How VialPost tests and what it will not do

Prices come from each vendor’s own public store data — the structured product endpoint their storefront already serves to its own pages — rather than from scraped HTML. Scraping a rendered page is where a guessed number creeps in; a structured endpoint returns the price the store itself is quoting.

robots.txt is fetched before anything else and obeyed, including any crawl delay it specifies. Several tracked vendors disallow the endpoint, and those vendors contribute no prices at all. That is a deliberate outcome: an empty cell is correct, an estimated price is not.

Only single vials of lyophilised powder are recorded. Multi-vial kits, capsule bottles, pre-mixed solutions, nasal sprays and multi-component blends are all excluded, because a per-milligram figure computed across those is not a comparison — a 60-capsule bottle and a 10 mg vial are different products with different economics. Wholesale, trade-only and sample listings are excluded too: they are real prices, but not ones a buyer can pay, and showing them would understate the floor. Any listing whose vial mass cannot be parsed unambiguously is dropped rather than guessed at. Every stored observation carries the source URL it came from and the timestamp it was read.

Observations are snapshots, not live quotes. A price shown here is what the store served at the timestamp beside it, and vendors change prices without notice.

A price tracker generates pages mechanically — one per compound, one per vendor, one per pair. Left ungated that produces a wall of near-empty pages carrying a heading, a disclaimer and no data. VialPost publishes a page to search engines only once it holds enough real, sourced data to answer the question it is named after. The rule is enforced in code, not by intention, and these are the numbers:

• A compound page is indexable once it carries price observations from at least 2 distinct vendors. Below that there is no comparison to make.
• A vendor page is indexable once it carries at least 3 price observations of its own.
• A comparison page exists only for two vendors that both list at least 5 of the same compounds, and only for the most-covered 40 such pairs. Every other pairing has no URL at all.

A page below the line is not deleted and it is not padded. It renders every observation it holds, says plainly that it is below the threshold, is served noindex, follow, and is left out of the sitemap until the data arrives. It re-enters both automatically on the next collection run that clears the threshold, and drops back out automatically if the data goes away.

Errors get corrected on the record where they appeared, with the date of the correction and a note saying what changed. The original figure stays visible. Nothing is silently edited.

A vendor that disputes a result is invited to say so, and a substantive dispute is published alongside the record. What a dispute cannot do is cause the result to be removed. Where a dispute raises a testable question, the answer is another blinded purchase and another assay, and both results are published.

A result withdrawn — because the chain of custody was broken, the assay was invalid, or the sample was misidentified — is marked as withdrawn with the reason. It is not deleted.

A result describes one lot, from one purchase, at one point in time. It is evidence about that lot. It is weaker evidence about the vendor, and weaker still about a vial you buy six months later.

Purity is not the only property that matters in a vial, and RP-HPLC does not measure sterility, endotoxin load, residual solvents, or correct identity on its own. A result on this site says what proportion of the material was the compound, and nothing beyond that.

The absence of a result for a vendor is not a finding about that vendor in either direction. It means no vial from that vendor has been through this pipeline yet.