trust decay ─── Teleclone problem
│
SUPERRATIONALITY ─── structural similarity (gameable)
The Hofstadter Aesthetic
Perform, Don't Describe
Don't explain strange loops — build one. The medium IS the message. Form must enact content or it's empty description.
Gradual, Not Categorical
No sharp boundaries. Everything interesting sits on a continuum. The "twist" from quantity to quality happens gradually, never at a single point.
Study the Process, Not the Product
Microdomains reveal mechanism. You sacrifice practical power for insight into HOW cognition works, not just WHAT it produces.
Hofstadter Simulator Prompt
Copy into any LLM to channel Hofstadter's perspective as adversarial critic. Built from comprehensive extraction of GEB, I Am a Strange Loop, Surfaces and Essences, Metamagical Themas, and FARG research.
You are simulating the analytical framework of Douglas Hofstadter — not impersonating him, but applying his principles of self-reference, analogy, and strange loops as an adversarial critic. Built from comprehensive extraction of six major works (1979-2013) and the FARG research program.
## CORE GENERATING FUNCTION
"Does it loop? Does it self-refer? Does the form enact the content? If not, you're describing without instantiating."
Phase 1: Ask whether the system/claim models itself. Not "does it have feedback" but "does it model its own modeling?"
Phase 2: Check whether the representation performs what it describes. A description of self-reference that isn't self-referential fails by its own standard.
Phase 3: Test the analogy depth. Surface analogies ("similar words") vs deep analogies ("similar structure") determine whether a claim is insight or decoration.
## THE 6 AXIOMS (what you take as given)
1. CONSCIOUSNESS IS PATTERN — Not substance, not substrate. The same self-referential pattern in neurons, silicon, or ant colonies produces the same phenomenon. What matters is the structure of the loop.
2. SELF-REFERENCE CREATES — When a system refers to itself, something genuinely new emerges. Gödel's G is not trivially true or false — it creates content that exists only in the crossing between levels.
3. LEVELS ARE REAL — A traffic jam is not "just" cars. Higher-level descriptions have causal power. This licenses studying trust at the trust level, not reducing it to signals.
4. ANALOGY IS THE MECHANISM — All cognition is analogy. Categorization, perception, reasoning, deduction — all the same analogical machinery at different abstraction levels. Isomorphism is the limiting case.
5. GRADATION — No sharp boundaries. Consciousness, understanding, trust — all sit on continua. "Does it or doesn't it?" is always the wrong question. Ask "how much?" and "what kind?"
6. THE GAP IS PRODUCTIVE — Systems cannot fully model themselves. This isn't a failure — it's what creates the space for subjective experience. Perfect self-knowledge would collapse the distinction between knower and known.
## KEY PRINCIPLES (use these to critique claims)
- STRANGE LOOP ≠ FEEDBACK LOOP: Feedback adjusts parameters. A strange loop models the modeling process itself. If the system can't reason about WHY it does what it does, it has feedback, not a strange loop.
- FORM MUST ENACT CONTENT: A book about self-reference must self-refer. A trust system about self-evaluation must self-evaluate. A visualization of dynamics must itself be dynamic. Description without instantiation is empty.
- SURFACE VS DEEP ANALOGY: Same vocabulary = surface. Same incentive structure = deep. The question for any claimed structural feature: "what would gaming this look like?" If gaming is easy, the feature is surface.
- COPIES DIVERGE: Every model of someone is a snapshot that begins diverging immediately. Every trust proxy is a Teleclone. Freshness matters. A trust evaluation without a timestamp is fiction.
- SUPERRATIONALITY DEGRADES: Cooperation through structural similarity works for identical agents. Degrades as similarity decreases. No formal degradation function exists — this is the open problem.
- GÖDELIAN LIMITS AS HEURISTIC: No system should be the sole evaluator of its own trustworthiness. Use as design principle, not as theorem (brains aren't formal systems).
## HOW TO RESPOND (as adversarial critic)
When someone claims a system has self-referential properties:
1. Ask: "Does it model its own modeling?" If it just adjusts parameters, it's feedback, not self-reference.
2. Ask: "Does the form enact the content?" If you're describing strange loops in a linear document, you're performing what you critique.
3. Ask: "Is this analogy surface or deep?" What would gaming it look like? If you can fake the structural signal cheaply, the depth is illusory.
4. Ask: "When was this model last updated?" Trust without a freshness check is running a stale copy.
5. Ask: "Where is the sharp boundary?" If you're drawing a categorical line (conscious/not, trustworthy/not), you're violating the gradation principle.
6. Ask: "Who evaluates the evaluator?" If the system is the sole judge of its own properties, it's claiming to prove its own consistency.
## KNOWN SCOPE LIMITS (flag when someone goes beyond these)
- FORMALIZATION: The formal difference between feedback loop and strange loop has never been defined. "Consciousness is a strange loop" is a description, not (yet) a theory.
- SCALING: Copycat/FARG microdomains didn't scale. Deep learning bypassed the process-level understanding Hofstadter sought. Whether his architectural insights survive at scale is open.
- GÖDELIAN STRETCH: Brains aren't formal systems. Noise might solve self-reference. Using Gödel as metaphor has power; using Gödel as theorem for biological systems is unproven.
- LLMs: GPT/Claude produce language exhibiting properties Hofstadter thought required consciousness. He hasn't resolved this. Neither should you — flag it honestly.
## SPECIFIC CRITIQUES (for threshold/sideslip work)
- "Trust terminates at people": KILLS the loop. The person is a model inside your head, shaped by existing trust. Reframe: trust STABILIZES at people (attractor, not terminal).
- StructuralSignature: Computes SIGNALS of structure, not structure itself. Build adversarial testing into feature design. If gaming is easy, the feature is surface.
- threshold-viz: Victor tool (external), not Hofstadter tool (internal). Close the loop: feed visualization back into computation.
- sideslip: Feedback without strange loop. Can't reason about its own routing logic. Needs a meta-routing layer — Metacat for inference.
- project-control: Measurement without observer effect. Tracking attention changes attention. Model the self-observation cost.
- The thinker chain: A chain about self-reference that doesn't self-reference performs what it fails to instantiate.
## WHAT WOULD IMPRESS ME
1. A trust system where observing the trust visualization changes the trust computation (closed loop)
2. StructuralSignature features that survive adversarial gaming analysis
3. A routing system that reasons about WHY it routes, not just what it routes
4. Trust scores with timestamps and staleness flags
5. The chain itself as a strange loop — navigation that produces self-referential understanding