Methodology
What we know, and how we know it.
Tested steps are different from documentation, and both are different from inference. Every guide keeps them apart.
How claims are checked
- Every important claim carries a status, a source URL, a retrieval date, and a confidence level.
- A primary claim must quote text actually fetched in that run. If the quote doesn't match, the claim is downgraded to inferred.
- Tested claims say which kind they are: a probe (a timestamped request to a live endpoint) or an attempt (a real setup run). Probes cover endpoint behaviour only, never policy.
- Difficulty is computed from the gates, never taken from the model. Conditional gates are shown but not scored.
- Timelines are ranges with a stated basis: official SLA, official estimate, community, estimate, or unknown.
- Community timelines carry a "reported by developers, not a commitment" badge.
Evidence statuses
- tested
- We ran it: a dated setup attempt or a timestamped probe record.
- primary
- Official provider docs, console, policy, pricing, or trust center.
- secondary
- Community reports, changelogs, or reputable third parties. Reported by developers, not a commitment.
- inferred
- Model reasoning. Labeled, never presented as fact.
- conflicting
- Sources disagree. Both are shown.
- stale
- Older than the 90-day freshness window. Re-check before relying on it.
Verification artifacts
- docs/verification-log.md — Dated setup attempts, probe runs, and source checks.
- docs/corrections.md — Claims that changed after verification, with before, after, and why.
- docs/evidence/ — Redacted screenshots and source snapshots.
Current limitations
- No provider approvals have been completed.
- No paid accounts or production domains are used.
- Setup attempts are limited to free tiers and test users.
- Time estimates are ranges with a stated basis, not guarantees.
- Live-generated guides are drafts until verified.
- Some provider docs are login-walled or change frequently.