The words, defined plainly.
Definitions we use across alt.qa, and what they do not mean.
Agent readiness
Whether AI agents acting for people can discover, understand and complete tasks on a website. A static score checks files; an assurance run checks behaviour.
Agent Readiness Assurance
Running real AI agents from multiple vendors (and behavioural personas) through a company’s own funnels before release, with evidence for each failure.
Agentic commerce
Shopping, booking or buying carried out by an AI agent on a person’s behalf, through a website UI or a commerce protocol.
Alt Matrix
alt.qa’s grid of personas (rows) × agents (columns) on a funnel; each cell is pass/fail with the blocking step and evidence.
Cell
One run: one persona × one agent × one goal × one target. The unit alt.qa meters and prices.
Behavioural approximation
A profile that mimics how a class of consumer agent behaves (pacing, confirmation style) using a vendor model under alt.qa’s identity, never a spoof of the real product.
Web Bot Auth
An IETF draft for bots and agents to sign HTTP requests (HTTP Message Signatures) so sites can verify who they are.
Confirmed finding
A failure reproduced in at least two cells or by deterministic replay. Only Confirmed findings become tickets.
Accessibility tree
The structure browsers expose to assistive technology: roles, names and states of elements. Many agents read it to decide what to click.
Stop before pay
The rule that test runs stop at the payment step unless in a sandbox with test cards.
Test Authorization
A signed record stating which properties, environments and actions an alt.qa agent may use, when, and until when.
Precision (finding)
Confirmed findings a human agrees with, divided by findings flagged. Published per persona with sample size and confidence interval.
MCP (Model Context Protocol)
An open protocol that lets AI assistants call tools and read data. alt.qa’s MCP server lets coding agents run a matrix and fetch failure bundles.
UCP, ACP and AP2
Commerce protocols that let agents check out without the web UI. They reduce UI failures at payment but not discovery and qualification on the site.
llms.txt
A proposed file describing a site for language models. Ahrefs found 97% of published llms.txt files never receive a request.