Science is not the problem. The record around it is.
Almost every scientist is honest. The few who aren't don't win by being clever — they win because almost nothing about how a study came to exist has to be written down anywhere a person can check.
Every one of these is true today.
- Because a study can be published without ever naming who paid for it.
- Because nobody has to prove the data they're describing exists.
- Because the person who owns the data is allowed to be the person writing the study.
- Because “that's confidential” is an accepted answer to “can we see it?”
- Because a study can be withdrawn and go on being cited for years.
- Because a withdrawal is a sentence in a PDF, not something a computer can act on.
- Because a study can be deleted outright and leave no trace it ever existed.
- Because the same small group can fund, write, review and cite each other, and nobody is counting.
- Because a researcher can name a funder on one paper and not the next, and nothing flags it.
- Because when an audit fails, “we couldn't check it” ends up looking the same as “we checked it and it's fine.”
- Because funding, authorship and citations sit in separate databases that don't join up.
- Because the databases that do exist can be edited, and quietly are.
- Because by the time anyone notices, governments have already acted on it.
- Because nobody's job is to go back and tell the people who acted on it.
Not one of those requires a conspiracy. They're just the conditions. Leave them in place and this happens on its own, over and over.
Take the conditions away and it stops working.
- Nothing can ever be deleted. A withdrawal is a new entry, not an erasure.
- Every claim carries its receipt — who paid, who wrote it, who actually saw the data.
- Funding is a structured link, not a footnote — so for the first time it can be counted.
- Citations point at a specific version, so a withdrawal reaches every paper that used it, instantly.
- “We couldn't verify this” is recorded as its own permanent fact, and never blurs into “verified.”
- Anyone can publish, with no gatekeeper — so nobody can be shut out, and nobody can be shut down.
- Trust is earned by what you're willing to show, and how it's calculated is published in full.
- The scoring algorithm is public and versioned — every change to it is permanent record too.
- We make no money on it. Nobody can pay to rank higher or to score better.
- It's all one record — so you can finally ask the question nobody can answer today.
When a hundred studies agree, how many separate funders are actually behind them?
Today that takes a journalist years, if anyone bothers at all. It's been done by hand before — in tobacco research, and in a case where a drinks company's own published list of the studies it paid for turned out to be incomplete. Here it's a click.
In 2020, every one of those conditions lined up at once.
A study built on data nobody was allowed to check paused a worldwide clinical trial and reversed national policy in two countries. Here is the same story, told as the record would have kept it.
It was withdrawn. People kept citing it anyway.
The record could be quietly changed
The record keeps its receipts
The Reef doesn't catch fraud. It makes “nobody can check this” something you can see on day one, instead of something you find out on day thirteen.
The databases scientists use still have this wrong.
These were found while checking the dates on this page against the two biggest open research databases. Records that can be edited don't stay correct on their own.
The publication date is wrong by three weeks
It says 1 May 2020. The paper came out on 22 May 2020. The error is longer than the paper's entire life before it was withdrawn.
The two main databases disagree by 48%
On how many times one related paper was cited: 3,368 versus 4,968. Same paper, same question, two very different answers.
Two more, if you want them
The data owner is listed at the wrong workplace
A parsing error puts him at a teaching hospital. The most important fact in this whole case — that the data belonged to a private company, not a hospital — is simply wrong in the record.
The withdrawal is dated to the day it was published
The record says the paper was withdrawn on 22 May, the day it came out. It was actually withdrawn on 4 June.
One study, and everything attached to it.
The same story drawn as connections instead of dates. Click anything to see what it is and what it's linked to. Nothing is ever removed from this picture — withdrawn work stays, marked.
Red means a warning is open. Grey outlines are things that should be here and aren't — like the funder, which no public database records for any of these papers. That gap is the whole reason The Reef exists.
Click anything on the map.
How it works
Anyone can publish here, and nothing can ever be deleted. What you earn is trust — and this page is the complete rulebook for how that's measured.
Anyone can publish. No gatekeeper.
No editor, no gatekeeper, no takedowns. Work published by someone completely anonymous stays visible forever and can never be removed. It just doesn't carry any weight, and we say so plainly rather than hiding it.
What you earn is trust, by showing your work. The more you're willing to put on the line, the more the record is willing to say about you.
The four levels
We check employers and funding against outside records using Flare's data connector, so you don't have to take our word for it either.
What publishing looks like
- You write the record. What the work claims, who wrote it, who paid for it, where the data came from, and which authors have actually seen it.
- You choose how much of the work itself goes on permanently. The three options below. That choice is a disclosure, so it counts toward your score.
- You pay the network to write it down. From that moment it is permanent. Neither you nor we can remove it.
- Everything after that is an addition, never an edit. A correction, a withdrawal, an audit result, a challenge from a reader. Each one attaches to the original, and the original never changes.
Where the work itself is kept
Option A is open from level 2 upward, and here is the reason, said plainly. Nothing on the record can ever be removed. So nobody anonymous should be able to write whatever they like into a place that can never be cleaned out again.
What it costs
Publishing is not free and we are not going to call it free. It costs whatever the network charges to write your record down. All of it goes to the network. None of it comes to us.
These figures are measured, not guessed. We ran the real thing and read the real cost.
The score
Every piece of work gets a score, worked out only from what its own record does and doesn't contain. Here is the entire calculation.
See it worked out on the study in this demo (scored under v0.1)
Warnings are a nudge, not a verdict
A warning means something about this record can be checked, and it didn't check out. It does not mean the science is wrong, and it does not mean anyone cheated. Simple automated checks in this field get it wrong 37–44% of the time, so we treat every warning as a reason for a person to look — never as a conclusion.
Warnings never lower your score. If they did, every argument about a warning would turn into an argument about a number. Keeping them apart means the two answer different questions: the score is what you were willing to show, the warnings are what a machine couldn't confirm.
All twelve things we check for
Counting how independent a consensus really is
This is the thing The Reef is really for. When lots of studies agree, you can ask a question nobody can easily ask today: how many genuinely separate funders are behind them?
A hundred studies that all trace back to one funder two steps down the chain is not a hundred studies agreeing. That's been shown by hand before — in tobacco research, and in a case where a company's own published list of the research it funded turned out to be incomplete — but no product does it, because the funding links aren't in the record in the first place. Here they are, from the start.
If we get it wrong, we change it in public
This page has a version number. Every change to the score, the levels or the checks is added to the record permanently, with the date and the reason. Nobody — including us — can quietly re-score the past.
Where a record like this stops helping
Worth being honest about the limits. Take the paper on where COVID came from — it was never withdrawn, the journal declined to withdraw it, and its authors have said the changes between drafts were ordinary scientific updating. That's disputed, and this demo doesn't take a side. It's here because it shows exactly where a public record helps and exactly where it stops.
What you'd have seen in 2020: the changes between versions, dated, as they happened — instead of reconstructed by journalists three years afterwards.
What you still wouldn't see: private emails and private chat. A public record holds what people published. It does not hold what they said in private, and pretending otherwise would fall apart the first time someone pushed back on it.
And if we are ever ordered to take something down: we cannot delete it from the record, and we would not try. What we can do is stop showing it on this site. If that ever happens, the removal itself becomes a permanent public entry with its reason attached, and the record stays readable straight from the network. Nothing here can be silently deleted. That is not the same as promising everything will always be on this screen, and we would rather say so now than be caught out on the difference later.