How this is made
Crawlspace is built from a research sweep, not from someone’s experience. This page describes exactly what that means, including the parts that are weak.
Every claim carries its own evidence
A claim only appears here if it came with a direct quote from a named source. The quote is on the page, one click from the claim. If a claim’s only evidence turned out to be the title of the page it cited - which proves which document was read and nothing about what it says - the claim was dropped rather than published.
Claims are separated by what kind of thing they are
The single most useful distinction on this site is between what an operator has documented and what the field merely repeats. They are never mixed together.
- Documented position
- What the operator of the system says in its own documentation.
- Measured finding
- Something someone actually measured, with the method attached.
- Practical effect
- What is observed to happen in practice, as distinct from what is documented.
- Changed on
- A dated change. This is where advice quietly stops being true.
- Disputed between sources
- Sources that do not agree, presented without picking a winner.
- Echoed without a primary source
- Widely repeated in the field, but traceable only to other repetitions. Listed because knowing something is merely echoed is worth as much as knowing it is documented.
Sources are graded by who published them
The grade is about who is speaking, never about whether they are right. A vendor study with a named method and a real sample is better evidence than a trade-press summary of it; the grade just tells you whose incentive is in the room.
- first party
- The operator of the thing being described, or the body that defines it.
- trade press
- A publication reporting on the field.
- vendor
- A company selling something in the field it is describing.
- other
- Anything else, including personal sites and archives.
The verdict is checked against the claims
The verdict at the top of each subject is written from that subject’s established claims and nothing else, then judged against those same claims by a separate pass. When the judge names a statement it cannot support, the verdict goes back once with those statements quoted, to be rewritten as the narrower thing the claims do support - naming the specific engine, date or study instead of generalising.
A verdict that still fails after that is not published. The subject then shows its claims with no verdict at all, because a summary this site cannot stand behind is worth less than the absence of one.
The judge fails closed. A reply it cannot read counts as a refusal, not as a pass - the opposite default would mean a broken check looked exactly like a clean one.
What this does not do
- It is not a ranking-factor list. Nothing here tells you what will move a position, because almost nothing in the public record actually establishes that.
- Absence of a claim is not evidence of absence. If a subject has three claims, that means three survived, not that three exist in the world.
- The grounding check catches invention, not every overstatement. Tested adversarially, it reliably rejects fabricated numbers and events. It is more permissive about a summary that says “always” where the claims only support“since 2015”.
- Dates are the date of the evidence, not the date a thing became true. A claim dated 2016 may have been true for years before, and may have quietly stopped being true since.
- Refused claims are not necessarily false. They are claims whose evidence did not support them as written. Several are probably true and badly sourced.
Corrections
If a claim here is wrong, or its quote does not say what the claim says, that is a defect worth fixing. Every claim carries its source so you can check it, which is the point. The full source list is here.