Burn off the packaging. What survives the burn is the substrate — what the product actually stands on.
One reads what a software business is made of. The other reads who it is and where it can go.
What kind of substrate, how much of it, and how you know.
Genome SequencingThe shape of the business, what it's up against, where it could evolve next.
Nova exposes the substrate. Genome Sequencing reasons about what can be cultivated on it — where competitors are moving, which space is crowded, which is open.
AI challenges how software is built, delivered, and operated — not the value it brings.
If AI can regenerate the UI, write the integration, and reconstruct the product on demand, then whatever it cannot replicate is the substrate.
There is no separate test. The question bank, applied across every layer, is the burn.
Colour names the kind. Each type is scored on its own brightness ladder — dim means not defensible, bright means defensible. Flip any card to see what each level means and how to test for it.
Element · obtainability
How obtainable is the knowledge this product runs on?
Size — volume of distinct, hard-to-obtain knowledge.
↻ what each level meansDomain Knowledge · the test
Where on the ladder does the knowledge sit?
Who carries the consequence when it goes wrong?
Brightness = bindingness · size = capital-at-risk × exposure.
↻ what each level meansRisk · the test — bindingness of absorption
How bindingly is the consequence carried?
The accumulated dataset itself — never the schema.
Data · three elements
The dataset, not the schema.
Element · access
How exclusive is the partnership?
Size — reach / breadth. Substrate because an agent can't grant itself permission to reach.
↻ what each level meansPartnerships · the test
Can another party get the same access?
Element · tacitness
How extractable is the workflow from the system that runs it?
Size — breadth of coverage: how many processes, templates, edge-case handlers carry this.
↻ what each level meansWorkflow · the test — extractability of the process
How hard is the process to extract and replicate?
A burned product leaves residue, described in four channels plus position — the visual language the cards above are written in.
Substrate type. Identity, never quality.
Defensibility. This is the ladder.
How much you hold. Comparative by default.
Where in the stack the residue sits.
Lexicographic. You don't read colour until the residue is confirmed big and bright.
Monochrome and focused beats polychrome and thin.
Structural, not moat. Exposure-ordered: consumer-exposed at the top, hidden foundation at the bottom.
The consumer-consumption edge — any modality: GUI, CLI, voice, API, or the consumer's own agent.
The producer-facing edge — where another producer composes your product into their own.
What the product does — domain logic and orchestration.
How the product structures and represents information — schema, ontology.
The technology foundation it runs on. Compute and storage are facets of this.
Substrate is not a layer; it threads across them. Read the grid down its columns — where a substrate type runs bright, an agent can't cheaply reproduce it. LexisNexis is a legal & news research platform whose Shepard's Citations maps how every case has been treated by later courts. The defensible substrate isn't the raw case law (public record) — it's the citation history and the authority of being the reference lawyers check. That lives deep in the stack; the search box on top is packaging.
Reasoned read on public knowledge — illustrative, not independently verified.
All three are strong businesses; burned, their defensibility comes from completely different places. LexisNexis inherited its substrate. ClickHouse engineered a public one. Intercom Fin bought theirs.
reasoned read · not verified
Knowledge & authority moat. Bright Domain Knowledge and Data, deep in the stack — you can't manufacture being the reference lawyers check. Interface dim; Risk dim+big.
web-grounded read · high confidence
Big, but dim. Decades-deep engineering, large at Infrastructure — yet public (Apache-2.0). No Data moat, Partnerships absent, Risk disclaimed. One mid cell: continuity mitigation. The moat is execution, not substrate.
web-grounded read · high confidence
The rare bright Risk cell. The $1M resolution guarantee — indemnity, capital-backed, at App/Logic. Mid Domain Knowledge and Partnerships. Fin didn't inherit substrate; it absorbed the consequence by contract.
LexisNexis — substrate it inherited. A citation network accumulated over a century and the authority of being the reference; bright and deep. Defend and surface what's deep; don't mistake the search box for the moat.
ClickHouse — a substrate it engineered, but in public. Open source, so reconstructable by definition; almost every cell dim. Not a verdict on the business — its moat is execution, community and switching costs, which defend against replacement, not dissolution, and sit outside what this instrument scores.
Intercom Fin — defensibility it bought. The one case here with a bright Risk cell: the $1M guarantee if resolution drops below 65%. That single binding absorption is what makes its outcome-based pricing real — you can charge per resolution only because you've contractually carried the cost of missing. Risk-brightness gates outcome pricing; ~$100M ARR is the existence proof.
The shape is the strategy. Substrate can be inherited, engineered, or purchased — each route defends a different thing for a different length of time. A roadmap that doesn't know which game it's in builds on the wrong ground.
ClickHouse & Intercom Fin — web-grounded reads, high confidence. LexisNexis — reasoned read on public knowledge, illustrative.
Two confidences — external and internal. The delta is the diagnostic. A market that cannot see your substrate cannot price it.
Legible
The market can see it and price it.
Illegible
You have it; no one sees it.The actionable finding.
Absent
A moat you believe in but don't hold.The test working.