The moat is the structured prompt — not the catalog UI

The claim

The defensibility is the artefact format — every prompt ships with typed metadata, evaluation cases, and a routing manifest — not the index page. Adjacent tools (notes apps, gist hosts, generic prompt libraries) cannot replicate this without giving up their core stance.

The evidence

Each artefact directory carries a metadata file (status, audience, triggers), a templated body in XML, a set of eval cases that gate promotion, and an optional skill-export manifest that emits the same prompt as an Anthropic skill. 38 of 52 prompts have cleared the gate and carry signed eval logs. A notes app stores text; a gist host stores files; neither can answer the question “is this prompt still passing its cases against the current model?” — the question that determines whether the prompt is safe to run today.

This shape compounds. Each new prompt adds a row to the routing layer (which model, which guardrail, which post-send verifier), and the routing layer is what keeps the catalog running as model versions change underneath it. Competitors building on a notes-app substrate would have to rebuild that layer from zero.

The comparable

Postman’s moat against curl was the same shape: anyone can send an HTTP request, but only Postman stored the request, the assertions, the environment, and the test history as one structured artefact. The artefact compounded; the UI was downstream. EVA carries the same wedge against the prompt.

What we ask for

A round sized to widen the artefact format — multi-model routing, per-prompt cost telemetry, and a shared registry — so the moat extends from one operator to many. The thesis and traction cards establish the why and the what; this card is the why-it-holds.

quadrantChart
  title Storage stance vs. LLM-native depth
  x-axis "Free-form text" --> "Structured artefact"
  y-axis "Generic" --> "LLM-native"
  quadrant-1 "Defensible"
  quadrant-2 "LLM playgrounds"
  quadrant-3 "Notes apps and gists"
  quadrant-4 "Prompt template libs"
  "Obsidian": [0.15, 0.2]
  "GitHub gists": [0.25, 0.3]
  "ChatGPT history": [0.4, 0.85]
  "LangSmith": [0.7, 0.9]
  "EVA": [0.92, 0.92]