The Framework
Five tests, six candidate moats, three verdicts.
The five tests
A candidate moat earns the name only if it can answer yes to all five.
- Publication. Could a competitor with full access to your published artifacts and a frontier model reproduce equivalent capability? If yes, the artifact was never the moat. Also called the Ingestion Test.
- Replication cost. What would it cost a credible competitor, including a foundation model provider moving up the stack, to build the same advantage from scratch? If the honest answer is months, it probably is not durable.
- Compounding. Does the advantage strengthen with each engagement, or merely persist? Persistent advantages decay. Compounding ones widen, or at least hold the gap open.
- Portability. If a senior practitioner walked out tomorrow, would the advantage leave with them, or is it embedded in workflows, data, relationships, and accountability they cannot carry in their head?
- Liability. When something goes wrong, is your firm the one on the hook? In regulated, high-stakes work this is a real moat.
The six candidate moats
| Moat | Where it lands |
|---|---|
| Trajectory advantage and knowledge-capture compounding | Most durable when properly built |
| Trust, accountability, and liability absorption | Most durable when properly built |
| Process experience and embodied judgment | Durable when institutionalized; fragile otherwise |
| Ecosystem position and distribution | Durable when institutionalized; fragile otherwise |
| Proprietary data flywheels in high-friction domains | Conditional or transitional |
| Governance and human–AI orchestration capability | Conditional or transitional |
Trajectory advantage is the gravitational center: three of the other five either depend on it or converge with it. And the most durable moats combine a structural barrier with a compounding mechanism.
Three verdicts
Every claimed moat leaves the filter as Durable, Conditional, or Fragile. Read the ranking as a mirror, not a scoreboard: pick the one or two moats you would actually claim in a pitch, and ask whether each is built, or merely asserted.
The tools that run these tests on your own situation live in the Living Edition.