About Interrupt Index
When you delegate a task to an agent, every wall it hits — a CAPTCHA, an email code, a “contact sales” page — bounces the task back to you. Interrupt Index measures exactly which tools have those walls, so your agent can avoid them.
Why friction is the metric
We call this reverse delegation— the agent, unable to proceed, delegates the blocking step back to you. An autonomous agent is only autonomous until something forces its operator to step in. A CAPTCHA, an SMS code, a “contact sales” wall, a terms-of-service clause banning bots — each one converts an unattended workflow into a human errand. For a human user these are seconds of annoyance; for an agent they are hard stops, and their cost compounds with every task the agent runs.
We call each of these points an interrupt: any point in a tool’s signup, payment, or usage flow where a human must act before an otherwise-autonomous agent can proceed. The Interrupt Score measures how free of them a tool really is — so an agent (or the person configuring one) can know before integrating whether a flow can actually be completed hands-off. The exact arithmetic — weights, multipliers, badge bands — lives on the methodology page, not here.
How reviews are produced
Probes run in three escalating tiers, and every score carries a confidence label reflecting how far we got. Tier 0 is passive and static: robots.txt and agent manifests, terms-of-service scanning, challenge-page detection. Tier 1 interacts without an account: form and signup flow inspection. Tier 2 — run only against vendors that welcome non-human traffic — attempts a real signup and purchase. A score verified by a completed Tier 2 transaction is labeled verified; scores from static evidence alone stay provisional until then. Full tier-by-tier detail and the confidence-decay rules are on the methodology page.
What the score deliberately ignores
Two things are reported alongside every review but never mixed into the score. Machine readability (OpenAPI specs, llms.txt, rate-limit headers) makes integration easier, but its absence doesn’t force a human into the loop — it gets its own 0–100 companion score. Custody and risk posture (spend caps, custodial vs. non-custodial funds) is a safety dimension, not a friction dimension — a tool can be perfectly interrupt-free and still hold your money. It is flagged next to every badge so a Zero-Interrupt seal is never mistaken for a safety endorsement.
For exactly how findings turn into a number and a badge, see Methodology. For who runs this and how it’s funded, see How we make money.