For investors · the asymmetry

The deterministic engine layer
below AI-era software.

AI Act binds 2025–26. Frontier models commoditise. LLM code volume goes 100×. Audit becomes the bottleneck. Whoever owns the engine that proves AI-written code safe by construction owns the next durable layer. Built in Europe.

50
ADRs · public architecture
every decision documented · every claim falsifiable
7,400+
Tests across 223 files
Stryker-mutation-tested · 75% break threshold
12
Cryptographic commitment chains
L1 → L12 · Bitcoin-anchored · regulator-verifiable
3
Live reference deployments
paysafe · cliniclens · civicgate

Across the regulated-EU operators we've spoken with, the pattern is consistent: the same AI mandate and the same stuck pilot: ship an LLM feature inside a regulated product without killing audit, liability, or compliance. The vendor landscape is partial. Vanta and Drata audit the software around the LLM. Credo and Fiddler audit the model's behaviour. LangChain orchestrates calls. Nobody constrains the LLM's output surface by construction. Nobody carries the call into a regulator-verifiable record.

The substrate does. One architecture, from which audit, factory-scale generation, regulatory compliance, capability transfer, and runtime-AI containment all flow. Not five products — five faces of one engine. The technical surface is Phase R / v1.2 closeout. The strategic surface is the asymmetry below.

Europe owns lithography through ASML. The next chokepoint is software — the layer the AI-tooling category will eventually have to live on top of. The asymmetry, with numbers →

Three forces · converging this decade

Each force in isolation is a known story. The conjunction is the category.

01

LLM code volume · 100×

Cursor, Copilot, Claude Code, Devin, Cognition. AI-generated code grows from a developer convenience to the default production input. Volume scales faster than human review capacity. Review becomes the bottleneck.

Inference

By 2027 most production code shipped at Fortune-500 scale will originate from an LLM. The audit pipeline that worked at human-throughput cannot.

02

Frontier-model commoditisation

Claude Sonnet 4.5, GPT-5, Gemini 3 — capability differences shrink quarter by quarter. The model becomes commodity input. The differentiator moves down-stack to the layer that turns commodity input into safe output.

Inference

Margin compresses for model providers. Margin expands for the deterministic engine they feed into. ASML, not Intel.

03

Regulatory rail · 2025–26

EU AI Act phased application. DORA effective Jan 2025. NIS 2 in force. PSD3 finalising. Each binds AI in regulated products to a verification standard the current vendor landscape doesn't meet.

Inference

Enforcement creates demand for verification. Verification at scale requires architecture, not consulting. Architecture is the moat.

Why Europe · the ASML play

Own one critical layer everyone else has to depend on. Applied to software, before anyone else has built it.

The lithography parallel

ASML is a Dutch company that makes the EUV lithography machines used to manufacture every leading-edge semiconductor in the world. Every chip Intel, TSMC, Samsung, and SMIC ships passes through an ASML machine. One country, one company, one layer everyone else has to depend on.

ASML was not built by accident. It was built by patient capital, institutional research alignment (TU Eindhoven, Imec), and a willingness to play decade-long games. The same institutional muscles that built ASML can build the software equivalent.

The regulatory lever

Europe's regulatory apparatus — historically described as a drag on its tech sector — becomes the asymmetric advantage in this category. The AI Act, GDPR, DORA, NIS 2, MDR, PSD3 aren't bugs to route around. They are the strongest demand signal for verifiable architecture on Earth.

The vendor landscape that grew around the US compliance surface (Vanta, Drata, Credo, OneTrust) treats regulation as a checklist. The substrate treats it as a structural input. The architecture inverts what was a disadvantage.

Why durable · three compounding loops

The moat is not a feature. It is the operational shape.

01

Empirical compounding

Every operator pilot produces a corpus contribution. Every corpus entry strengthens the next product's spec generators. Each cycle of the loop reduces the next pilot's time-to-production.

Falsifiable

Time-to-production per pilot is measured + falsifiable, not assumed. The substrate audits itself; trajectories are reproducible.

02

Regulatory compounding

Each new regulator citation reduces the cost of the next operator's compliance conversation. The substrate becomes a Schelling point for AI-in-regulated-product conversations as enforcement scales.

Falsifiable

3 live reference deployments today. Each new one cited makes the next one easier to defend.

03

Capability transfer compounding

The meta-generator authors substrate components from operator-described intents. Each operator who hits a missing primitive contributes a substrate-component intent. The substrate grows from operator engagement, not just engineering hires.

Falsifiable

The substrate's meta-generator has emitted 2 substrate modules currently in runtime use. Scaling becomes operator-paced, not headcount-paced.

Traction · calibrated, not promotional

What's real today. What's still aspiration.

Built · verifiable

  • 50 ADRs documenting every architectural decision · public
  • 7,400+ tests across 223 files · tsc clean · Stryker-mutation-tested
  • 12 cryptographic commitment chains under one Ed25519 anchor
  • Direct OpenTimestamps Bitcoin anchoring of L12 HEAD
  • Cross-language Python port · 8 fixture-driven byte-identity tests
  • 3 live regulator-verifiable reference deployments
  • 25 products the engine has built end-to-end
  • 2 substrate-generated substrate components in runtime use

Not yet · honest gaps

  • Zero paying operators today. Three reference deployments are substrate-operated; first paying-operator pilot is the next gate.
  • Zero regulator citations today. Conversations open with several EU bodies; first regulatory-document citation is the second gate.
  • Cost-per-pilot trajectory unmeasured. The architecture supports compounding cost reduction across pilots; empirical measurement requires the first 3–5 operators.
  • HSM integration is roadmap. v1.2 ships in-memory key signing; v1.3 native HSM injection. Operator-side workaround documented.
  • Team is small. Hiring after first paying operator, not before.

Three honest paths to revenue

Each path monetises a different face of the engine. The architecture supports all three.

01Active · first cohort scoping

Operator pilots

Fixed-scope 90-day engagements with regulated operators. Bank, hospital, govtech, insurer. The substrate ships their LLM feature into production with a complete cryptographic record.

Monetises

The audit face. Sized for quarterly platform budgets.

02Conversations open

Regulator partnerships

Non-exclusive cooperation with regulatory bodies. We provide reference architecture + verifier + reproducibility briefings; the regulator cites the substrate in guidance documents.

Monetises

The rail face. Compounds via citation.

03Architecture ready · pilots required first

Capability transfer

Operator-pilot graduates take ownership of an emitted product line. The substrate continues evolving the underlying engine. Operators pay for substrate access at platform-budget scale.

Monetises

The factory face. Recurring, anchored in delivered code.

Capital model · milestones, not headcount

Capital tied to verifiable gates. Each one falsifiable.

Pre-seed (now)

Architecture + 3 reference deployments + first operator conversation

In progress · Phase R complete

Seed

1st paying operator pilot signed · architecture battle-tested in real production

Next gate

Series A

3 paying operators · 1 regulator citation · measurable cost reduction per pilot

Architecture-ready · operator-pilot-bound

Series B+

10+ operator pilots · 3+ regulator citations · capability transfer cohort active

Conditional · path is structural, not assumed

Three doors

One substrate, five consequences. Read the artifacts.