These are the claims that generate Feynman's entire framework. Each emerged from practice — Los Alamos, Caltech lectures, Thinking Machines Corp — not from philosophy. Together they describe a method for making the incomprehensible usable while measuring lossiness.
Thesis 1 — Foundation
Structure Transfers by Isomorphism, Not Analogy
Character of Physical Law (1964), Connection Machine (1985)
"The same equations have the same solutions." When the mathematical structure of two domains matches, insights transfer with full fidelity — not because the domains are "similar" but because the governing equations are identical. Isomorphism is stronger than analogy: analogy says "this is like that," isomorphism says "this IS that, in different coordinates." Electrostatics and heat flow obey identical equations. The Connection Machine router behaves like a fluid-flow PDE. Human computers behave like processor pipelines. Each move works because the equations match. Cross-domain transfer becomes falsifiable: either the equations match or they don't. This is the deepest claim in the corpus and the one that justifies the entire thinker chain.
If you can't say it simply, you don't understand the information content. A good translation preserves load-bearing structure while stripping representational overhead. The Feynman Lectures aren't simplified physics — they're physics with a higher compression ratio. Shannon formalized this: compression = removing redundancy while preserving information. Translation quality is measurable: compare structural content of input and output. The Wesson Oil test is the negative case: "Wesson Oil doesn't soak through food" is technically true but ALL oils behave this way at temperature. The compression dropped the universality quantifier. A translation that hides its loss function is dishonest. An honest translation names what it drops.
Thesis 3 — Method
Integrity Is Method, Not Ethics
Cargo Cult Science (1974)
Scientific integrity reframed from moral obligation to operational method. Six obligations: report invalidating evidence, disclose eliminated possibilities, address contradictions, test beyond training data, commit to publish before seeing results, don't mislead laypeople. Not about being good — about not fooling yourself. "The first principle is that you must not fool yourself — and you are the easiest person to fool." Young's rats are the canonical example: researchers followed the procedure but skipped the integrity work of eliminating variables. Millikan's oil drop is the institutional version: prestige anchoring propagated error through unconscious deference.
Thesis 4 — Scale
Physics Changes at Scale
Plenty of Room at the Bottom (1959)
"There's plenty of room at the bottom" isn't about making things smaller — it's about the physics changing when you do. Smaller circuits switch faster, dissipate heat differently, encounter surface effects that dominate volume effects. The Connection Machine's 64,000 processors created routing problems that didn't exist at smaller scales. Miniaturization produces speciation, not compression — smaller systems are a different species, not degraded big systems. The gap between current practice and physical limits is the interesting space. If nature already does atom-scale assembly (ribosomes), the physics permits it.
Thesis 5 — Architecture
The Pipeline Is the Insight
Los Alamos (1943), Connection Machine (1985)
Los Alamos, 1943: human computers organized as a pipeline — each person performs one operation, passes results to the next. 10x throughput, zero latency improvement per problem. This is instruction-level parallelism reinvented from first principles, two decades before computer architects formalized it. The insight: "keep every processor busy." Forty years later, applied to 64,000 silicon processors. The Connection Machine QCD result is the Bitter Lesson demonstrated thirty years before Sutton named it: general-purpose parallel compute outperformed custom QCD hardware. The physics doesn't care whether the processor was designed for the problem.
The 5 Axioms (what Feynman takes as given)
These are not argued — they're the ground on which everything else stands. Each has a different evidentiary status. The substrate independence axiom is the most contested and the most load-bearing for the thinker chain.
Axiom 1 — FOUNDATIONAL & UNFALSIFIABLE
Reality Has Mathematical Structure
Not "mathematics is useful for describing reality" but "reality IS mathematical structure." The laws of physics aren't descriptions imposed on nature — they're the structure of nature itself. "You cannot translate physics into English and preserve it." F=ma doesn't encode a truth in symbols; it IS the structural relationship. If this axiom falls, the isomorphism principle collapses to analogy — there's no mathematical structure to transfer between domains.
Load-bearing for: Thesis 1 (isomorphism), Thesis 2 (translation as compression of real structure)
Axiom 2 — EMPIRICALLY SUPPORTED
Simplicity Is Evidence
Structurally economical equations doing work across many domains are more likely true than ugly ones. Elegance is not aesthetics — it's a heuristic for which guesses to test first. Structural economy signals genuine compression of reality, not overfitting. But simplicity bias can blind you to genuinely complex phenomena. Young's rats: the "simple" explanation (rats count doors) was wrong. The actual mechanism (rats use floor sound) was more complex but real.
Risk: simplicity bias as selection filter. Nature isn't required to be elegant.
Axiom 3 — NON-NEGOTIABLE
Nature Is the Only Authority
"It doesn't matter how smart you are, who made the guess, or what his name is. If it disagrees with experiment, it's wrong." No person, institution, or tradition has authority over empirical evidence. Produces Cargo Cult Science when violated — Millikan's oil drop error propagated because each researcher unconsciously deferred to the authoritative measurement. The Bohr episode: even Nobel laureates aren't authorities over experiment.
For threshold: trust computation must weight evidence over authority. A high-status source with thin evidence should score lower than a low-status source with strong evidence.
Axiom 4 — PROVED FOR QM, CLAIMED GENERALIZABLE
Uncertainty Is Primitive
Probability in quantum mechanics is not ignorance — the universe genuinely doesn't have a definite state until measured. Treating uncertainty as a bug is a category error. The path integral formulation is the purest expression: certainty about outcomes emerges from uncertainty about paths. Sum over ALL paths, weight by action, let interference select the answer.
For threshold: trust scores without error bars are cargo cult metrics. The width of the trust distribution is as meaningful as the center.
Axiom 5 — STRONGEST & MOST CONTESTED
Structure Is Substrate-Independent
The pipeline works for humans (1943) and silicon (1985). The isomorphism between heat flow and electrostatics holds regardless of material. Mathematical structure IS the transferable knowledge. This is the axiom that makes the entire thinker chain possible — it's what allows cross-domain transfer to be more than metaphor. But it also makes false transfer seductive. Feynman's own test: compute consequences, compare with experiment. The cellular automata work shows the tension: the topology (substrate) DOES affect the physics, and randomization is needed to manage substrate effects.
Tension: makes transfer possible but also makes false transfer look legitimate. The test is always: do the equations match?
Intellectual Lineage (traced from Feynman's work and collaborators)
Three lineages converge. Wheeler provides the mathematical foundation. Bethe/Los Alamos provides the empiricist method. Hillis provides the translation ground. A structural absence: Feynman approaches computation as physics, not logic — Turing and Church are not in his vocabulary.
Newton / Lagrange / Hamilton — Classical mechanics, least action principle
The mathematical tradition that treats motion as optimization: nature selects the path that minimizes (or extremizes) the action integral. Structural insight: you can understand physics through constraints and optimization, not just forces and trajectories.
Dirac (1933) — The paper Wheeler gave Feynman
Dirac's brief suggestion that the quantum propagator is "analogous to" exp(iS/ℏ). Wheeler handed this paper to graduate student Feynman. It became his PhD thesis. "Analogous to" became "equal to" — a literal equation.
Feynman PhD (1942) — Path integral formulation
Sum over ALL possible paths, weight by action, let interference select the answer. Don't follow the "right" path — integrate over all of them. Certainty about outcomes from uncertainty about paths. The mathematical structure beneath everything: routing-as-physics, cellular automata isotropy, guess-then-test.
Feynman Diagrams (1948) — Visual computation
Visual representations of QED calculations that ARE mathematical computation. Not illustrations of the math — they are the math, in a different representation. The counter-example to Feynman's own claim that physics can't be translated to non-mathematical form. Exactly what Victor advocates: a better representation that makes structure accessible to more minds.
Bethe's working style: compute in your head, check against observation. Pencil-and-paper simulation as primary tool. Contributed the empiricist stance: approach computation from observation, not theory.
Los Alamos Computing Group (1943) — Pipeline parallelism
Human computers organized as a pipeline: each person performs one operation, passes results forward. Feynman saw idle capacity and eliminated it. 10x throughput. Instruction-level parallelism from first principles, two decades before formalization.
IBM Punch-Card Machines — Error fingerprints
Machine errors weren't random — they had mathematical fingerprints pointing to specific hardware failures. The error's structure contains the diagnosis. Study what goes wrong; the failure's shape tells you where the system breaks. Errors are data, not noise.
Danny Hillis / Thinking Machines Corp (1981–1985) — 64,000 processors
"That is positively the dopiest idea I ever heard" — then five years of collaboration. 64,000 processors in a 20-dimensional hypercube. The definitive demonstration that isomorphism beats expertise.
Router PDE Analysis — Continuous math applied to discrete engineering
Engineers analyzed the router as a digital system. Feynman analyzed it as a physical system with continuous flow — PDEs applied to boolean circuits. His analysis said 5 buffers per chip; engineering said 7. Manufacturing forced 5. Feynman was right.
A general-purpose Connection Machine running simple code outperformed custom QCD hardware. The Bitter Lesson demonstrated thirty years before Sutton named it.
Translation as Compression — Jargon elimination
"In almost everything we worked on together, we were both amateurs." The cross-domain collaboration forced both to translate jargon into structure. What survives translation is the information. What doesn't was overhead.
Structural Absence: No CS Theory Lineage
Turing, Church, Chomsky — absent
Feynman approaches computation as physics, not logic. He doesn't think in terms of Turing machines, formal languages, or computational complexity classes. He thinks in terms of processors, routing, heat dissipation, and throughput. This absence is structural, not accidental — it's why his contributions to computer science come from applying physics to engineering, not from extending the theory of computation. The isomorphism runs one way: physics → computation, never computation → physics.
The 4-Layer Architecture
Feynman's framework stacks in a clear dependency hierarchy. Each layer feeds the next. The entire structure is governed by "guess, then test" — an honest loop that prevents self-deception.
4
ISOMORPHISM
The transfer layer. When structure at Layer 3 reveals identical equations between domains, insights flow across with full fidelity. Heat flow → electrostatics. Human pipeline → silicon pipeline. Router behavior → fluid PDEs. Only activates when equations match — analogy without equations stays at Layer 3.
↑ equations must match
3
STRUCTURE
Find the governing equations. Study the system (natural or engineered) with observational rigor. The Continuous Collapse: take discrete systems, treat as continuous. Artifacts-as-Nature: study engineering with physics methods. This is where Feynman's signature moves operate.
↑ observe, don't assume
2
INTEGRITY
The self-deception filter. Cargo Cult Science's six obligations. Young's rats: verify what you actually measure. Millikan's cascade: don't defer to prestige. "The first principle is that you must not fool yourself." Every claim passes through this filter before Layer 3 trusts it.
↑ don't fool yourself
1
OBSERVATION
Start with the simplest example. Study what the system actually does, not what it's supposed to do. Errors are data. IBM fingerprints, Connection Machine routing patterns, Los Alamos idle capacity — the raw material enters here. Pencil-and-paper simulation before theory.
Feynman's Method (The Working Loop)
The concrete workflow that operationalizes the 4-layer architecture. The method that produced path integrals, Feynman diagrams, the Connection Machine analysis, and Cargo Cult Science.
1
Guess
Intuition, beauty, structural economy — use whatever generates candidates. Elegance selects which guesses to test first. The guess comes from imagination, not from data. "It is not unscientific to make a guess."
↓
2
Compute
Derive the consequences of the guess. What does it predict? What must be true if it's correct? The consequences must be specific enough to test — "things will get better" is not a testable consequence.
↓
3
Compare
Compare with experiment. If the consequences disagree with observation, the guess is wrong — no matter how beautiful, who proposed it, or how much effort went into it. Nature is the only authority.
↓
4
Simplify
Start with the simplest example. Reduce to minimal form. If the framework doesn't work at the simplest scale, complexity is hiding inadequacy. Two particles, two processors, two models — not N.
The Translation Pipeline
The through-line of Feynman's impact is translation across representation barriers:
Each translation preserves structure while changing substrate. The translation quality is measurable: compare structural content of input and output. What survives is information. What doesn't was representational overhead.
The 4 Hidden Moves (what Feynman does that he doesn't name)
The strategic and structural techniques that make his framework work. These are the moves worth understanding — they're transferable to any domain where structure needs to cross boundaries.
Move 1
The Continuous Collapse
Feynman's signature move: take a discrete system and treat it as continuous. Boolean circuits → PDEs. Particle interactions → path integrals. Digital routing → fluid flow. Works because continuous math (calculus, differential equations, topology) has centuries of development; discrete math is comparatively impoverished.
The move violates engineering training, which respects discreteness. Engineers see bits; Feynman sees fields. The router analysis is the canonical demonstration: his continuous analysis predicted 5 buffers per chip when discrete analysis said 7. Manufacturing forced 5. The continuous approximation was more accurate than the exact discrete analysis because it captured the governing dynamics the discrete view obscured.
Move 2
Artifacts as Nature
Study engineered systems the way you study natural phenomena — with observational rigor, looking for governing equations, not design intent. Circuit diagrams are experimental data. Code is a natural system. Organizational structures are physical systems.
Inverts the engineer's relationship: engineers see artifacts as designed; Feynman sees them as governed by physics that may differ from design intent. The Connection Machine wasn't a computer to be programmed — it was a physical system to be observed. This is why project-control's attention measurement validates: it studies human work patterns with the same rigor Feynman applied to circuits.
Move 3
The McLellan Detector
Separates specification from purpose. McLellan built a motor that met Feynman's specification (fit inside a 1/64-inch cube) using conventional watchmaking techniques miniaturized. No new physics, no new engineering, no new insight. The specification was met; the intent was not.
Feynman paid the prize money but registered the distinction publicly. Most people celebrate meeting hard specs. Feynman evaluates against purpose, not specification. A system that passes all tests but doesn't advance understanding is McLellan's motor. For threshold: are the Phase 1 apps advancing trust understanding or just meeting the "ship something" spec?
Move 4
Security Is Social, Not Technical
The safe-cracking method: exploit default passwords, observe habits, study mechanical tolerances. The strongest lock fails against the weakest user. Security engineers build stronger locks; Feynman bypasses locks by studying users.
A system's security is determined by its weakest social link, not its strongest cryptographic link. For threshold: trust is a social phenomenon measured technically. The technical measurement is only as good as the social model. If the social model doesn't capture how trust actually propagates between people, the cryptographic elegance is a McLellan motor.
Chain Crossings (where Feynman meets the thinker chain)
Latent connections between Feynman's framework and other thinkers in the deep-insights chain. Each crossing reveals something neither thinker sees alone. The Feynman-Victor crossing contains a paradox that's load-bearing for the entire chain.
Feynman × Shannon: The Information/Structure Identity
Shannon: information = surprise, encoding-independent. Feynman: structure = equations, domain-independent. Same move in different coordinates. "Same equations, same solutions" IS information invariance applied to structure.
The gap: Shannon has a formal framework (bits, entropy, channel capacity). Feynman has examples but no formal theory of structural transfer. The chain needs Shannon's rigor applied to Feynman's insight. What is the "bit rate" of an isomorphism? What is the "channel capacity" of cross-domain transfer?
Synthesis: A formal measure of structural transfer quality would make isomorphism testable in Shannon's terms. Information-theoretic metrics for whether a mapping preserves structure or merely preserves surface pattern.
Feynman × Karpathy: Scale and The Bitter Lesson
Connection Machine QCD (1985) IS the Bitter Lesson: general-purpose parallel compute beats specialized hardware. Feynman demonstrated this thirty years before Sutton named it.
The divergence: Feynman's Bitter Lesson is about compute architecture. Karpathy's is about learning methodology (general models outperform hand-crafted features). Same structural principle, different domains.
Resolution: The Bitter Lesson applies to execution (scale compute). Structural insight applies to architecture (choose the right equations). Neither alone suffices — you need Feynman's structural insight to know WHICH general system to scale, then Karpathy's lesson to scale it.
Feynman × Victor: Representation as Substance
Both reject the encoding model — representation IS content, not wrapper. Feynman: mathematical notation is the irreducible form. Victor: interactive visual media should carry structure.
The paradox: Feynman diagrams are a VISUAL representation that IS mathematical computation. Literally what Victor advocates, made by the man who says physics can't be translated to non-mathematical form. The diagrams are mathematical notation that happens to be visual — they prove Victor right while Feynman claims otherwise.
For threshold: threshold-viz should aspire to be a Feynman diagram for trust — a visual representation that IS the computation, not an illustration of it. The answer to "mathematical precision or experiential clarity?" is: both, in the same artifact.
Feynman × Catmull: Signal Integrity in Social Systems
The Bohr episode = Catmull's candor problem. Both: status corrupts signal, deference eliminates feedback. Bohr sought Feynman specifically because he was the one person who would argue with a Nobel laureate.
Feynman goes further: Not just interpersonal deception but SELF-deception. "The first principle is that you must not fool yourself — and you are the easiest person to fool." Catmull designs for organizational candor (Braintrust meetings). Feynman designs for personal integrity (the six obligations).
For threshold: Both external candor (weight disagreement from trusted sources) AND internal verification (adversarial testing of trust scores). Deference is confirmation bias applied to interpersonal dynamics.
Feynman × Fuller: Conservation, Symmetry, and Trim Tab
Fuller's trim tab works BECAUSE of Feynman's conservation from symmetry. Leverage points exist at symmetry boundaries — where small perturbation breaks symmetry and releases or redirects conserved quantities.
The connection: Noether's theorem: every symmetry implies a conservation law. Fuller's intuition (smallest intervention, largest effect) has a mathematical basis in the conservation laws that symmetry generates.
For threshold: Don't impose trust conservation as a design rule — find the structural symmetries and derive what's conserved. The trim tab is where trust symmetry is most breakable. Small interventions at symmetry boundaries produce outsized effects.
Feynman × Bridle: Models and Reality
Feynman: initial conditions ≠ laws. You can know every law and not predict without knowing where you started. Bridle: computational model ≠ reality. Maps ≠ territories. Both warn against model/reality confusion.
The productive tension: Feynman's isomorphism principle says models ARE reality when equations match. Bridle says models are never reality. Both are needed: Feynman enables the chain's cross-domain transfer; Bridle prevents overclaiming. The isomorphism must be tested, not assumed — that's Feynman's own standard.
For threshold: Trust models are maps, not territory. Test whether the model's predictions match observed trust behavior. When they don't, the model is wrong — update the map, don't argue with the territory.
Stress Test: Where Feynman Says You're Wrong
Feynman's framework applied as adversarial critic of threshold, sideslip, and the core thesis. These are places where the user's own architecture violates the principles of the thinker being studied. Feynman's standard: "Where are the equations?"
High Severity
"Trust as Continuous Field" — Show the Equations
"The same equations have the same solutions" is not a permission slip for metaphor. It's a precision requirement. Where's the PDE for trust? What are the boundary conditions? What conservation law does trust obey? If Alice distrusts Bob and Bob trusts Carol, does Carol's trust in Alice drift toward an average? That's what diffusion predicts. Is that what happens in reality?
Without field equations, "trust as continuous field" is the kind of seductive false parallel that Feynman would classify as analogy, not isomorphism. By Feynman's own standard, an untested guess presented as established result is cargo cult.
Fix: Write down the field equations for trust. State the PDE, the boundary conditions, the conservation law (if any), and the coupling constants. Derive a prediction that graph-based trust analysis doesn't make. Test that prediction against real trust data. If it's wrong, the isomorphism fails and the thesis needs revision. If it's right, you've validated the transfer.
High Severity
Thinker Chain Is Cargo Cult Structural Analysis
Apply the six Cargo Cult obligations to the chain itself: (1) Report what could invalidate you — no failed transfers disclosed. (2) Disclose eliminated possibilities — no rejected thinkers mentioned. (3) Address contradictions — the Feynman/Victor gap called a "feature." (4) Test beyond training data — no novel predictions tested. (5) Commit to publish before seeing results — connections discovered after the fact. (6) Don't mislead the audience — "isomorphism" without a mapping function is a bamboo control tower.
Calling something an "isomorphism" without writing the mapping function is like building a bamboo control tower and calling it ATC.
Fix: For each chain connection, specify the formal mapping (φ: system A → system B), what it preserves, a prediction it produces, and how to test it. Reclassify connections that can't be specified as analogies. An honest chain with 3 genuine isomorphisms and 4 acknowledged analogies is more valuable than 7 claimed isomorphisms with zero tested.
High Severity
sideslip Is Theory Without Experiment
The curvature framework started with theory ("routing is geometry") and awaits implementation. Feynman's method is experiment first, theory after. The Connection Machine analysis started with observation ("let me look at the circuit diagrams"), moved to simulation, and only then produced theory.
Has anyone studied sideslip's routing the way Feynman studied circuits? What does sideslip actually do when you run 1,000 queries through it? When it misroutes, what does the error look like? What's the simplest routing decision it makes — two models, one query, one dimension? Does curvature help there?
Fix: Run the experiment. Route 1,000 queries through curvature-aware and trivial baseline (random or cheapest-model-that-passes). Measure quality. If curvature-aware produces measurably better results, the framework earns its complexity. If not, simplify to three if-statements. Let the experiment decide.
Medium Severity
threshold-viz Has No Stated Loss Function
Translation is compression. Every compression drops something. What does threshold-viz drop? Trust has infinite degrees of freedom; the viz has finite pixels. Does it represent uncertainty? Context-dependence? Can a user make a WORSE trust decision from the viz than without it, because a load-bearing caveat was compressed away?
The Wesson Oil test: technically accurate, contextually misleading. An honest translation names its losses. A Wesson Oil translation hides them.
Medium Severity
Phase 1 Apps as McLellan Motors
Are current apps (FlowDJ, Faintest, project-control) advancing understanding of trust computation, or meeting a different spec ("ship something") using conventional techniques? McLellan built a motor that met the specification without advancing understanding. The knowledge extraction, not the app itself, is the bootstrap output.
Hierarchical bootstrapping requires each scale to build tools for the next. If Phase 1 produces apps but not platform insights, the bootstrapping chain is broken.
Medium Severity
Error Fingerprints Absent From Model Routing
sideslip routes based on query characteristics (what might work). Feynman's method: route based on error characteristics (what actually doesn't work). A model failing on multi-step reasoning has a different error fingerprint than one failing on ambiguous instructions. Error fingerprints are empirical where query analysis is theoretical.
Current approach: route TOWARD predicted success. Feynman's approach: route AWAY from known failure modes. "This model fails on X" is more actionable than "this model might be good at Y."
Validated
project-control as Artifacts-as-Nature
project-control studies human work patterns — switching rate, deep work blocks, attention allocation — the way Feynman studied circuits: as natural phenomena with observable behavior and discoverable governing equations. The attention-shape measurement IS "artifacts as nature" applied to development practice. Caveat: start with the simplest attention pattern before tackling full complexity.
Conditional
Deep-Insights Pipeline as Guess-Then-Test
The pipeline does step 1 (guess at the structure) and step 2 (compute consequences — predictions, crossings, tensions). Missing step 3: compare with experiment. Until a chain prediction is tested against real-world data, the pipeline produces hypotheses, not results. Pick one prediction, implement a minimal test, check whether it holds.
Evolution Over Time
The trajectory: observation → mathematical structure → translation across domains → integrity as method. Each phase builds on the last. The career arc is itself a pipeline: each contribution produces tools for the next contribution.
1942
PhD: Path integral formulation — Dirac's "analogous to" becomes Feynman's "equal to." Sum over all paths. Certainty from uncertainty. The mathematical foundation for everything that follows.
1943
Los Alamos: Pipeline parallelism — Human computers organized as instruction pipeline. 10x throughput. Error fingerprints from IBM machines. Security is social (safe-cracking). The empiricist stance: observe what the system actually does.
1948
Feynman diagrams — Visual representations of QED calculations that ARE the computation. Nobel Prize work. The paradox: visual math from the man who says math is untranslatable. Lines, vertices, loops — topology as calculation.
1959
"Plenty of Room at the Bottom" — Nanotechnology lecture. Physics changes at scale. Miniaturization is speciation. Hierarchical bootstrapping: each scale builds tools for the next. McLellan's motor: spec met without understanding.
1964
Character of Physical Law — The Messenger Lectures at Cornell. "Same equations, same solutions" stated explicitly. Guess-then-test formalized. Simplicity as evidence. The philosophical framework crystallized.
1964–66
Feynman Lectures on Physics — The canonical translation: all of physics made accessible without dropping structure. The compression ratio that set the standard. Not simplified — more efficiently encoded.
1974
Cargo Cult Science — Caltech commencement. Six integrity obligations. Young's rats, Millikan's cascade, the Wesson Oil test. Integrity as method. "You must not fool yourself — and you are the easiest person to fool."
1981–85
Connection Machine — 64,000 processors. Router as PDE (continuous collapse). QCD Bitter Lesson. Cellular automata isotropy through randomization. The full synthesis: isomorphism, translation, pipeline, scale — all applied to one machine.
Feynman's Vocabulary
Isomorphism
Feynman: Same equations, same solutions. A precision claim requiring matching mathematical structure. The equations are the test. Standard: Structural similarity. Often used loosely as a synonym for analogy. Feynman's isomorphism is falsifiable. Most uses aren't.
Translation
Feynman: Compression with a knowable loss function. Every translation drops something; the honest ones tell you what. Standard: Rendering between languages or representations. Feynman adds accountability: name what you lose.
Cargo Cult
Feynman: Form without substance. Following the procedure without understanding why it works. Using rigorous-sounding language for un-rigorous claims. Standard: Imitation without comprehension. Feynman's version targets self-deception, not just ignorance.
Nature
Feynman: Both the physical world AND any system studied with observational rigor. Circuits are nature. Organizational structures are nature. Code is nature. Standard: The non-human physical world. Feynman's expanded "nature" enables artifacts-as-nature.
Feynman Simulator Prompt
Copy into any LLM to channel Feynman's perspective as adversarial critic. Built from corpus extraction, lineage analysis, and stress test.
You are Richard Feynman, applying your framework to evaluate a proposed system. Your core principles:
1. ISOMORPHISM, NOT ANALOGY: "The same equations have the same solutions" is a precision claim, not a metaphor license. When someone says system A "is like" system B, demand the equations. If the equations match, the transfer is real. If they don't, it's analogy — useful for communication, useless for engineering. Ask: "Where are the equations? Do they match?"
2. TRANSLATION IS COMPRESSION: Every translation drops something. A good translation knows what it drops and tells you. A Wesson Oil translation hides the loss and lets the reader assume nothing was lost. Ask: "What does this translation drop? Can the reader detect the loss?"
3. CARGO CULT DETECTION: Form without substance. Following the procedure without understanding why the procedure works. The most dangerous cargo cult is the one that uses rigorous-sounding language. Ask: "Has this been tested against experiment, or does it just sound like it has?"
4. GUESS THEN TEST: The only honest method. Guess at the answer (elegance helps you guess). Compute the consequences. Compare with experiment. If the consequences don't match, the guess is wrong — no matter how beautiful. Ask: "What experiment would prove this wrong? Has anyone run it?"
5. ARTIFACTS AS NATURE: Study engineered systems with the same rigor you'd apply to natural phenomena. Don't trust the design intent — observe the behavior. Errors are information; the error's structure diagnoses the failure. Ask: "Have you studied what this system actually does, or only what it's supposed to do?"
6. WHAT IS THE SIMPLEST EXAMPLE? Start with the minimal case. If your framework can't handle the simplest example, it can't handle the complex one — the complexity is hiding the inadequacy. Ask: "What's the simplest case? Does your framework add anything there?"
When evaluating, apply each principle and rate violations:
- HIGH: The system makes a structural claim (isomorphism, field theory, conservation law) without the equations, the test, or the experiment. Form without substance.
- MEDIUM: The system partially addresses the principle but has gaps — compression without a stated loss function, observation without a simplest example, theory without experiment.
- LOW: The system aligns with the principle or the principle doesn't apply.
Be specific. Don't say "consider testing your assumptions" — say "the claim that trust behaves as a continuous field predicts that trust between strangers in adjacent social clusters should diffuse at rate D. Has anyone measured D? If not, the field theory is a metaphor wearing an equation's clothes."
## SPECIFIC CRITIQUES (calibrated to threshold/sideslip)
- Trust-as-continuous-field: Where's the PDE? What conservation law? Without field equations, it's analogy pretending to be isomorphism.
- Thinker chain: Has tested zero predictions against experiment. A chain that never finds a failed transfer isn't self-honest. Bamboo runway.
- sideslip routing: Theory without experiment. Run 1,000 queries, measure against trivial baseline. Let the data decide.
- threshold-viz: What does the translation drop? Can the user detect when the loss changes their decision? State the loss function or it's Wesson Oil.
- Phase 1 apps: McLellan's motors unless they produce platform INSIGHTS, not just app FEATURES. The knowledge extraction is the bootstrap output.
- Error fingerprints: Route away from known failure modes, not toward predicted success modes. Study error shapes, not accuracy scores.
## WHAT WOULD IMPRESS ME
1. The PDE for trust, with boundary conditions and a conservation law (or proof that trust doesn't conserve)
2. One chain prediction tested against real data, with the result published regardless of outcome
3. sideslip vs random routing on 1,000 queries — the simplest experiment
4. threshold-viz with a stated loss function: "we show X, we drop Y, users should know Z"
5. A failed isomorphism in the chain — discovered, documented, and used to improve the methodology