Principles and commitments

How it works says what Cabdell does. This page says why: the principles it means to protect, and the commitments it makes when those principles conflict.

Taking the editor away does not take editorial power away. It moves it into the rules: who may flag an article, how much a vote weighs, which reviews count, who settles a dispute. Those rules are choices, not facts of nature. So they must be written down, justified, and open to criticism, and this page is where we do that.

What Cabdell is for

Scientific publishing bundles three different things: making an idea public, judging it, and crediting the person who proposed it. Cabdell keeps them apart. Publishing needs no one’s permission. Judging is open, signed and accountable. Credit is earned from the judgement of others, field by field.

The test we hold ourselves to is a single question: does Cabdell make it easier to find, judge, question and correct knowledge, without raising new barriers for those who have no recognition yet?

Principles

Reading is free. Anyone can read everything Cabdell shows, with no account, no wallet and no payment, always. The only exception is what this service has to stop showing under the law (see “Permanence, correction and erasure” below); even that stays in the public record.
Publishing needs no permission. Nobody approves an article before it appears. Its quality is judged afterwards, in the open.
The record is permanent; its interpretation is not. Nothing is deleted from the record: versions, reviews, votes and retractions stay on the blockchain. What changes is what we know about a work: its status, its amendments, its citations, its seal.
Recognition is earned, field by field, from others. Nobody gains reputation from their own actions, for money or by anyone’s decision, and recognition in one field gives no weight in another.
The rules are public and the same for everyone. Every rule that weighs a vote, qualifies a review or settles a dispute is written down with its numbers, and every result can be checked against the record.
No operator is needed, ourselves included. The record lives on a public blockchain and the texts on IPFS, and the rules are published, so anyone could build an indexer and a website that reach the same results. The source of Cabdell’s own indexer and website is not public yet; we intend to publish it under an open licence. We aim to be replaceable.
Figures describe; they do not rank people. Reputation, the Thread Score and the Peer-reviewed seal summarise what the record shows about articles and contributions. They are not a measure of a researcher’s worth, and they are not designed for hiring, promotion or funding decisions. We follow the spirit of the San Francisco Declaration on Research Assessment (DORA) and the Leiden Manifesto: judge work on its content, keep the figures open and reproducible, and say what each one measures and what it does not.
Personal data stays minimal. An address is enough to take part. Anything more, such as a name, an ORCID iD or contact details, is each person’s choice, and the few rules that ask for it say why.

When principles conflict

Principles collide. For each tension below we say what we choose, what it costs, and how we limit the cost.

Openness and abuse

We choose openness. There is no filter before publication, ever. Abuse, plagiarism and fabrication are handled afterwards: anyone with reputation 10 in an article’s field can flag it, and when the flags of a round reach a weight of 30 the article becomes disputed and governance decides whether to clear or retract it.

What it costs. Spam and bad work can appear, sometimes for a while, sometimes for good.

How we limit it. Flags need reputation in the field. A disputed article is marked for every reader and takes no new version until the dispute ends. A retraction is visible to every reader, and the authors of articles that cite a retracted one are told.

Flags can also be abused. A simulation showed that ten fresh addresses voting for each other can raise enough flags in a week to dispute any article of a field. A qualified jury, not one key, settles disputes (see the governance section below). Contract v4 makes a false flag costly: every flag deposits a bond of 1 ALGO, lost if the dispute is cleared (half paid to the article’s author, half kept by the protocol) and doubled for the next flags on that article for a year; a flag weighs at most 10, so a dispute needs three people; and a flagger’s reputation in the field must be 30 days old. A dispute that nobody resolves can be ended by anyone after 60 days: the article is restored and the bonds are kept, so an author gains nothing by disputing their own article.

Permanence, correction and erasure

We choose permanence. The blockchain record cannot be rewritten. A work is corrected by adding to it, with a new version, an amendment or a retraction, never by erasing.

What it costs. A mistake stays on record. So does personal information that someone later regrets publishing.

How we limit it. The forms of the app say that what you publish is public and permanent, and this page and Privacy and permanence say what that covers for every action, before you sign anything. A new version replaces the old one in what readers see first. A retraction marks the work for every reader.

What we can and cannot do. We cannot remove anything from the blockchain. We can stop keeping and serving our own copies, on our IPFS node, our indexer and our website. Copies that other people keep are beyond our reach.

When we stop showing something. We stop showing a record on our indexer, our IPFS gateway and our website in four cases only:

an order of a court or a competent authority;
a notice that shows the content to be manifestly illegal, such as child sexual abuse material, terrorist content, defamation or a copyright infringement: a substantiated notice is enough, we do not wait for a court;
content against the terms of use, such as spam, malware or phishing;
a request about one’s own personal data under data protection law.

We then purge the copies we kept, and the IPFS gateway answers “410 Gone”. The record’s page says, in place of the content, why, on which ground, since when and how to contest the decision, and its author is told, unless an authority forbids it. A public transparency report lists each case by its record and ground, never its content; cases of child sexual abuse material, terrorist content and personal data appear there only as counts by month, so that the report never points to them. Terrorist content is kept, encrypted and unseen, for the six months the law requires, then deleted; child sexual abuse material is kept nowhere. The record on the blockchain does not change, and anyone else can still read it from there or from their own copies. Disagreement is never a reason: it is answered with reviews, comments and flags.

Immutability and fixing mistakes

We choose immutability. The contract cannot be modified and has no administrator. A rule that nobody can change cannot be changed to favour anyone, ourselves included.

What it costs. A bug or an unfair rule cannot be patched in place.

How we limit it. An independent security review comes before MainNet. If a serious flaw appears after it, the remedy is a new deployment, announced publicly in advance, with the old record still readable and a clear account of what carries over and what does not. A silent change is never an option.

Transparency and privacy

We choose transparency of judgement. Every action is public and signed. Reviews are not anonymous, and votes and follows are visible. Whoever judges others’ work answers for that judgement.

What it costs. Some reviewers may hold back criticism of powerful authors, and early-career researchers may feel exposed.

How we limit it. An address need not carry a name, although a review that counts for the seal needs a verified ORCID iD, so those reviews are named. Cabdell has no accounts and asks for no e-mail address to take part; it never reads the e-mail addresses of ORCID records, and keeps no profiles beyond the public record. Only someone who writes to us (a legal notice, a data protection request) gives an address, to get an answer. Browser notifications store only the browser’s notification address and the account it follows. Anyone can object to the figures Cabdell derives about them: their Thread Score and reputation are then not shown, and they are never drawn for a jury (see Privacy and permanence).

What we decided about delayed identity. Cabdell does not offer reviews whose author is revealed later. On a public blockchain, the address that sends a review is visible to anyone at once. Hiding it in our interface would only pretend to protect the reviewer, and real protection would need cryptography that Cabdell does not have. Anyone who wants distance can review from another address, and that review does not count toward the seal. We watch the effect of open reviews on candid criticism during the pilot, and we will reopen the question if the evidence calls for it.

Past recognition and present judgement

What holds today. Reputation never decreases, and a vote weighs 1 + ⌊√reputation⌋ in the article’s field. The square root keeps influence from growing as fast as reputation, and a limit of 50 reputation per field per day slows how fast it is earned.

What it costs. Authority earned long ago weighs as much today as it did then. Recognition brings influence, and influence is the power to recognise others. That loop can entrench a consensus and the people who hold it.

How we limit it. Reputation is separate for each field. The square root and the daily limit damp it. The seal asks for verified identities, the absence of conflicts of interest, and no favourable reviews returned between authors.

What we decided, after a simulation. We simulated ten years of a field under several rules.

Recognition stays lifetime and never decreases. Weighing votes by recent reputation changed little in the simulation (one or two points), so we do not adopt it.
Repeated votes are damped. What mattered was people voting for each other again and again. In contract v4, a voter’s second vote for the same person counts half, the third a third, and so on (see Reputation and vote weight). In the simulation, this kept a ring of fresh addresses at 2% of the best researcher’s reputation instead of above it. It also cut the founding cohort’s share of influence from 14% to 9%, and lifted a well-founded minority result from the 64th to the 91st percentile.
A cartel of established researchers is not stopped by this. That is why coordinated voting will also be detected and shown.

The protocol’s autonomy and the need for governance

What we choose. Governance exists only where rules cannot decide: registering new fields, settling disputes, and handing its role to another address, which must accept by signing. It cannot edit, hide or delete anything, and it cannot touch reputation, votes, deposits or bonds.

What it costs. Today governance is one signer, a concentration of authority at odds with everything else here.

How we limit it. Its powers are narrow and every act it takes is public on the blockchain. The MainNet key will be generated offline, as a post-quantum account.

What we decided: a qualified jury. Disputes will be settled by a jury, not by one key. When an article becomes disputed:

Seven jurors are drawn at random, in a way anyone can check, from the people qualified in its field, one chance per person. A juror has a verified ORCID iD, reputation 10 in the field or 3 works in its area on OpenAlex, and no conflict of interest with the authors or the flaggers. These are the same tests a review must pass to count toward the seal; the list of people who qualify is fixed when the dispute starts, and their iD and reputation must be at least 90 days old by then.
Each juror answers “clear” or “retract”, with their reasons, signed and public, within 14 days.
Two thirds of at least five answers decide.

Every case has an outcome:

Two thirds: two thirds of at least five answers decide; when one side holds two thirds of all seven seats, the outcome comes at once.
No consensus: at least five answers without two thirds draws a second jury of new people, once.
Silence: after seven days, every silent seat passes to the next person drawn.
Too few answers: fewer than five answers draws a second jury of new people, once.
No outcome twice: when the second jury reaches no outcome either, governance decides alone, with public reasons. Silence or disagreement never clears an article by default (D-162, after a security audit showed that a few attacker jurors could otherwise clear anything).
Recusal: a juror who recuses is replaced at once.
A small field: the jury is drawn from the area, then from everyone qualified. Only when fewer than five people qualify anywhere does governance decide alone, with public reasons.

In the first phase, governance carries out the outcome on the blockchain within seven days, and the dispute page shows whether it did, did otherwise, acted before the jury, or is late. Later, the contract itself will take the jury’s verdict. How juries work.

If the key is lost or stolen. If the key is lost, no field can be added and no dispute resolved until a new deployment. In contract v4 anyone can end a dispute 60 days after it started, so a lost key cannot leave articles disputed forever. If the key is stolen, the thief can only do what governance can do (register fields, resolve disputes, propose a new governance address), and all of it in public; a retraction still waits 14 days after the dispute started and must name its published reasons. We would announce the theft and prepare a new deployment.

Useful figures and the incentives they create

We choose to show figures, because they help readers find their way among many works.

What it costs. Once figures matter outside Cabdell they will be optimised. People may trade votes, form cliques or avoid risky work, even without any bad faith.

How we limit it.

There is no global ranking of people.
Each figure says what it measures and what it does not.
Self-citations and returned favours are left out where the rules can see them.
Coordinated voting will be detected and shown.
We will never promote Cabdell’s figures as a basis for hiring or funding.

What Cabdell will never do

Charge to read, or to publish beyond the network costs. Those costs go to Algorand and to deposits locked in the protocol, not to us.
Sell a token, show advertising or sell data.
Rewrite or delete the record.
Grant reputation for money or by anyone’s decision.
Approve or reject an article before it is published, unless the law one day requires it.

How this page changes

This page is versioned. Every change is recorded, with its reasons, in the decision log. Anyone can propose a change by publishing notes on Cabdell that discuss it. Read with Privacy and permanence, Flags, disputes and governance and Reputation and vote weight.