Every published number traces to a claim ID
Behind the site is an internal research base. Its unit is the claim: one statement, the source it came from, the exact wording that source used, the date the source was captured, and an identifier. A number reaches a published page only by way of a claim. A number that has no claim behind it does not get published, and the sentence that wanted it gets rewritten or dropped.
Calculated figures are held to the same standard with an extra requirement. A derived statistic has to show its formula, the inputs it used, the date it was computed, and it has to survive a validity check before it exists. A derivation that only works by assuming a number nobody has measured is not published as an estimate. It is not published.
Sources are captured as dated snapshots and quoted verbatim
Sources are not read and summarized from memory. Each one is fetched and stored as a dated snapshot, and the load-bearing wording on any page has to appear word for word inside the snapshot it cites. That check is mechanical, so paraphrase drift and half-remembered figures fail it rather than pass quietly.
A capture is also classified before it is judged. A fetch that returned a navigation menu, a cookie banner, a login wall or a missing page is recorded as exactly that, and never described as thin or weak content, because those are different findings with different consequences.
Competitor pages are used to find questions, never as a source
Ranking pages are read to learn what buyers are asking, which sources exist, and what the market already covers. They are evidence of intent and a route to primary material. They are never quoted, paraphrased, or treated as authority for a factual statement. When a competitor page cites a study, the study gets tracked down and captured, and the study is what gets cited.
The same discipline applies in reverse, and it is the mistake this project has made most often. Writing that competitors do not cover something is a claim about the captured set, and it needs the same evidence as any other claim. Before a sentence here says none, only, or every, the captured pages are re-read at the moment of writing.
Vendor figures are attributed, never restated as fact
Software vendors publish a lot of numbers about themselves. They appear here only as that vendor's own statement on a named page on a named date, and never as this site's assertion about how intake works or what a product delivers. Availability documented on an official page means the vendor documents it, which is a weaker statement than the vendor having it work in a firm, and the pages say which one they mean. Research published by a vendor that is itself compared on the same page is treated as that vendor speaking, not as independent research.
Community evidence is voice of customer, not measurement
Practitioner discussion is collected for what people say the job feels like: the language, the complaints, the workarounds, the questions they ask before buying. It can never establish how common something is, what a normal result looks like, or what caused what. No prevalence, no benchmark, no causality comes out of a discussion thread.
Two further constraints apply to that material. Scope belongs to the speaker, so a market, a firm size or a figure is only attached to the person who actually said it, never inferred from the forum, the thread topic or a neighboring comment. And threads about buying software contain promotional participants, including vendors, affiliates and accounts running lead generation. Those are identified in the capture and nothing is drawn from them. Several people agreeing inside one reply chain is agreement, not independent corroboration, and it is not described as convergence.
Legal and regulatory questions are held, not hedged
Intake sits next to questions about professional conduct rules. Those questions are real, buyers ask them, and this site does not answer them.
Published rule text may be quoted verbatim from a captured official source with its citation and its capture date, because reporting what a published rule says is research. What does not get written, in any form including a hedged one, is a statement that a firm, product, service, arrangement or described practice is permitted, prohibited, required, risky or compliant. A sentence a reader could act on as legal guidance does not ship here regardless of how it is qualified, and the honest version of that rule is to say so plainly rather than to publish the guidance with a disclaimer under it.
Where model rule text has been collected, it is model text. This project has compared it against one state's adopted text and assumes nothing about any other jurisdiction. No qualified reviewer has read any of the legal source material held here. That review is recorded as outstanding, and every page that draws on the material says so on the page.
Where the evidence stops
Known boundaries are published rather than left for a reader to discover.
- Reddit is not reachable by the collection stack used here. Every supported route returns a refusal, and Reddit appears in the majority of the search result sets captured for this subject, so this is a material gap rather than a footnote. Several other large discussion platforms refuse the same way.
- When a platform blocks collection, the operator can paste the content from his own browser. Those captures are labeled operator-supplied. They carry no automated collection record and no content hash, by design, because hashing hand-pasted text would assert an integrity guarantee the capture does not have. They are a weaker form of evidence than a collector capture and are never presented as one.
- This site publishes no first-party benchmark data. It holds no dataset of its own, and no client data of any kind is published.
- Search demand measurement for this subject disagrees between tools, so a low reading is treated as a measurement limit and not as proof that nobody is asking.
Corrections are recorded, not made silently
When something published here turns out to be wrong, the correction is recorded rather than quietly edited away. The research base keeps a dated account of what was found wrong, what changed, and which review found it, including reviews run by a second model instructed to attack the work rather than confirm it.
That log includes corrections that were themselves later found to be wrong, which is the reason for keeping it. A correction is verified against the original source, not against the finding that prompted it.
Who to tell when this page is wrong
Corrections are welcome, and a specific one is worth more than a general one: name the page, the sentence, and the source that contradicts it. Contact details are on the about page.