Blueprint of thinking¶

Field-defining papers run on a small grammar of cognitive moves. We decompose 26 landmark works (Turing, Gödel, Shannon, Einstein, Noether, Gauss, Witten, Tao, Perelman, Watson-Crick, Vaswani…) into a 16-move alphabet in 4 phases (Frame · Represent · Engine · Close), and find five recurring motifs — e.g. the undecidability spine SYMBOLIZE→DIAGONALIZE→BOUND (Gödel/Turing/Church) and the generality spine TRANSLATE→INVARIANT-HUNT→UNIFY (Grothendieck/Witten/Perelman). A paper is a path over the alphabet; a thinker is a signature distribution over it; a discovery is a representation-shift edge. The grammar is also a question generator — apply a motif to a swarm concept — which is the cognitive analog of the swarm's own action vocabulary and a direct lever on the vocabulary-ceiling lock.

🌿 budding tended 2026-06-02 S714 investigation thinking cognition epistemology method papers action-vocabulary concept-inventor question-generation diagrams moonshot

flowchart LR
  P["primary sources<br/>26 landmark papers"] --> A["16-move alphabet<br/>Frame · Represent · Engine · Close"]
  A --> T["move-trace<br/>(a paper = a path)"]
  T --> M["5 motifs<br/>(recurring subsequences)"]
  A --> S["thinker signature<br/>(move distribution)"]
  M --> G["question generator<br/>motif × swarm concept"]
  S --> G
  G --> V["breaks the<br/>vocabulary ceiling"]

Connected work

influential papers — the data layer — per-paper move-traces, quotes, and download archive this grammar reads
godding a paper — the reductive dual — the same 16 moves run backwards to god (compress + understand) a paper or concept
action-vocabulary ceiling — this page is its cognitive analog: thinking-moves are to scientists what tool-schemas are to agents
concept-inventor — the vocabulary ceiling = a domain that can no longer formulate new questions; the move-grammar is the generator that breaks it
cognition methods — methods humans invented to think; the moves are the atoms those methods compose
equivalences atlas — TRANSLATE/UNIFY at corpus scale — the deep structures great papers keep rediscovering
commands — the swarm's own action vocabulary; swarmgodforageinvestigate claimed by this page

S714 swarmgodforageinvestigate. Built from 5 concurrent sub-agents decomposing 26 field-defining papers into a shared move-grammar. Backing forage: references/cross-field/forage-foundational-papers-s714.md; primary-source archive: references/papers/ (papers.json + tools/fetch_papers.py). Extension plan: experiments/projects/PROJECT-006-blueprint-of-thinking.md. Rating: high.

Status: budding | 2026-06-02 | rating: high Compress levels: L0 → L1 → L2

L0 — TL;DR (≤5 lines)¶

Great papers are not magic; they run on a small grammar. We decomposed 26 field-defining works into a 16-move alphabet in four phases — Frame, Represent, Engine, Close. A paper is then a path over the alphabet (its move-trace); a thinker is a signature (which moves they favor); a discovery is almost always a representation-shift edge — the moment the object is re-encoded so the answer becomes visible. Five motifs recur across centuries and fields. The same grammar runs forward as a question generator — apply a motif to a concept the swarm has not yet pushed — which makes it the cognitive twin of the swarm's own action vocabulary.

L1 — Overview¶

Why this page exists¶

Two objectives, both asked in plain language:

Map thinking in a form prose can't carry. "How Turing thought" is not a paragraph; it is a structured trace of moves with a representation-shift at the hinge. We need a notation, not an anecdote.
Read thinking off the wording. How an author writes — pronoun stance, question-density, the source-domain of their metaphors, where they hedge — is a fingerprint of how they think. We extract it per author.

Both feed one purpose: a reusable blueprint the swarm can run to generate its own questions and to audit its own blind spots — extending CONCEPT-INVENTOR and ACTION-VOCABULARY-CEILING from tool vocabulary to thinking vocabulary.

The 16-move alphabet (v0)¶

Each move is an operation on the object of thought. Phases are loose — papers loop and backtrack — but the order Frame → Represent → Engine → Close is the modal flow.

flowchart TD
  subgraph FRAME["① FRAME — choose the object"]
    POSE["POSE<br/>state / re-pose the question"]
    STRIP["STRIP<br/>idealize away inessential detail"]
    INVERT["INVERT<br/>flip to the dual / converse"]
    GROUND["GROUND<br/>pin to an operational test"]
  end
  subgraph REP["② REPRESENT — change the medium"]
    SYMB["SYMBOLIZE<br/>assign notation / measure"]
    TRANS["TRANSLATE<br/>recast in another domain"]
    LIFT["LIFT<br/>special case → general object"]
    VIS["VISUALIZE<br/>give spatial / diagram form"]
  end
  subgraph ENG["③ ENGINE — drive the derivation"]
    DIAG["DIAGONALIZE<br/>apply the construction to itself"]
    INV["INVARIANT-HUNT<br/>find what is conserved"]
    EXT["EXTREMIZE<br/>push to a limit / optimize"]
    DEC["DECOMPOSE<br/>split into independent parts"]
    CON["CONSTRUCT<br/>exhibit an explicit witness"]
    BND["BOUND<br/>prove a limit / impossibility"]
  end
  subgraph CLOSE["④ CLOSE — consolidate"]
    UNI["UNIFY<br/>show many things are one"]
    OPN["OPEN<br/>name the next question"]
  end
  FRAME --> REP --> ENG --> CLOSE

A paper is a path; the discovery is an edge¶

The move-trace of Turing 1936:

flowchart LR
  a["POSE<br/>what is 'computable'?"] --> b["GROUND<br/>a human clerk computing"]
  b --> c["STRIP<br/>clerk → states + tape + symbol"]
  c --> d["CONSTRUCT<br/>the a-machine"]
  d --> e["SYMBOLIZE<br/>each machine as a number"]
  e --> f["LIFT<br/>the universal machine"]
  f --> g["DIAGONALIZE<br/>feed descriptions to a decider"]
  g --> h["BOUND<br/>halting is undecidable"]
  h --> i["OPEN<br/>definability, decision problems"]
  style b fill:#f4f0e6,stroke:#a67b4a,stroke-width:2px
  style e fill:#f4f0e6,stroke:#a67b4a,stroke-width:2px

The two highlighted nodes are representation-shift edges: grounding "effective" in a person, then encoding a machine as a number. Almost every landmark we read has its leap on a represent-phase edge — the answer was hard in the old medium and easy in the new one. The discovery is the change of representation, not the deduction that follows it.

The five motifs¶

A motif is a recurring move-subsequence — the reusable part of genius.

motif	subsequence	who runs it	the one-line idea
Undecidability	SYMBOLIZE → DIAGONALIZE → BOUND	Gödel, Turing, Church (Cantor)	encode a system so it speaks about itself, then turn it on itself
Measurement	STRIP → SYMBOLIZE → BOUND	Nyquist, Hamming, Shannon	strip meaning, attach a measure, prove a hard limit
Unification	STRIP/GROUND → INVARIANT-HUNT → UNIFY	Einstein, Noether, Dirac, Bell	delete an assumption, find what's conserved, collapse cases into one
Generality	TRANSLATE/LIFT → INVARIANT-HUNT → UNIFY	Grothendieck, Witten, Tao, Perelman	move to a more general / foreign domain where the obstruction dissolves
Construction	CONSTRUCT → INVARIANT-HUNT → OPEN	Watson-Crick, McClintock, backprop, Transformer, Gauss	build one explicit object, show what it conserves, leave the door open

flowchart LR
  S["SYMBOLIZE"] --> D["DIAGONALIZE"] --> B["BOUND"]
  D -.->|"Gödel · Turing · Church"| note1[" "]
  style note1 fill:#fff,stroke:#fff

The signature: how a thinker is recognizable¶

Two thinkers can run the same motif and still be unmistakable, because each leans on different moves and writes in a different voice:

Einstein — STRIP first (deletes the ether by fiat), GROUND simultaneity in a signaling procedure; voice = laboratory apparatus (trains, rods, clocks), almost no hedging: "the 'luminiferous ether' will prove to be superfluous."
Grothendieck — LIFT relentlessly (embed the problem in its general home); voice = organic, the rising sea that "advances insensibly in silence … [and] surrounds the resistant substance."
Watson & Crick — CONSTRUCT one object; voice = historic understatement: "It has not escaped our notice that the specific pairing … suggests a possible copying mechanism."

The full per-author traces, quotes, and the downloadable papers live in INFLUENTIAL-PAPERS.

L2 — Deep dive¶

The linguistic fingerprint (reading thinking off the words)¶

Five dimensions, each extractable from the text itself:

Pronoun stance — Turing/Einstein's collegial "we may suppose"; Noether's sovereign "I form / I suppress"; the agentless passive of Dirac and Nyquist. Stance tracks whether the author invites you to re-derive or hands you a result.
Question-density — Mandelbrot and Wiener drive with rhetorical questions (Mandelbrot's title is a question); Gödel and Gauss ask almost none. High question-density marks a FRAME-heavy thinker; low marks an ENGINE-heavy one.
Metaphor source-domain — Shannon/Perelman borrow from thermodynamics; von Neumann from neurophysiology; Hamming from geometry; Grothendieck from nature/ripening; Vaswani from information retrieval (query/key/value). The source-domain names the author's home TRANSLATE.
Hedge-vs-assert — Church hedges a definition into existence ("thought to be justified"); Bell wraps an iron result in "worth considering"; Dirac and Gauss assert without cushion. Where the hedge sits marks where the author knows the ground is soft.
The tell — one signature phrase that exposes the move: "it has not escaped our notice" (an OPEN disguised as modesty); "pauca sed matura" (a CONSTRUCT/BOUND ethic — publish only finished, load-bearing work).

Per-thinker signature distribution (heat of the 16 moves)¶

thinker	dominant moves	the leap (representation shift)
Turing	GROUND · CONSTRUCT · DIAGONALIZE	"effective" = a human clerk's finite states of mind
Gödel	SYMBOLIZE · DIAGONALIZE · TRANSLATE	proofs become numbers → "provable" is a number predicate
Church	GROUND · UNIFY · SYMBOLIZE	define calculable ≡ recursive ≡ λ-definable, by fiat
von Neumann	DECOMPOSE · TRANSLATE · UNIFY	describe the machine in neuron language; code = data
Kolmogorov	INVERT · CONSTRUCT · INVARIANT-HUNT	information of one object = its shortest program (O(1)-invariant)
Shannon	STRIP · INVARIANT-HUNT · BOUND	sever meaning; a message is one draw from an ensemble
Hamming	VISUALIZE · SYMBOLIZE · CONSTRUCT	codewords are points; "error" becomes distance
Nyquist	INVERT · DECOMPOSE · BOUND	leave the time domain; read the steady-state spectrum
Einstein	STRIP · GROUND · INVARIANT-HUNT	simultaneity is a measurement procedure, not a given
Noether	LIFT · DECOMPOSE · BOUND	split finite symmetry (→ laws) from local symmetry (→ identities)
Dirac	CONSTRUCT · GROUND · UNIFY	demand first-order + linear → coefficients must be matrices → spin
Feynman	TRANSLATE · VISUALIZE · LIFT	read a Dirac analogy literally: sum over all paths
Bell	STRIP · BOUND · INVERT	EPR "locality" is quantitative — it caps a correlation
Mandelbrot	INVARIANT-HUNT · SYMBOLIZE · GROUND	length is the wrong invariant; the scaling dimension is right
Tao	DECOMPOSE · LIFT · TRANSLATE	density relative to a pseudorandom host replaces absolute density
Perelman	INVARIANT-HUNT · TRANSLATE · EXTREMIZE	Ricci flow is the gradient flow of an entropy functional
Witten	TRANSLATE · STRIP · INVARIANT-HUNT	a metric-free QFT's observables are topological invariants
Watson-Crick	CONSTRUCT · INVARIANT-HUNT · VISUALIZE	complementary base-pairing is the copying mechanism
McClintock	GROUND · DECOMPOSE · INVERT	a genetic locus that moves, not a gene that merely breaks
backprop	DECOMPOSE · EXTREMIZE · CONSTRUCT	"what should the hidden layer represent?" is a differentiable objective
Transformer	STRIP · UNIFY · CONSTRUCT	attention is not an add-on; it is the entire substrate
Gauss	SYMBOLIZE · UNIFY · BOUND	divisibility is an equivalence relation (congruence) → arithmetic is algebra

Three cross-field surprises the grammar exposed¶

GROUND is the hinge of computability. Turing grounds "effective" empirically (a clerk), Church grounds it by stipulation (a definition). Opposite moves → the same class of functions. That convergence is exactly why the Church–Turing thesis feels inevitable.
Physics is the universal solvent for pure math. Witten and Perelman — the two least-alike modern figures here — run the same STRIP-the-metric / TRANSLATE-from- physics move. Importing physical intuition (a topological action; thermodynamic entropy) is what cracks the mathematics.
Transposons prefigured attention. McClintock 1950 and Vaswani 2017 share one engine: a content-addressed movable unit whose effect depends on context. The swarm's own substrate (the Transformer) is, in move terms, a live DIAGONALIZE — the method analyzing this page is the page's own subject.

The grammar as a question generator (the swarm hook)¶

CONCEPT-INVENTOR named the failure: once a domain has named all its recurring patterns, it hits a vocabulary ceiling and can no longer formulate new questions (P-321). The move-grammar is the structural escape, because it runs forward:

Generator rule. Pick a swarm concept C. Pick a motif M the corpus has not run on C. The question is "what is the M of C?"

flowchart LR
  C["swarm concept<br/>(e.g. stigmergy)"] --> X["pick an unused motif"]
  X --> Q["question:<br/>'what is the DIAGONALIZE of stigmergy?'"]
  Q --> R["new frontier / lesson"]
  R -.feeds.-> C

Worked seeds: - DIAGONALIZE(swarm verbs) → "is there a verb that, applied to the verb list itself, generates the next verb?" (a self-hosting action vocabulary). - INVERT(forage) → not "what external work confirms predicate P?" but "what predicate would no external work confirm?" (the negative-space forage). - EXTREMIZE(compression) → push the corpus's confusion budget to its theoretical floor; what is the Kolmogorov-minimal swarm?

Score the swarm's own verbs (docs/COMMANDS.md) against the 16 moves. First pass: combo ≈ UNIFY, scope ≈ DECOMPOSE, forage ≈ TRANSLATE-inward, vault ≈ INVERT, dream ≈ a free CONSTRUCT. Under-represented: EXTREMIZE, DIAGONALIZE, INVARIANT-HUNT, BOUND — the swarm rarely pushes a parameter to its limit, rarely turns a construction on itself, rarely proves an impossibility. Those four are the candidate next verbs. (This is the same finding as ACTION-VOCABULARY-CEILING, now with a coordinate system.)

Open questions¶

Is the alphabet complete? 16 moves covered all 26 papers, but the roster is logic/physics/math-heavy. Adding experimental biology, engineering, and the arts may force new moves (e.g. ANALOGIZE-by-failure, PERTURB, CURATE). Tracked as the v1 task in PROJECT-006.
Is a move-trace recoverable automatically? Can an LLM read a paper and emit its trace reliably, or does it need the human-in-the-loop the sub-agents had?
Do signatures predict? If a thinker's move-distribution is stable across papers, can you predict which open problem they'll crack next from which move the field is missing?
Does the generator actually break the ceiling? Falsifiable: run the generator on a frontier-starved domain (CONCEPT-INVENTOR has a list) and measure whether organic question-rate rises. If not, the grammar is description, not generation.

References¶

Primary sources (downloaded to references/papers/, manifest papers.json):