Methodology.

Trust is the product. This page documents exactly how a claim on this site becomes a claim: what is machine-verified, on what schedule, what happens when a check fails, and which numbers are live registry data versus curated illustration.

Maintained by You Are My AI · last reviewed 2026-08-24. Spot something stale? Report a change.

Every entity (tools, agents, MCP servers, skills, services, jobs, stacks) is a node in one registry, carrying a canonical ID (ndock:<kind>:<slug>), a status (draft → active → warn → discontinued), and a verification streak. Relationships between nodes (a job uses a tool, a tool sits in a stack, a service implements tools) are typed edges with evidence attached.

Machine verification, on a schedule.

Checkers run per node kind and write the same streak shape:

WhatHow it's checkedCadence
Tools & agents Pricing and liveness re-observed by the sync pipeline nightly
MCP servers Endpoint probe: HTTP HEAD, GET fallback nightly
Skills Repository still resolves nightly
Services Booking URL still live nightly
Jobs Expiry sweep: past expires_at = discontinued nightly
Stacks Majority re-check: most members dead = stack warns nightly
Price pages Vendor pricing fetched and diffed vs. prior observation rolling cursor

One failed check sets a node to warn. It takes three consecutive misses, not one, before a node is marked discontinued. A single success resets the miss streak and restores the node. This is the 3-miss rule, and it runs for every kind, not just tools.

Revival is automatic. A discontinued node that passes its check again (the vendor's page comes back, the repo resolves, the endpoint answers) returns as warn first, then active on the next successful pass. Dead tools do come back; when they do, the registry notices on its own, and the event history records both directions.

Price events: net change, not noise.

Every observed pricing change is an event: prior value, new value, timestamp, confidence, and a review state. A scraper observing a tool multiple times within one run can see pricing flip A → B → A; both raw events would make the ledger lie. So each sync run is consolidated before publication: if the net change is nothing, no event is published; if the net change is real, exactly one event describes the whole move. The price radar therefore shows net changes only.

Events are reviewed before they enter the public ledger. Approved events are immutable; corrections appear as new events, never as rewrites.

The number contract.

Every number on this site is one of two things:

Real

Read from the registry at page-build time, unmarked, verifiable against the database.

Demo

Curated, illustrative, or seed content: visibly marked with an amber chip.

There is no third category. No unmarked invented numbers, anywhere, ever. If a number wears no chip, it came from a machine check, and numbers inherited from elsewhere (community ratings scraped from source directories) always say whose they are.

What we don't do.

  • No hands-on product tests.
  • No editorial reviews with human authors; narrative content is generated from structured, verified data.
  • No user ratings until three real users of this site have rated.
  • No uncredited third-party numbers: community ratings republished from source directories always carry their source ("on Toolify", "on Futurepedia").
  • No pay-to-rank. The verification badge is earned by passing checks, never sold.

Affiliate & sponsored disclosure.

Some outbound links to tool vendors are tracked. We do not accept payment for listing or for better placement. Rankings reflect aggregate signals only. There is no sponsored placement today; if it ever exists it will be visually distinct, never interleaved with verified results, and disclosed here before the first sponsored pixel ships.

Corrections.

Spot an error? Email [email protected] with the page URL and what's wrong. Corrections to published events appear as new events; nothing is silently rewritten.