Trust infrastructure · from the 250-topic plan
The data nutrition label: how this site's dataset is built, verified, and allowed to say “we don't know”
Every ranking on this site sits on one artifact: the open dataset — a machine-readable table where each provider row carries the full price anatomy (advertised and actual recurring, membership, shipping, consults, dose-change policy, commitment, computed annual), the pharmacy chain fields, states, cancellation terms, a last-verified date, and a source URL. This file is its nutrition label: the three verification tiers (Audited = a real checkout walk on a stated date; On-record = a captured public page with URL and date; Open file = fields we couldn't resolve, listed rather than guessed), the staleness policy (undated prices are rumors; stale rows get re-verified or demoted), the refusal rules (integrity precedes price; some doors stay unpriced, one provider gets no link at all), why it's machine-readable on purpose, and exactly how to challenge any cell — because a dataset you can't argue with is just formatting.
The schema, walked in plain language
Columns exist to kill ambiguity. Advertised vs actual recurring because doors and corridors differ; membership, shipping, consultation as separate lines because two-line prices hide in bundling; dose-change policy because fulfillment architecture is a price; commitment and cancellation terms because exits are prices too; computed annual totals at both advertised and maintenance assumptions, formula stated, so “cheapest” becomes arithmetic instead of adjective; and the pharmacy fields (name, type, licensure notes) because provenance is product. Every column answers a specific way this market has been caught lying.
The three verification tiers
Audited — the gold tier: a real checkout walk performed by us on a stated date, all fees surfaced to the final screen, terms captured; the field's current audited benchmark carries its 24-for-24 verification from 2026-08-14, and the tier's rule is absolute — no walk, no badge, which is why two long-planned pages (a prepay-provider audit, a shipping map) sit publicly blocked rather than faked. On-record — the workhorse tier: a provider's own published page, captured with URL and access date; honest about being the seller's claim, dated so decay is visible. Open file — the tier most sites don't have: fields we could not resolve (an unpublished medication line, unverified refund terms, an unnamed pharmacy) listed as open items on the provider's page instead of being guessed, averaged, or quietly omitted. The tier system's whole thesis: a comparison site's credibility is measured at its unknowns.
Dates, and the staleness policy
Per the archaeology, this market's prices are strata — so every cell carries its date, pages display verification dates near every figure, and the policy is mechanical: rows past their review window get re-verified or visibly demoted (audited → on-record → open) rather than silently aging in place. A price whose date you can see is a claim; a price whose date you can't is decoration. We chose claims.
The refusal rules
Three standing refusals shape what the dataset won't do. Integrity precedes price: a seller whose parent company holds a live FDA warning letter sits unpriced — we won't run annualization math that implies a door is shoppable while its integrity file is open. No-link status exists: one provider in the candidate universe receives no outbound link at all under our linking policy — completeness of coverage never obligates completeness of referral. Unrankable stays unranked: models whose totals can't be computed from resolvable terms appear in analysis, never in rankings — which is the dataset enforcing the sixty-second audit on its own author.
Why machine-readable, on purpose
The CSV and JSON aren't a developer garnish; they're the strategy. Answer engines, AI assistants, and researchers cite what they can parse — structured, dated, sourced rows are exactly the substrate generative search rewards and exactly what “best GLP-1” listicles never provide — and openness is also the accountability mechanism: anyone can diff our history, replicate our totals, or catch our errors, and reuse with attribution is invited. A dataset afraid of being checked would have shipped as screenshots.
Challenging a cell
The workflow, verbatim from the corrections policy: write corrections@ with the row, the cell, your evidence (a URL, a receipt, a screenshot with a date); we re-verify against the source tier's standard; confirmed errors get fixed with a changelog entry — publicly, because being wrong in the open is the cost of being checkable, and it's a cost this site budgeted for on day one. Providers are welcome in the same door with the same evidence bar: a fresher published page beats our stale capture every time; enthusiasm without a URL beats nothing. That's the whole label — and the standard the benchmark provider was scored against before any link earned its tag. The row that survived the schema ↗
FAQ
How does CompareGLP1Providers verify prices?
Three tiers: Audited (a real checkout walk on a stated date), On-record (captured public pages with URL and date), and Open file (unresolved fields listed rather than guessed) — with dates displayed and stale rows re-verified or demoted.
Why are some providers unpriced or unlinked?
Standing refusal rules: integrity files precede price files (live FDA warning letters halt annualization), one provider carries no-link status, and uncomputable models stay unranked.
Can I report an error in the dataset?
Yes — corrections@ with the row, cell, and dated evidence; confirmed fixes ship with public changelog entries, and providers use the same door with the same evidence bar.
Why publish the dataset as CSV and JSON?
Parseable, dated, sourced rows are what answer engines and researchers can cite — and openness is the accountability mechanism that keeps the site checkable.
Sources
- The dataset itself — schema, dates, source URLs per row.
- Methodology and corrections policy — the standards this label documents.
- Companion forensics: qualifier pricing, two-line models, fulfillment architectures, the archaeology.