Site / Methodology

Methodology

How AI Tool Graph turns official sources, pricing pages, documentation, and updates into public registry facts.

/methodology

What the method produces

AI Tool Graph turns source material into structured registry facts. The method is designed for repeatable fields such as identity, official links, capabilities, pricing routes, evidence, freshness, and review state.

Relationship and context fields are modeled separately and only become public when source-backed and ready.

The goal is not to decide which tool a reader should buy. The goal is to make factual records inspectable enough that downstream editorial products, internal research, and public readers can see where a claim came from.

identity

Canonical records

Tool, company, suite, platform, model, and project identities are separated when the source material supports that split.

access

Capability and delivery facts

Capabilities, input and output types, delivery modes, and supported user surfaces are tracked as separate fields.

pricing

Pricing routes

Commercial access paths are recorded as routes with their own source URL, checked state, owner, and review state.

evidence

Source-backed context

Evidence rows keep source type, used-for labels, freshness, and review state visible instead of burying them in prose.

Source priority

Primary sources are preferred for facts that affect identity, access, pricing, or publication state.

Official sources are the default authority. Third-party pages can help discovery, but they are weak evidence for canonical facts and should not become primary pricing evidence.

Level 1	Official pricing pages, documentation, changelogs, product pages, company announcements, terms, and security or trust pages.
Level 2	Official help centers, status pages, official blogs, public release notes, and vendor-owned support materials.
Level 3	Marketplace listings, app-store listings, developer mirrors, and official partner listings when they add useful access context.
Level 4	Third-party references, media summaries, community notes, or search snippets. These are treated as discovery aids, not primary evidence.

Freshness rhythm

Profile freshness and pricing freshness are separate. A stable official identity page can remain current while a pricing route needs a new check, and a stale pricing route should not automatically make every profile field stale.

Records are rechecked when an official source changes, a source update is reported, a record is prepared for public publication, or a downstream projection needs a stronger source state.

Freshness fields should describe what was checked, not imply a broader review than the source supports.
Pricing checks should remain attached to pricing routes or direct plans rather than a generic page timestamp.
Unclear or stale fields should stay visible as incomplete instead of being converted into confident copy.

Internal review state

Source strength is kept in the facts model and review workflows. Public pages should expose source links, freshness, and review-needed states without turning source review into product quality marks.

official

Official and recent

A field is backed by a strong official source, the claim is unambiguous, and the checked state is current enough for the public record.

partial

Official but partial

The source is official, but the field may have limits, plan conditions, regional differences, or missing detail.

review

Ambiguous or stale

The source trail is incomplete, old, indirect, or likely to need human review before publication.

Source update loop

Source updates are part of the method. A useful report points to the affected page, names the exact field or sentence, and includes the strongest official source available.

A source review can update a value, mark a field for review, mark a field stale, add a missing source, or keep a claim out of public facts until it is reviewable.

Preferred evidence: official pricing, documentation, product, trust, changelog, or announcement pages.
Useful context: affected route, field name, source URL, and why the current record is wrong or incomplete.
Rejected updates: unsourced promotional claims, broad opinions, private account screenshots, or requests to change ordering for commercial reasons.

What the method refuses

The registry does not turn subjective judgments into canonical facts. If a page needs buyer guidance, subjective ordering, or editorial recommendation, that belongs outside the shared facts registry.

No invented prices, star-style labels, verification dates, official source counts, or review labels.
No chosen-for-you labels, numeric quality marks, or popularity ordering as registry facts.
No conversion of context-specific workflow signals into global tool facts.
No affiliate-first calls to action as evidence or methodology.