Home / Methodology

How we work

Everything on this site is generated from tracked, dated, reproducible data — and every method below states what it can't see.
SOURCES vendor pricing pages App Store · storefronts FETCH scheduled scripts raw snapshots kept EXTRACT sandboxed model: page text → JSON only VALIDATE schema-checked, confidence scored STORE every fact dated + source URL REBUILD → STATIC PAGES every page regenerated from the database on each refresh refresh loop
ranking score = App Store rating × 10 + log₁₀ reviews × 6 + confirmed features × 0.75 + free tier × 3

Pricing

Collected automatically from each vendor's public pricing page on a schedule. Every figure carries the capture date. Non-USD prices are converted at the ECB reference rate and marked ≈. Raw page snapshots are archived for every data point. Limit: usage-based pricing is summarized from published tiers; negotiated/enterprise pricing is invisible to us.

Ratings & merchant sentiment

Ratings and review counts come from the Shopify App Store's published structured data. Sentiment themes are synthesized from each app's most recent reviews — paraphrased, never quoted, counted by theme. Limit: recent reviews overweight the current product version; sample sizes are shown.

Feature matrices

Extracted from vendor App Store listings against a fixed per-category taxonomy. Unstated features are marked unknown ("—"), never guessed. Limit: listings understate features; vendor docs may confirm more.

Brand stack detection

We scan brand storefronts' public HTML for vendor script and CDN signatures. Presence of a vendor's code is strong evidence it's in use; absence proves nothing (server-side or inner-page tools escape detection). Confidence is shown per detection.

Rankings

Category rankings use a disclosed formula — App Store rating ×10 + log₁₀(review count) ×6 + confirmed features ×0.75 + free tier ×3 — computed from the data above. Products without ratings are listed unranked rather than scored.

Independence

No vendor pays for placement or ranking. We currently have no affiliate relationships; if that changes, affected links will be disclosed on the page. Verdicts marked "auto-generated" are computed from data; editorial verdicts are signed.

Corrections and contact

Every number here is measured, and measurement can be wrong. If something about your product or storefront is inaccurate, email hello@dtcproof.com with the evidence and we will correct it and note the date. Detection reflects what a public page loaded on the date shown — server-side and checkout-only tools are invisible to us, so absence is never a claim.

Data last refreshed 2026-09-08. Machine-readable summary: /llm-info.