Methodology
Titer measures the asset before it brokers the deal. Every listing carries a published Titer Score and, where applicable, a Proof of Function run — both computed the same way every time, described here in full.
The Titer Score (rubric v1)
An integer from 0–100 across five weighted components. It is computed by a person against this rubric and stored with a short justification — never inferred, never estimated, and never shown as a full score when only partial information is available. A listing with no score renders Unscored, in grey.
Grades: A ≥80 · B 65–79 · C 50–64 · D <50.
Proof of Function
For AI agents and APIs, we run the asset against a fixed request set in a sandbox and publish the result: success rate, p50/p95 latency, and cost per task at the listing's current prices, plus an error taxonomy. The run is timestamped and versioned by harness — we link to this page, not to raw logs, because the harness and request set are what make the number comparable across listings.
The Model Risk block
Every AI agent listing, and any digital business with inference in its cost of goods sold, discloses its primary and fallback model provider, inference cost as a percentage of revenue, gross margin if token prices doubled, and what the moat actually is — a prompt, proprietary data, a workflow, distribution, or an integration. No incumbent marketplace prices this risk; it's the thing every AI-asset buyer is actually afraid of, so we make the seller answer it in plain English before a buyer has to ask.