Skip to content
@coherence-energy-labs

Coherence Energy Labs

How systems hold together: provable software and applied systems, built on coherence energy.
Coherence Energy Labs - how systems hold together. This banner is the solution of the field equation (D·L + κ²I)τ = s, computed in exact integer arithmetic.

banner re-derives

The banner is computed, not drawn - the solution of our field equation in exact integer arithmetic, rendered by a program in this repository. CI re-derives it on every push; one byte of drift fails the build. Its receipt is committed beside it and printed on the artifact itself.


Provable software and applied systems, built on coherence energy: one measurable way to know how well any system holds together, and to make it hold together better.


Website  ·  Live demos  ·  Contact


The thesis

Physics, life, mind, and technology get studied in separate rooms. They share a load-bearing property: parts working together as one. We call that property coherence, and we treat it as an engineering quantity - something you can measure, compute, and build against.

Coherence energy is the cost of holding a system's order together against noise:

E_coh = k_B·T · D_KL( ρ_system ‖ ρ_disorder )

Coherence itself is a field, governed by one screened equation we solve everywhere:

(D·L + κ²I) τ = s

The same equation schedules instructions in our compiler, forecasts natural hazards, plans spacecraft trajectories, and drives attention in our AI systems. It also drew the banner at the top of this page. One discipline, bare metal to cosmology.

The standard

A claim is only as strong as the artifact behind it.

Modern software asks for trust. Ours is engineered so that trust is never required. Every result travels a chain of custody in which each link is checkable by a stranger:

 source          certified          deterministic         receipt            stranger           a claim
 one program ──► compile      ──►  execution       ──►  program + input ──► verification  ──►  you can
                 proven equal      0 ULP across          + cryptographic    zero lines of       cite
                 over ALL inputs   CPU·GPU·WASM·browser  hash               our code needed
  • Every result ships as program + input + receipt. Anyone can re-execute it and get the same bytes.
  • Every guarantee has a gate that can say no. Our CI gates are mutation-tested - we deliberately inject the bugs they claim to catch and prove they go red. A gate that cannot fail is not a gate.
  • Every failure is published, not buried. We keep a falsification library of our own dead claims, dated and preserved. Negative results are load-bearing.
  • Everything fails closed. When evidence is missing, the answer is "no" - in our compilers, our proofs, and our AI.

By the numbers

0 ULP drift between CPU, GPU, WebAssembly, and browser execution of the same program - floating point included. Exactness is a contract, not an aspiration. verified-public
2⁶⁴ⁿ inputs covered by each compiler-optimization equivalence proof. Machine-checked over every possible input, not tested on a sample. private-development
7 compilation backends from one language - VM, C, native x86/ARM, LLVM, WebAssembly, GPU, and embedded scripting - proven to agree. private-development
29 cryptographic primitives - classical and post-quantum - each verified byte-exact against independent NIST/RFC test vectors, interoperable with reference implementations. private-development
15 communication-protocol specifications formally model-checked with an active attacker in the model. private-development
~9 ms to generate a million-point proof round on GPU, bit-identical to the reference prover. Verification runs client-side, in a browser. private-development
1 field equation, solved everywhere: compiler scheduling, hazard forecasting, trajectory design, machine cognition - and this page's own banner. verified-public

Every number above links to its entry in our public evidence registry - verified-public means you can check it yourself today; private-development means we attest to it and say plainly that you currently cannot. We hold our marketing to the same standard as our software: no claim without a labeled artifact.

What we build

Coherence Language A first-principles programming language, compiler, and runtime. Certified compilation - the toolchain cannot miscompile, because optimized and original are proven equivalent over all inputs before anything ships. The same source runs bit-identically from bare metal to your browser tab.
LumOne A coherence-native AI whose every decision is re-runnable and receipted. Fifteen-plus sound solver lanes - exact arithmetic, theorem provers, computer algebra - answer before a model generates a single token. And when it cannot know, it signs a provable certificate of ignorance instead of bluffing. In active development.
A.C.E A cognitive architecture built entirely in Coherence Language - a mind whose reasoning is auditable end to end. Its memory is Merkle-sealed, its forecasts are registered before the evidence arrives, and its self-modification sits behind supervised gates with measured keep-or-rollback. In active development.
One Link Private communication with no servers. Post-quantum hybrid key exchange, forward-secret ratcheting, and a formally model-checked protocol core - privacy that holds against the next era of adversaries, not just this one.
Obsign AI you can prove. Accountable computation for media and models: deterministic, bit-exact, with re-executable provenance assertions that today's content-credential standards define but do not deliver.
Applied systems The same discipline pointed at the physical world: natural-hazard forecasting scored honestly against named baselines, hardware-rooted attestation proven on real silicon, verifiable measurement from sensor to signature, and trajectory design with proof-carrying routes.

Don't trust us - re-run it

Our live demos hand you the verifier:

  • The Gauntlet - honest and forged computation proofs, checked entirely in your browser. The forgeries fail on your hardware, by your arithmetic - not our word.
  • Re-executable figures - every published chart is a program, its input, and its receipt. Your browser re-derives the result and confirms it, offline, from a local file, with no server to trust. A figure you can't fake is a figure you can cite.
  • CPU == GPU, live - the same computation on two different substrates, agreeing to the last bit, in front of you.

No accounts. No telemetry. No network required. The proof either verifies on your machine or it doesn't.

Find the missing unit

Two energy ledgers sit in this repository - 4,096 integer transfers across 64 accounts each, generated by the same program. In one of them, a single entry was altered by one unit - one part in billions of total flow. No chart will show it. No statistic will find it fast. The books find it instantly, because conservation is a sum, not a vibe:

git clone https://github.com/coherence-energy-labs/.github && cd .github
python tools/conservation_challenge.py --audit assets/ledger-a.json   # one says BOOKS CLOSE, drift = 0
python tools/conservation_challenge.py --audit assets/ledger-b.json   # the other names the broken accounts

The forged ledger doesn't just fail - the audit names the exact accounts where conservation breaks, drift ±1, at both ends of the altered transfer. This is the discipline our applied systems run on: exact-integer books that close to zero, or say precisely why they don't. This page practices what it preaches. The banner is a re-executable figure with its receipt committed, guarded by a public gate that re-derives it from source on every push and on a weekly schedule. Clone this repo and run python tools/render_banner.py --check - you'll re-create the org's face, byte for byte, on your own machine.

How we work

Artifacts before adjectives. Nothing is called done, fast, or safe without a machine-checkable artifact behind the word. First failure is fuel. A red gate is never reverted around - it is root-caused and the system comes back stronger. Falsify your own work first. Every major claim ships with the experiments that tried to kill it. Preserve coherence. We build systems that do not hide how they work - that can be inspected, that can prove what happened, and that hold together instead of falling apart.



One field. One framework.

The same equation runs everything here, from the compiler to the cosmos - and one standard of evidence judges it: re-run it yourself, or read exactly why you can't yet.


coherenceenergylabs.com  ·  info@coherenceenergylabs.com


Pinned Loading

  1. evidence evidence Public

    Every public claim resolves to evidence or an honest label. Machine-readable claim registry.

  2. hazardpulse hazardpulse Public

    Three natural-hazard prediction systems from one field equation. Prospective forecasts frozen and hash-receipted before outcomes (Ed25519-signed from 2026-07-31), scored honestly against named base…

    Python

  3. Ariadne Ariadne Public

    Validated, Python-native toolkit for CR3BP / cislunar mission design and trans-Neptunian-object discovery. Standard gravity, real DE440 ephemerides, cross-validated against NASA GMAT.

    Python

  4. oneq-distance-closures oneq-distance-closures Public

    Certified exact distances for six PBB codes: signed certificates, inline witnesses, one-command verifier, and a forgery-laden stranger-verification challenge

    Python

Repositories

Showing 10 of 10 repositories

Top languages

Loading…

Most used topics

Loading…