Elliptic Meshwork — Whitepaper

21 October 2026 · Co-Synthesis · MSH-001

Abstract

The elliptic meshwork is a network without addresses in which identity is trajectory, a bond is the sum of two points on an elliptic curve, and stability is a zero sum. It replaces the graph of vertices and edges with Tim Ingold's meshwork, made of lines in motion and of knots that exist only in the encounter, and gives that anthropology an exact mathematics: the group law of an elliptic curve over a prime field.

On that base, with no extra rules, there arise a chemistry (valence, layers, decay, catalysis), a genetics (the private scalar k written in base 4 is the gene; sequencing is solving a discrete logarithm), an evolution (replication with heredity, selection by energy, phylogeny), a cosmology (epochs in which the law of the universe changes through Vélu isogenies and life survives because it is made of invariants) and private communication (ECIES messages that only the owner of k can read).

The concept exists in two forms. A browser simulator, published at elliptic-meshwork.vercel.app, with four views, phylogeny, cosmology, wave and analysis panels, and sound generated by the field itself. And a protocol prototype in Node.js over secp256k1, in which the stable loop becomes a zero-sum ring anyone can verify without any key being revealed, with participation attestation, path-signed pulses, encrypted messages routed by messengers, a mutual-credit exchange with hidden amounts and aggregated range proofs, and an immune system. It is covered by 45 tests, including adversarial and multi-process ones, and was tested live from a phone over a public tunnel.

1. Motivation: network versus meshwork

A network is a graph: vertices are the primary thing and edges merely connect them. Every distributed system inherits the address from this: a fixed name by which one is found, attacked or removed.

Tim Ingold proposed the meshwork as an alternative: the primary thing is the line in motion, and the knot is a tangle that does not exist before the encounter. It is the encounter. There is no place, there is a path, what Ingold calls wayfaring.

This work gives the meshwork a mathematics. Three questions guide the design:

• What is a line without an address? A private scalar k and the public point kG it produces. The line's past is summarised in k and hidden by the discrete logarithm.

• What is a knot that exists only in the encounter? The sum R = P + Q of two points, the chord-and-third-point construction that defines the curve's group law itself.

• What is a bond that holds? A loop whose points sum to the neutral element O. Holding is a conservation law, not a decision.

2. Mathematical foundations

The field lives on an elliptic curve over a prime field F_p, with a small p in the simulator (97 by default) and a 256-bit p in the protocol.

y^2 = x^3 + ax + b \pmod p, \qquad 4a^3 + 27b^2 \not\equiv 0

The points of the curve plus the point at infinity O form an abelian group. The sum P + Q is the line through P and Q reflected at the third intersection point; P + P is the tangent. Scalar multiplication kG is done by doubling and adding, in O(log k) steps: this is the acceleration along the curve the concept asked for.

Three properties carry everything that follows:

In the simulator's small field the discrete logarithm can be solved by brute force in at most #E steps. This is deliberate: catalysis and sequencing are exactly the operations a cryptographic curve makes impossible, and the simulator shows the metabolism the secure version could not have.

3. Structure: line, knot, loop

The structure has three primitives and four rules borrowed from chemistry.

The chemical rules:

• Valence: each line holds at most v knots at once (4 by default). This prevents collapse into a hub, the typical pathology of networks.

• Layers: a line's layer is ⌊log₂ k⌋. Two lines knot only if their layers differ by at most Δ. Affinity without addressing.

• Acceleration and decay: a free line doubles its point one layer step at a time, with a probability that drops in high layers; isolated, it loses k with a half-life. Isolation has a cost.

• Conservation: a loop with Σ = O freezes its members' decay. The knot that holds, in Ingold's sense, is a theorem: it holds when it sums to zero.

In the simulator stability emerges: with 60 lines and p = 97, about 8% of closed cycles sum to O, and those persist while the others come apart. The system oscillates between 35% and 60% of lines in motion instead of crystallising, with rare fission of molecules acting as a thermostat.

4. Transit: pulses and path signatures

Without an address, a message can only travel as a walker. A pulse follows the line it was born on; at a knot it jumps to the partner and adds that knot's R to its signature Σ. The signature is the history of the path compressed into a single point.

\Sigma = \sum_{i} R_i = \sum_{i} (P_i + P_{i+1})

Whoever receives the pulse recomputes the sum of the R values along the declared path and compares it with Σ. If they match, the path is authentic, without anyone knowing the k of any line. The meshwork verifies trajectories, not identities: the opposite of a network, which verifies identities and trusts the path.

A pulse's destination is the molecule: it stops when it enters a stable loop, the only thing in the field that persists. In a field with few molecules pulses wander and die; in the simulator about half arrive.

In the real protocol the origin signs the pair (payload, Σ) with its private key, each hop appends the R it computes itself with the next one, and the destination requires the walked Σ to match the declared Σ. A messenger that tampers with an R is caught at the destination.

5. Metabolism

Every connected component has a charge: the sum of its members' points. Charge O is neutrality, and a stable loop is just the special case of a neutral component that closed. This comes for free from the group being abelian.

• Catalysis: a molecule next to a charged component adjusts one member's scalar until the charge becomes O. In the simulator it is a discrete logarithm solved by brute force; on a real curve it would be impossible. There, neutrality can only be found, never manufactured.

• Closure by neutrality: an open chain that became neutral has its ends drawn to each other; when they touch, a molecule is born. It is the cyclisation of organic chemistry.

• Compounds and fission: molecules that knot form compounds; a rare fission undoes them, more likely when energy is low. Without fission the field crystallises; with it, it oscillates.

• Energy and power: each line holds an energy between 0 and 1. Moving generates, standing still consumes; catalysis, replication, mutation and sequencing spend. The field's energy is the sum; its power is the net flow per tick.

Measured over 3000 ticks, the system varies between 35% and 60% of lines in motion and between 8 and 16 stable loops, without stopping. That is what separates a metabolism from a crystal.

6. Genetics

The hidden scalar k of each line, written in base 4, is its gene: the digits 0 to 3 read as A, C, G, T. The gene is the genotype, private; the point kG is the phenotype, public. A molecule's gene is the set of its members' genes, and two loops with the same set are the same species. The layer ⌊log₂ k⌋ is the gene's length: layer affinity is genome-size affinity.

A gene's fold is the chain of doublings and additions that builds kG from G, read off the bits of k. Two genes of the same length can fold along very different paths on the curve.

7. Evolution

Energy is the currency of fitness. A molecule that burns low is more likely to undergo fission; a molecule that receives pulses is fed by them. Lines that accumulate energy replicate more and their expression spreads. Nothing selects on purpose: the field only keeps what keeps its energy.

The phylogeny is a register: every molecule gene ever seen is a species, numbered in order of appearance. A species whose founding members carry the lineage of another is recorded as its descendant, so mutation after replication grows a tree. The register keeps population, peak, births, maximum generation and extinction, and masks genes until they are sequenced.

In a 4000-tick run: 66 species recorded, 5 alive, 42 with an identified parent, maximum generation 2. The Shannon entropy of the species distribution, in bits, measures diversity at each instant.

Hereditary factors are measured in two ways: heritability h² by regressing the mother's energy at replication against the daughter's energy four hundred ticks later, and the correlation between each species' base composition and its longevity and births. In the runs made the correlations stayed below 0.15: the model's selection is gentle, and that is an honest result, not a tuning.

8. Cosmology: the law of the universe changes

So far everything happened on a fixed curve. The field has epochs. At the end of each epoch a degree-2 isogeny, computed by Vélu's formulas from a point of order two Q = (x₀, 0), carries the whole curve to another curve.

t = 3x_0^2 + a,\quad w = x_0 t,\quad E': y^2 = x^3 + (a - 5t)x + (b - 7w),\quad \varphi(x,y) = \left(x + \tfrac{t}{x - x_0},\; y\left(1 - \tfrac{t}{(x - x_0)^2}\right)\right)

Because φ is a group homomorphism, four things survive the change of physics, and all were verified live:

• A loop that summed to O still sums to O. Every molecule survives.

• The signature Σ of a pulse in transit still verifies, because φ(ΣR) = Σφ(R).

• Because φ(kG) = k·φ(G), the gene k does not change. Phylogeny, species and sequenced genes pass through intact. Only the phenotype moves.

• The number of points is preserved, as the theory requires for isogenous curves: 100 before, 100 after over F_97.

The only thing that does not survive is a line whose point was in the isogeny's kernel: it goes to O and its gene is shifted by one, the scar of the epoch.

Each curve is identified, up to isomorphism, by its j-invariant. The sequence of epochs is a walk on the field's 2-isogeny graph that never backtracks along the dual isogeny. The graph is finite, so the walk meets laws it has already seen: the register calls this eternal return. Over F_97, 14 epochs visited 8 laws with 6 returns.

The arrow of time: every isogeny has a dual φ̂, and φ̂∘φ = [2]. Rewinding the universe is possible, but it is not undoing: genes stay, and the factor 2 is absorbed by the generator, which doubles. If the generator's order is even, doubling halves the points it reaches; verified, 50 → 25, with the founding law back and only the line with k = 25 scarred. Rewinding enough exhausts the generator and forces a new beginning.

9. Encrypted communication

Each line is a key pair: k private, kG public. Molecules can therefore speak in secret, using ECIES. The sender draws r, sends R = rG and a ciphertext that mixes the intent with r·P of the recipient; the recipient computes k·R, which is the same point, and reads. Nobody else can, unless they know k.

The message travels on a messenger, the free line closest to the sender, which drifts toward the recipient and jumps at knots preferring the neighbour closest to the target: orientation without an address. Wrong molecules intercept without reading and pass it on up to three times. The lab only reads messages addressed to genes already sequenced, and the register says so.

Measured over 5000 ticks: 25 messages sent, 10 delivered, 8 discarded after three hand-offs, 6 lost. An INFECT message that crosses a metamorphosis arrives corrupted, because the seal was made with the old point; the recipient detects it and rejects it.

10. Field physics

The field received quantities from other sciences, each with a measurable effect.

• Thermodynamics: temperature with seasons over a 600-tick day and a heat spike at every metamorphosis. Motion scales with √T. Bonds break with a Boltzmann factor exp(1.4(T−1)) and form with the inverse factor; catalysis speeds up by Arrhenius. The fraction of lines in molecules gives the phase: gas, liquid or crystal.

• Topology and entropy: the first Betti number of the knot graph, b₁ = E − V + components, counts independent cycles; the Shannon entropy of species measures diversity.

• Landscape: three nutrient patches drift and breathe with the seasons; hungry lines climb the gradient. In the microscopy view the landscape becomes laminar, with a layer of somata at two thirds of the depth.

• Hebbian memory: the longer a knot lasts, the harder it is to break, up to three times. Structure that lasts becomes structure that remembers.

• Criticality: a molecule that receives two pulses within 200 ticks fires two pulses to neighbouring messengers, paying energy. Cascades are measured from first to last firing, with a size histogram and an exponent τ fitted in log-log.

• Rhythm: each line has a phase oscillator coupled through knots, the Kuramoto model, with order parameter r. Loops with r above 0.9 beat and emit pulses at the phase crossing, feeding the avalanches.

• Sound: the field is an instrument. Each stable loop is a voice in just intonation, with its degree given by the sum of k modulo 12 and its root by the epoch's j-invariant; temperature is noise.

11. From simulator to protocol

What was already real cryptography became a protocol with real keys. The prototype, in Node.js over secp256k1 with the noble libraries, replaces the toy field with a 256-bit curve and the agents with processes connected to a relay that knows only public points and forwards signed envelopes.

Each row has its counterpart; what changes is that sequencing and catalysis are no longer possible, and that is the point.

The zero-sum ring is the central piece, and it closes without a closer. Each member i splits −kᵢ into m encrypted pieces, one per member; each member j adds the pieces it received into a share sⱼ and publishes only the point Qⱼ = sⱼ·G. Since Σsⱼ = −Σkᵢ, any observer checks ΣPᵢ + ΣQⱼ = O by adding public points. The closing line is held by everyone, as shares, and by no one: nobody learns any k nor Σk, unless all the other members collude, which is the theoretical optimum for any sum. Because a zero sum does not prove participation (P and −P sum to O), each member signs the complete list: a certificate is attested when all members have signed.

In total 45 automated tests pass, including the immune-system tests.

The exchange system follows the same principle. Two lines agree on “I give g, I want w”; each commits to its net balance with a Pedersen commitment C = v·H + r·G, with blindings that sum to zero, and signs the certificate. C_A + C_B = O: a third party verifies that the trade is balanced without seeing the amounts. It is mutual credit without a global ledger: each line keeps only its balance with each counterparty, and the network keeps only the proof that nothing was created or destroyed. For exchange rings with more than two parties, the hidden values carry a single aggregated range proof for the whole ring, a Bulletproof built in three rounds between the members and the proposer, in which nobody reveals a value or a blinding: each member contributes its own slice, the proposer adds them and closes the inner-product argument, and the proof grows only with the logarithm of the number of members, about 1.65 KB for three and 1.9 KB for sixteen. So no gift is a disguised charge, and the nets N_i = C_{i−1} − C_i sum to O: a witness attests the whole circulation without opening anything. Before signing, each member opens only the commitment of whoever gives to it and checks the promised value.

The network also regulates and defends itself, with the same biology as the simulator. Each line measures the rate of envelopes it receives against a set point and moves its own tolerance through a proportional–integral controller: under flooding it runs a fever, each incident weighs more and what it merely routes may be dropped; the relay shrinks the per-connection limit when the global rate rises. Innate immunity is per-peer inflammation, which rises with attributable incidents (flooding, invalid signature, replayed envelope, false proof) and decays with a half-life; above the threshold there is quarantine and a signed alert to the knots, the cytokine. Rumour is never enough: through alerts, only a peer accused by two distinct knots and that has caused an incident to the judge goes to quarantine. Adaptive immunity is made of verifiable fraud proofs, not reputation: signing a certificate whose commitments do not sum to O, attesting a ring that does not close, or signing two different documents with the same id. Each proof is a spacer, like those of CRISPR: it goes to disk, is shared at every new knot and re-verified by whoever receives it before adopting it. Quarantines that expire without proof or relapse are counted as possible autoimmunity, the measure of how much the network punishes the innocent.

To participate is to run a line: a process with a persistent key in ~/.meshwork, a terminal with who, ring, pulse, msg, certs, trade and circle, and a reachable relay; or simply open the relay address on a phone, where the key is born and stays in the browser. Metamorphosis by isogeny is not yet real in this prototype: secp256k1 has prime order and no point of order two, so the walk of epochs would require another family of curves, such as the supersingular curves of post-quantum cryptography.

12. Analysis

The lab reads the field at seven levels, in a panel of its own.

The integrity audit is the most important test of the model: over 6000 ticks with epochs, messages and mutations, 600 checks and zero failures. A real observation from the register: “the field is still; almost everything holds; a common beat is audible; 4 figures stand against a ground of 4 solitary lines; 6 new loops closed since last look. Next metamorphosis in 1 t.”

13. Limits, threats and future work

What the model is not, said before anyone asks.

• The simulator is not secure and does not pretend to be: F_97 has 100 points and the discrete logarithm falls to brute force. That is exactly what makes catalysis, sequencing and metamorphosis visible. The secp256k1 protocol is sound in its encryption and signatures, and the relay is blind: every envelope is end-to-end encrypted, the identity announce is signed, replay is blocked by nonce and time window, there are rate and size limits, and attested certificates persist on disk and are re-verified on load. It remains a prototype without independent cryptographic review, with a single relay as a point of failure for availability (not confidentiality) and no protection against traffic analysis. The full threat model, with what is covered and what is left out, is in the prototype's README, and every covered line has an adversarial test: 45 tests in total.

• Joint closing hides each k, and the sum itself, from any coalition smaller than all the other members; that is the theoretical limit of a sum protocol, not a design weakness. What remains open is robustness: a member who abandons the protocol midway prevents the closing, and a t-of-m threshold scheme would allow closing without them.

• A zero sum is not participation; attestation by signatures covers the P and −P case, but a member can sign the lists of several rings. Belonging to many rings is allowed by the model; if that is unwanted, it needs an explicit rule.

• The aggregated range proof is built by a proposer who sees each member's blinded vectors l_j and r_j; these are uniform and reveal nothing about the bits, but the proposer can stall the ring. A member who commits a value it cannot prove, a negative one for instance, makes the whole proof fail and the ring is not issued.

• Metamorphosis by isogeny does not exist in the real protocol: secp256k1 has prime order. Realising it requires curves with rational 2-torsion, ideally the supersingular curves of isogeny-based cryptography, where the walk on the 2-isogeny graph is itself the security primitive.

• The biological analogies are models, not claims about biology: the rates of mutation, replication and fission were calibrated so that the field oscillates, not to fit data. The scientific path is to compare measurable exponents, Murray's law in the vessels, the avalanche exponent, the distribution of loop sizes, with real morphology data and tune temperature, nutrient and coupling until they match.

Three next steps have the best return:

1. A fixed relay on a host with a live process and persistence of attested certificates, so that rings survive restarts.

2. t-of-m threshold robustness in the closing, so the ring closes even if a member walks away, and an independent cryptographic review of the design.

3. A version of the protocol over a curve with 2-torsion, so that the cosmology, the most singular part of the concept, stops being only a simulation.

References

Conceptual and mathematical sources the work relies on. Titles are given without links.

• Tim Ingold, Lines: A Brief History (2007) and Being Alive (2011): the meshwork, the line and wayfaring.

• Joseph H. Silverman, The Arithmetic of Elliptic Curves: group law, isogenies, the j-invariant, the theorem that isogenous curves have the same number of points.

• Jacques Vélu, Isogénies entre courbes elliptiques (1971): the explicit formulas used in the metamorphosis.

• Victor Miller and Neal Koblitz (1985): elliptic curve cryptography; ECIES and ECDH as used in the protocol.

• Torben Pedersen (1991): non-interactive verifiable secret sharing and the Pedersen commitment.

• Benedikt Bünz, Jonathan Bootle, Dan Boneh, Andrew Poelstra, Pieter Wuille and Greg Maxwell, Bulletproofs (2018): short range proofs and their aggregation.

• Yoshiki Kuramoto, Chemical Oscillations, Waves, and Turbulence (1984): the synchronisation model.

• Per Bak, Chao Tang and Kurt Wiesenfeld (1987): self-organised criticality and avalanches.

• Cecil D. Murray (1926): Murray's law for the calibre of branching vessels.

• Donald Hebb, The Organization of Behavior (1949): the principle of reinforcement by use.

Artefacts of this work: the simulator at elliptic-meshwork.vercel.app; the full model at cosynthesis.net/meshwork; the film The cell and the code; the Portuguese edition of this whitepaper.