The Canadarm2 robotic manipulator above Earth in low orbit

Independent research and engineering house

58.97° N / 05.73° E

From the abyss
to deep space.

AI and software, tested against physical machines. Evidence decides what gets built—a venture is the last step, not the first.

Inspect current work
Systems on record
05Research / Systems / Operations / Ventures
Asset case, manual QA
59Manufacturable AI assets
Starlink Wall tests
08Node 20 / 22 / 24
Live open-source systems
02Knowledge graphs / Link diagnostics

01 / Operating model

One house / full system

From first principles
to fielded systems.

The work crosses disciplines, then survives deployment. What proves useful keeps moving.

  1. Research

    Name the decision.

    Map the evidence, mechanism, unknowns, and first failure before choosing what to build.

    EVIDENCE / MECHANISM / INVALIDATOR

  2. Systems

    Integrate the machine.

    Electrical, mechanical, software, AI, and control designed as one system.

    EE / ME / SW / AI / CONTROL

  3. Operations

    Make it survive contact.

    Safety, supply, deployment, cost, and lifecycle carried past the prototype.

    FIELD / SAFETY / SUPPLY / COST

  4. Ventures

    Carry proof into the world.

    A focused business formed only when a working system has earned a commercial test.

    PRODUCT / CUSTOMER / ECONOMICS

What proves useful keeps moving

Frame 02 / Sea level

Open water, camera at the surface

Sea state / 0 m

02 / Research in progress

PCB Compiler / Updated 12 July 2026

Compiling plain-language intent into physical copper.

AI models write circuits that look correct but fail in the physical world. We are building a deterministic pipeline to compile plain-language intent into verified, manufacturable PCBs. If we succeed, hardware development becomes as fast, cheap, and agile as software.

The mission
Text-to-manufacturable hardware
Roadblock
Datasheet qualifiers
Next test
12 V solenoid bench

Finding 01 / The datasheet footnote

Our agent selected a low-pressure valve, missing a page-18 footnote that requires a 0.5 bar differential pressure to open. Before drawing copper, we are teaching the compiler to catch these qualifiers.

Read the engineering log

03 / Current systems

Technology / Working example / Claim boundary

Current systems and working examples.

Each row separates the technical work from the prototype, venture, or open build used to test it.

  1. System / 01

    Manufacturable AI assets

    REFERENCE MASK / SAFE AREA / PRINT BLEED / ASSET QA

    World Patches

    59-ASSET CASE / MANUAL QA

  2. System / 02

    Vision teleoperation

    HAND LANDMARKS / EXPIRING INTENT / SUPERVISOR / LOCAL SAFETY BOUNDARY

    Teleoperation control bench

    TESTED BENCH / SIMULATED LINK

  3. System / 03

    Knowledge graph systems

    DOCUMENTS / ENTITIES / COMMUNITIES / GRAPH INSPECTION

    GraphRAG Workbench

    OPEN SOURCE / V2.0.0

  4. System / 04

    Terminal link diagnostics

    NETWORK PATH / ORBITAL GEOMETRY / PASS WINDOWS / DISH REACHABILITY

    Starlink Wall

    OPEN SOURCE / V0.2.0 RELEASE

  5. System / 05

    Terminal space-weather instrumentation

    OVATION / SOLAR WIND / DARKNESS / STABLE ANSI COMPOSITOR

    Aurora Wall

    OPEN SOURCE / PRE-V1 / 23 TESTS

04 / Standard of proof

Claim classes / Source order / Currency

Claims earn
their place.

Every public statement on this site carries a class: what supports it, and what would prove it wrong.

  1. Class 01 / Verified fact

    Directly supported by a current primary or authoritative source.

  2. Class 02 / Calculation

    Reproducible from stated inputs and method.

  3. Class 03 / Observed result

    Produced by a dated test with conditions and artifact.

  4. Class 04 / Inference

    Reasoned from named evidence; the reasoning is visible.

  5. Class 05 / Hypothesis

    Testable and not yet established.

  6. Class 06 / Judgment

    A recommendation from stated evidence and constraints.

An inference does not inherit the confidence of its sources

An SR-71 crossing a twilight sky with both afterburners lit

05 / Technical decisions

Evidence before commitment

What needs
to be true?

Send the decision and the evidence already in hand. The next step is the smallest test that could change the answer.

Email Lyon Industries
  1. 01 / Decision

    Which decision needs evidence?

  2. 02 / Evidence

    What is already known?

  3. 03 / Failure

    What breaks first?

  4. 04 / Test

    What can be checked next?

Frame 03 / Propulsion

SR-71 · Photograph NASA

J58 / Full afterburner / Twilight