A record of scientific claims, the evidence under each one, and the exact test still needed before the claim is settled.
Peer-reviewed / Verified — not the same as true.
Every tag tells you how far the checking has gone — not whether the claim is right. The badges are explained on each entry and on the Method page.
Check a claim
Heard a claim and want to know where it really stands? Search the registry in plain language.
Find open hypotheses
A claim with no settled answer yet, and a specified test that would give one. For labs looking for the next move.
Suggest a claim
Know a claim that's repeated more than it's tested? Send it in. Every suggestion is public, and the ones taken up become entries held to the same six-field standard.
One claim, one schema, one standard of proof — applied the same way to every entry, including the ones we'd like to be true. These are the rules that keep the registry honest as it grows.
Peer-reviewed / Verified — not the same as true.
Every tag tells you how far the checking has gone — not whether the claim is right. Seven peer-reviewed papers can still carry a claim that drifted. Learn the badges:
Peer-reviewedexperts read the paper before it was published — not that it's correct
Verified at sourcewe confirmed the quote against the publisher — not that the claim is right
Replicatedanother lab re-ran the study and agreed — the strongest signal, yet still not a final verdict, since a field's understanding keeps evolving
What this is, and isn't
This is a record of scientific claims, the evidence under each, and the specific test that would settle it. It is not a list of things that are false, and not a forum. A claim appears here because its test hasn't been run or its evidence has been misread downstream — not because someone dislikes it. The one rule that makes it different from an opinion site: every claim, including every competing hypothesis, goes through the same six fields and the same test standard. A challenger idea gets no free pass for being the challenger.
The six-field entry
Claim as stated
Quoted from the paper, with a citation. A separate line records how it's been reported publicly when that differs from the paper.
Evidence base
What the study actually did: design, sample, and what was measured.
Untested assumption or unaddressed alternative
The specific gap, stated neutrally — what wasn't measured or ruled out. No guesses about motive.
What would settle it
The concrete next step toward an answer — usually a specific test, with the predicted result if the claim holds and if it doesn't. When the move is re-running an existing study, the entry labels it a required replication. For a claim that can't be tested directly, this is the trail to follow instead: the path back to its earliest source.
Current status
Replications, published critiques, author responses, and whether the test has been run.
Entry author and conflicts
Who wrote the entry and any stake they have in the outcome.
Evidence tiers
Claims sit at different strengths of evidence, and most drift happens when one strength is quietly swapped for another. This ladder is universal — it works for biology, AI, physics, economics, anything. Each entry then defines what the tiers mean concretely for its own field, and collecting that domain-specific version is part of writing the entry. Entry 001 spells out the prion version: present in a sample → acts in cultured cells → seeds an effect in an animal → causes clinical disease.
T1Observed or associated. The pattern is seen, or two things correlate. No controlled test yet.
T2Shown under controlled or model conditions. Demonstrated in a lab, a simulation, or a simplified stand-in for the real thing.
T3Shown in the real system, partially. Demonstrated in the actual target, but limited — small scale, one setting, or preliminary.
T4Established in the real system, measured. The full claim demonstrated in the real-world target, with a quantified result others can reproduce.
Domains
Every entry files under a Core › Sub › Topic path: one of 14 core science categories, then a sub-science, then the specific topic. A field is placed under its best-fit primary category and cross-linked to the others it spans, so an interdisciplinary claim is findable from every angle without being duplicated. The full field index (~1,000 sub-sciences) sits behind these cores.
Three separate questions, never collapsed into one. Each has its own vocabulary so a reader can't mistake "we checked the quote" for "an expert signed off" for "the science is settled."
Hypothesis — is the science settled?
A claim Under Investigation may carry no hypothesis status at all. It gets one only once a testable form of the claim is found — "Open Hypothesis" is earned, not a default.
Open HypothesisNo adequate test of the claim has been run.
Under TestA test is in progress.
Resolved — SupportedThe test was run and the claim held up.
Resolved — Not supportedThe test was run and the claim didn't hold up.
Resolved — MixedResults conflict.
Peer review — has an expert vetted this entry?
UnreviewedDrafted; no qualified reviewer has checked it yet.
Open for reviewA reviewer is invited or working on it.
Peer reviewedA domain reviewer has signed off.
Source verification — did we confirm each quote?
Verified at sourceThe quote was confirmed against the publisher's own page.
Located, not readWe have the specific source in hand — a consistent citation or the document itself — but haven't read it in full yet.
UnlocatedNo source in hand yet. The node is named from general knowledge and points to where a primary source should be, but none has been pulled. The weakest state — common early in an investigation.
UnreadablePaywalled or otherwise unavailable to us.
Identity unresolvedWhich paper the citation points to isn't established.
Per-source tags
Peer-reviewedExperts read the manuscript before publication. Not a claim that anyone repeated the study.
PreprintPosted publicly without peer review.
ReplicationWhether another lab re-ran it and got the same result — the test that actually settles a claim. "Unknown" means we haven't checked; "Not a primary test" marks reviews and cross-field citations, which have nothing to replicate.
FilesThe original paper or dataset attached to the node, so anyone can check it directly.
Entry flags
ActiveThe claim is cited in current work.
LegacyAn older open question, often one that current work quietly rests on.
ContestedTwo credible groups disagree on the right test; the entry shows both and picks neither.
What makes a good test
Concrete and pre-specified.
It names the measurement, the design, and the threshold — before anyone runs it.
Symmetric.
It states what result would support the claim and what result would sink it. One-sided isn't a test.
Runnable in principle.
A lab could actually do it, and the entry names roughly what that would take.
It has a checkable outcome.
A test names a result to look for ahead of time, then checks whether it appears. "This deserves more study" is a fine starting point — the entry's job is to turn it into a test by naming what result would answer it.
Writing neutrally
Bias usually enters through word choice, not claims. The tables below are the house style: the left column is what to avoid, the right is what to write instead. This guide is public on purpose — readers can hold every entry to it.
Words that assume motive
Avoid
Use instead
willfully ignored, masked, hid, suppressed
not addressed, not measured, not reported
wants the public to believe
concludes, states, reports
overlooked, missed
outside the study's scope, not tested in this design
Words that close an open question
Avoid
Use instead
proven, verified, confirmed
supported by, consistent with, replicated in
debunked, disproven, myth
not supported by, contradicted by [study]
fact, settled, established
widely accepted, current consensus
Words that put people on teams
Avoid
Use instead
mainstream, establishment, institutional
the study, the authors, the field's current consensus
Big Pharma, industry-funded (as a dismissal)
funded by [X], per the conflicts section
skeptics, believers, so-called, alleged
researchers who propose [alternative], or drop it
Words that add heat
Avoid
Use instead
shocking, alarming, troubling
notable, unresolved
crisis, scandal
open question, replication gap
always, never, completely
in [n] of [n] studies, in this sample
The test for any entry: could the study's authors read it and agree it describes their work accurately, even if they dislike the conclusion? If yes, it's neutral enough.
Conflicts, provenance, and honesty
01Every entry names its author and their stake. Any paid relationship touching a claim is disclosed on the entry.
02Quotes are marked located or verified at source. AI-assisted research is disclosed, with reviewer and date.
03Peer-reviewed is not the same as true. Verified is not the same as true. The badges describe process, not correctness — seven peer-reviewed papers can still carry a drifted claim.
04Paid review work is firewalled from public status. Paying never changes a rating, and this page says so plainly.
The night sky
The Map
Every claim is a star. The more it connects — to its fields, and to the claims that rest on it — the brighter it burns. Colour says what state the claim is in, so a bright one is never mistaken for a true one.
Every star is an entry — hover to see what it connects to, click to open it
Brightness & size — how much turns on the claim: how many unresolved claims rest on it, or wait on it to be settled. A claim that would settle a test someone is running now shines brightest; a claim nothing waits on stays dim however much has been written about it. Brightness is leverage, never truth.
Color — the claim's evidence state:
Slate — under investigation, with no testable form of the claim found yet. No verdict has been offered.
Yellow — an open hypothesis: a test exists and has not been run.
Blue — under test right now.
Green — tested and held up, or independently replicated. The paths that look right so far.
Red — tested and did not hold up.
Violet — contested; two credible camps disagree on the test.
Two ways to read it
The overview sky
The whole registry at a glance. Claims cluster into their domains, so you can see where a field is crowded, where it's quiet, and which corners glow with popular-but-untested ideas. Zoom from a domain down to a sub-science and then a single claim. This is the front door — the sense of the whole.
A claim's constellation
Click any star to center it and see only what it touches: the claims it rests on, the ones that cite it, and the domains it spans. Small enough to read no matter how large the registry grows — the working view, for tracing one claim and finding the open tests around it.
What to settle next
The same measure as the brightness above, written out. A claim is here because
unresolved work waits on it — directly, or further up a chain. This is the list to read if you are
deciding where to spend effort; the sky is for seeing the shape, this is for choosing.
What you are looking at
Every star is a registry entry and every line is a connection the registry records: a solid line is a dependency one entry declares on another, a dashed line is a field the two share. A field holding most of the registry is not drawn as a line — it would join everything to everything — but it still pulls those claims together, so a crowded field reads as a cluster with a faint haze between its members rather than a thicket. Position is computed from those links, so claims that depend on each other sit together; the layout is seeded, so the same registry always draws the same sky. Nothing here is illustrative, and there is no filler: an empty region means nobody has filed a claim there yet.
Who's behind it
About
Assumption Check is built and operated by Priority Nexus LLC — an independent effort to turn down the noise around scientific claims and show people what the evidence actually says.
Priority Nexus LLCPrioritizing BoundariesOperator of the Assumption Check registry
Why this exists
Most confusion isn't a shortage of information — it's an excess of noise. Data, headlines, and urgent claims pile up faster than anyone can check them, and the original finding, the actual study and the real measurement, ends up buried under layers of citation that drift further from the source with each repetition.
Assumption Check is a flashlight for that dark. It traces a claim back to what was actually measured, marks how far the checking has gone, and names the one test that would settle it. Nothing is called true for being popular, peer-reviewed, or often repeated — only when the evidence holds.
The tools to do this at scale arrived just in time. Recovering a buried source and mapping how a claim drifted across a literature once took a specialist weeks; it can now be done carefully, in the open, and checked by anyone.
Core values
Identity
Every claim, source, and reviewer is named. You always know what you're looking at, and who stands behind it.
Transparency
The method, the wording rules, and each entry's confidence level are public — so the work can be held to its own standard.
Integrity
Findings follow the evidence, not funding, popularity, or pressure. Paid review never changes a public rating.
02The paid review service (Assumption Check for labs) runs separately from the public registry. Paying for a review never changes a claim's public status.
03Claims are scored the same way regardless of who holds them or how widely they're believed — and the score says how far the checking has gone, never that a claim is simply "true."
Get in touch
For collaboration, review, or funding enquiries: theaistherapist@gmail.com · Priority Nexus LLC.
To propose a claim for the registry, use the public suggestion form (a free GitHub account is needed), or email it to the address above.
