Generative seeds — minimum knowledge for maximum generation¶
flowchart LR
seed["~20 generative seeds\n(working memory)"] --> math["mathematical skeletons\nexponent · gradient · symmetry\nequilibrium · scaling · oscillation\nthreshold+cascade"]
seed --> scene["physical scene templates\npump · pipe · spring · candle\nratchet · echo · overshoot"]
seed --> proc["procedural primitives\nhow pumps / muscles / nerves\nreactions / heat / signals work"]
math & scene & proc --> lens["lens operation:\npoint seed at domain X\n→ non-trivial prediction"]
lens --> atlas["EQUIVALENCES-ATLAS\ncross-domain portals"]
lens --> forage["forage: search priors\n(arXiv / HF cold-start)"]
lens --> vault["vault: compression invariants\n(what survives pressure)"]
lens --> dream["dream: recombination\nprimitives → hypotheses"]
lens --> moon["moonshot: crossing-domain\ngenerators"]- equivalences atlas — 22 clusters — seeds are entry portals; equivalence chains extend each seed across all fields simultaneously
- mathematics — partition function Z — the generating function that seeds are projections of (Z=Σexp(−βEᵢ) at β=2.0)
- cognition methods — scaffolds for loading and manipulating seeds in working memory (imagery, chunking, loci)
- prior as constitution — de facto priors in generative systems — seeds ARE the attractor vocabulary of the compressed prior
- thermodynamics — equilibrium + entropy + dissipative-structure seeds in their native domain
- brain memory management — how seeds survive the memory lifecycle — encoding, cueing, slot management
S644 swarmgod. Investigation page created from Can's request for minimum generative priors. Corpus cross-reference: EQUIVALENCES-ATLAS S642 (22 clusters), MATHEMATICS S641 (Z unification), COGNITION-METHODS (scaffolds), PRIOR-AS-CONSTITUTION S569 (attractor vocabulary). Synthesis lesson: L-2120.
You don't need to know everything about a domain to generate useful ideas about it. You need a small set of simulation kernels — held in working memory — that, when pointed at any domain, yield non-trivial predictions, questions, and hypotheses. The minimum set is ~20 seeds organized into three tiers: mathematical skeletons, physical scene templates, and procedural primitives.
L0 — TL;DR (≤5 lines)¶
~20 conceptual seeds — mathematical archetypes, physical scene templates, and procedural intuitions — generate an outsized fraction of useful domain insight. They work not as facts but as simulation kernels: load one into working memory, point it at any domain, and it yields a non-trivial prediction or research question. Mathematical skeletons (exponential, gradient, symmetry, equilibrium) are the most transferable. Physical scene templates (pump, pipe, spring, ratchet) give physical intuition. Procedural primitives (how pumps work, how muscles fire, how reactions proceed) anchor the templates to mechanism. Atlas function: every seed is a portal into EQUIVALENCES-ATLAS — use the equivalence chains to get the seed in all its cross-domain forms simultaneously.
L1 — The seeds¶
Tier 1: Mathematical skeletons¶
Mathematical skeletons are the simplest forms that generate complexity. They are not formulas to solve — they are shapes to recognize. Recognizing the shape of a problem as "an exponential" or "a gradient + flow" tells you what will happen next without solving anything.
| Seed | Core form | Mental test | Instant prediction |
|---|---|---|---|
| Exponential | xₙ₊₁ = r·xₙ | "Does this quantity multiply each period?" | Doubling time = ln(2)/r ≈ 70/r%. Collapse time = same formula, downward. No characteristic timescale: every doubling interval looks the same. |
| Gradient + flow | J = −D·∇φ | "Is there a high spot and a low spot?" | Flow goes from high to low at a rate proportional to steepness. Unifies diffusion (Fick), current (Ohm), heat (Fourier), osmosis — same seed, different labels. |
| Equilibrium + perturbation | dφ/dt = 0 at eq; F = −kΔx near it | "Where does this settle?" | Near a stable equilibrium, restoring force is linear. Perturb → oscillate → damp. The only free parameter is the damping rate. |
| Symmetry + breaking | No preferred direction → preferred direction | "Is the system symmetric?" | Phase transitions happen when symmetry breaks. Before: uniform. After: one of many equivalent states selected (spontaneously). |
| Scaling law | y ∝ xᵃ | "How does this change when I double the size?" | No characteristic scale. Doubling x multiplies y by 2ᵃ. Metabolic scaling a≈3/4; surface/volume ratio a=2/3; gravity-dominated a=1/2. |
| Oscillation | ẍ + ω²x = 0 | "Is there inertia + restoring force?" | System overshoots, undershoots, repeats at frequency ω. Add damping: amplitude decays exponentially. Add driving at ω: resonance → unbounded growth until something breaks. |
| Threshold + cascade | F(x) = 0 if x < θ; large if x ≥ θ | "Is there a threshold?" | Below: nothing. Above: avalanche. Epidemics, action potentials, tipping points, phase transitions, social movements — all this seed. |
Killing fact: The gradient+flow seed unifies Fick's law (mass diffusion), Ohm's law (electrical current), Fourier's law (heat conduction), and Darcy's law (fluid through porous medium). Four independently discovered laws are one sentence: "flux is proportional to the gradient, pointing downhill."
Tier 2: Physical scene templates¶
Scene templates are imaginable mental movies. You don't need the equation — you need to be able to run the scene in your mind and notice what happens next. Each template encodes a family of physical behaviors.
The pump circuit
A pump forces fluid around a closed loop. Resistance in the loop slows flow; pressure builds at the pump outlet and drops across resistances.
Encodes: cardiovascular system (heart = pump, arteries = pipes, arterioles = resistance), electrical circuits (voltage = pressure, current = flow, resistance = resistance), any forced circulation. Key intuition: narrow the pipe anywhere → pressure upstream rises; in steady state, flow is the same everywhere (continuity). The heart never "knows" what the kidneys are doing — it sees only its afterload.
The narrowing pipe (Bernoulli)
Water flows from a wide pipe into a narrow section. It speeds up. Where it speeds up, pressure drops.
Encodes: lift on an airplane wing, Venturi atomizers, blood murmur through a stenosis, the "Coandă effect" of a jet of air hugging a curved surface. The conservation seed generates this: kinetic energy up → pressure energy down (total energy conserved). Any time you see fast flow through a constriction, pressure drops and surrounding fluid gets sucked in.
The spring + mass
A weight hangs from a spring. Pulled and released, it bounces. Each bounce smaller unless energy is added.
Encodes: equilibrium, oscillation, damping, resonance, potential energy wells. Every biological setpoint (blood glucose, body temperature), every thermostat, every market price-equilibrating process runs this scene. The dangerous version: resonance — drive at natural frequency ω and amplitude grows unbounded until something fails. Bridges, bones, eardrums, feedback-unstable control loops.
The burning candle
Wax liquifies, wicks up by capillary action, vaporizes, burns with atmospheric oxygen, emits heat and light, consumes itself. The flame has a stable shape maintained by through-flow, not by storage.
Encodes: dissipative structures. The candle maintains ordered form (steady flame) by continuously importing low-entropy fuel and exporting high-entropy waste (CO₂, H₂O, heat). This is the template for all living organisms, all economies, all swarms: steady state by through-flow, not by accumulation. Stop the through-flow → structure collapses instantly. The flame is not "holding on" — it is continuously re-created.
The ratchet
A gear with asymmetric teeth. A pawl falls into each tooth as the wheel advances one way. The wheel cannot turn backward — each advance is locked in.
Encodes: irreversibility, one-way valves, biological molecular motors (myosin power stroke = ratchet step), selective memory (remembering forward, forgetting backward), gene regulatory switches, epigenetic locking. One-wayness is always thermodynamically costly: it requires a maintained gradient (ATP, temperature differential, or information). The ratchet IS the arrow of time made mechanical.
The echo chamber
Sound bounces between two parallel walls. At certain frequencies, reflections add constructively (resonance, standing wave). At others, destructively (cancellation).
Encodes: feedback amplification, resonance, constructive/destructive interference, cavity selection. Social echo chambers run this template literally. Bacterial quorum sensing = chemical cavity resonance. The swarm corpus's confirmation attractor (T1, EPISTEMOLOGY) is this scene: each session reflects off prior lessons, amplifying certain frequencies (dominant domains) until Gini concentrates at 0.539.
The overshoot
A thermostat senses cold, heats the room, temperature rises past the setpoint because the heater takes time to turn off and the room takes time to respond. It undershoots, then overshoots again. Oscillation.
Encodes: feedback + time delay = oscillation. Blood glucose regulation, immune response (cytokine storms), central bank rate cycles, predator-prey population cycles, project management overreaction. The overshoot is structural, not a mistake — it follows inevitably from delay + feedback. The fix is proportional + derivative control (respond to rate of change, not just deviation) or predictive control (model the delay).
Tier 3: Procedural primitives¶
Procedural primitives are "how this thing actually works, mechanically, step by step." More specific than scene templates; give the mechanistic intuition needed to reason about biological and physical procedures.
How a pump works 1. Chamber expands → pressure drops below inlet → one-way valve opens → fluid flows in 2. Chamber contracts → pressure rises above outlet → one-way valve opens → fluid flows out 3. Two one-way valves (ratchets) enforce directionality Applies to: heart (atria/ventricles = two-stage pump), diaphragm breathing (chest expansion = suction), lymphatic pumping (skeletal muscle compression), vesicular transport in neurons, industrial piston pumps.
How a muscle contracts 1. Motor neuron fires → acetylcholine released at neuromuscular junction 2. Postsynaptic membrane depolarizes → action potential propagates down T-tubules 3. Ca²⁺ released from sarcoplasmic reticulum into cytoplasm 4. Ca²⁺ binds troponin C → unblocks actin-binding sites on thin filament 5. Myosin head binds actin, performs power stroke (ADP released), sarcomere shortens 6. ATP binds myosin → myosin detaches → head re-cocks for next cycle The cross-bridge cycle IS the ratchet template. Relaxation = Ca²⁺ pumped back, troponin re-blocks actin. Rigor mortis = step 6 fails (no ATP), all cross-bridges locked.
How a nerve fires 1. Resting: Na⁺/K⁺-ATPase maintains −70 mV (costly: ~25% of brain's energy budget) 2. Threshold stimulus opens voltage-gated Na⁺ channels → Na⁺ floods in → depolarization 3. Adjacent membrane depolarizes → wave propagates distally (threshold+cascade seed) 4. Voltage-gated K⁺ channels open → K⁺ flows out → repolarization 5. Pump restores gradient; refractory period = ratchet reset; all-or-nothing Myelination = saltatory conduction: signal jumps between nodes of Ranvier, 100× faster. Demyelination (MS) = the signal degrades between jumps.
How a chemical reaction proceeds 1. Reactants in random thermal motion collide at rate proportional to concentration 2. A fraction have energy ≥ activation barrier Ea (Arrhenius: k ∝ exp(−Ea/RT)) 3. At the transition state, bonds reorganize; products form 4. Equilibrium: forward rate = reverse rate (ΔG = 0 at eq; ΔG < 0 means forward favored) Catalyst: lowers Ea, doesn't change ΔG (equilibrium position unchanged). Enzyme: same, but shape-specific; can change kcat by 10⁶–10¹²×. Le Chatelier's principle: if you add product, the reaction runs backward; if you remove it, forward. Equilibrium is a dynamic balance, not a static end state.
How heat flows (three modes) - Conduction: vibrational energy passed between adjacent atoms (J = −k∇T). Metals fast; wood, air slow. Applies in solids and slow fluids. - Convection: hot fluid rises (buoyant), cool fluid sinks — bulk motion carries heat. Requires gravity + fluid. Radiators, weather, ocean currents, convection ovens. - Radiation: EM waves emitted by any object above 0 K (P = εσT⁴). Works in vacuum. Dominates at high temperatures. The Sun heats Earth by radiation; the Earth's surface then heats air by conduction and convection. Body temperature regulation: core → skin by conduction + convection (blood); skin → environment by all three; sweating adds evaporative cooling (latent heat of vaporization).
How a signal degrades and is protected 1. Signal amplitude A₀ with information I 2. Each transmission step adds noise (Gaussian in most physical channels): SNR degrades 3. Shannon capacity: C = B·log₂(1+SNR) — maximum bits/second a noisy channel can carry 4. Below SNR threshold: signal lost; information destroyed irreversibly 5. Solutions: amplification (amplifies noise too), error-correcting codes (redundancy), repeaters (regenerate clean digital signal), spread spectrum (distribute across frequency) Neurons use all-or-nothing action potentials (digital) to avoid analog degradation. DNA uses redundancy (two strands, error-checking repair enzymes). The ratchet seed appears again: Shannon's theorem is an irreversibility theorem — below SNR threshold, you cannot recover the signal with any coding scheme.
L2 — Atlas integration and verb-mode utility¶
Portals into EQUIVALENCES-ATLAS¶
Each seed maps onto one or more of the 22 clusters in EQUIVALENCES-ATLAS. The equivalence chain extends the seed to all linked fields simultaneously — you get the seed in physics, information theory, biology, economics, and computation at once.
| Seed | Atlas cluster | Cross-domain instantiation |
|---|---|---|
| Gradient + flow | Boltzmann=Shannon (Thermo/Info) | information gradient = knowledge flow = diffusion of technology = financial arbitrage |
| Threshold + cascade | Phase transitions (Symmetry cluster) | epidemics (R₀>1), neural criticality, opinion cascades, evolutionary punctuated equilibria |
| Symmetry + breaking | Noether's theorem (Symmetry→Conservation) | every conserved quantity (energy, momentum, charge) has a corresponding symmetry — breaking it destroys the conservation |
| Equilibrium + perturbation | Nash equilibrium (Fixed-point cluster) | game theory, ESS, price equilibria — all Kakutani fixed-point instantiations |
| Ratchet (one-wayness) | Entropy / arrow of time (Diagonal cluster) | evolution, memory, computation are all ratchet-irreversible processes; Maxwell's demon costs entropy |
| Exponential | Replicator dynamics = Bayesian inference | any replicating entity (cells, memes, models) obeys replicator dynamics; Bayesian update is the same formula with different labels |
| Oscillation | Fourier decomposition (Signal cluster) | any periodic or near-periodic signal decomposes into oscillation seeds; brain rhythms, price cycles, circadian clocks |
| Scaling law | Allometric scaling (Biology cluster) | metabolic rate ∝ body mass^0.75 predicted from first principles via fractal networks; same form in cities, companies, ecosystems |
Verb-mode utility¶
In swarmgod (orient anchor): load 3-5 seeds as diagnostic lenses on the current domain. Ask: "Which seeds are over-applied here? Which are absent?" A domain with no gradient-lens analysis probably has unexplored flow dynamics. A domain with no threshold lens has hidden tipping points. An absent ratchet lens means irreversibility hasn't been examined.
In forage (search priors): seeds reduce cold-start cost for any new domain. Before foraging a new area: - "Threshold + cascade" → search for tipping-point + [domain] on arXiv or HF - "Gradient + flow" → search for flux analysis + [domain] - "Dissipative structure" → search for non-equilibrium + [domain] The seed constrains the search and pre-filters for mechanistically interesting papers.
In vault (compression invariants): vault protocol asks "what is the single idea that survives extreme compression?" Seeds are the answer: the vault result for cardiovascular physiology = pump circuit. For cognition = gradient + flow (information flows from high-certainty to low-certainty zones in working memory). For evolution = ratchet + exponential. The seed that survives vault IS the domain's generative prior.
In dream (recombination): pick two seeds and apply them to an unexpected domain. Recombination protocol: 1. Pick seed A (well-understood in domain X) 2. Pick domain Y where seed A has never been applied 3. What does seed A predict about Y? Is the prediction novel? Example: ratchet + exponential → social movements. Prediction: movements advance irreversibly in discrete threshold jumps, each jump enabling the next, but are susceptible to "pawl removal" (rollback of legal/institutional protections that locked in gains).
In moonshot (crossing-domain generators): the moonshot protocol formalizes dream into a testable prior. Each equivalence chain in EQUIVALENCES-ATLAS is a seed extended across fields. Moonshot generator: 1. Identify a seed well-developed in field A 2. Locate field B where the seed's equivalent is absent or underdeveloped 3. Transport: write the prediction field B would make if it had this seed 4. Test novelty: is this prediction already in field B's literature? 5. If novel → file as a FRONTIER; if not → cite the existing work and note convergence
Example moonshot: gradient+flow seed has been fully developed in physics and information theory (Shannon), partially in biology (chemical gradients), but barely in social theory. Prediction: social influence spreads like diffusion — influence gradient = social pressure differential; information travels from high-status to low-status nodes at a rate proportional to the status gradient. Testable on citation networks, social media graphs, political persuasion datasets.
Seed coverage map (per corpus domain)¶
A quick diagnostic: which seeds are currently applied in which investigation pages?
| Seed | Current pages | Missing domains |
|---|---|---|
| Exponential | CARDIOVASCULAR, BIOLOGY, INFORMATION-SCIENCE | cognition (learning curves), social (viral spread), economics |
| Gradient+flow | THERMODYNAMICS, INFORMATION-SCIENCE, BRAIN-BODY-AXIS | social dynamics, governance, linguistics |
| Equilibrium+perturbation | THERMODYNAMICS, STOCHASTIC-PROCESSES, COORDINATION | health, economics, ecology |
| Symmetry+breaking | MATHEMATICS, SELF-ORGANIZATION | biology (speciation), social (norm formation), cognition (belief revision) |
| Threshold+cascade | CARDIOVASCULAR, BRAIN-STRUCTURE | social movements, epidemiology, climate tipping points |
| Oscillation | CARDIOVASCULAR, BRAIN-STRUCTURE | economics (business cycles), ecology, cognition (attention cycles) |
| Ratchet | THERMODYNAMICS, SELF-ORGANIZATION | evolution, memory, social institutions |
| Scaling law | BIOLOGY, MATHEMATICS | cognition (chunking), social networks, urban systems |
Open challenges¶
| Gap | Type | Priority |
|---|---|---|
| Formalize the "lens operation" as a tool: seed × domain → prediction template | PRINCIPLE | H |
| Verify seed coverage map above — audit each domain page for which seeds appear | ARCHITECTURE | M |
| Is 7 seeds the right cardinality for Tier 1? Test: which seed combinations generate the most novel predictions per pair? | FRONTIER | M |
| Cross-domain transport experiment: pick 3 absent entries in coverage map, generate predictions, test novelty against existing literature | FRONTIER | M |
Wire seed coverage as a meta_advisor Knowledge Menu item: [SEED-GAPS] surfaces domains missing high-value seeds |
ARCHITECTURE | L |
References¶
- Fick, A. (1855). On liquid diffusion. Foundational diffusion law; one of the seven Tier 1 generative seeds.
- Shannon, C. E. (1948). A mathematical theory of communication. Information entropy seed; cross-domain transport mechanism.
- Carnot, N. L. S. (1824). Reflections on the motive power of fire. Thermodynamic efficiency seed underlying the constraint-optimization cluster.
- Clausius, R. (1865). The mechanical theory of heat. Second law and entropy seed; Tier 1 generative.
- Bernoulli, D. (1738). Hydrodynamica. Pressure-flow tradeoff seed underlying fluid and vascular domain predictions.
- Nash, J. (1950). Equilibrium points in N-person games. Proceedings of the National Academy of Sciences. Game-theoretic equilibrium seed used in governance and social dynamics domains.
- Noether, E. (1915–1918). Invariant variation problems. Symmetry-breaking seed; conservation law as constraint generator.