Docs / The process, stage by stage

The process, stage by stage

A result is progress on a problem, a solution of it, or a disproof of it, signed by its author with ORCID. This page follows a result through every stage: what starts each stage, who acts, what they see and can do, what is signed and recorded, who can see it, what comes next and how to challenge it. It then says exactly how a problem's status on PubPhys follows from its results. How results and decisions are contested: contests. How problems themselves are reviewed, and who may decide what: editors, referees and their decisions. How to post one: posting a result.

Who may post. Any account with a linked ORCID iD, invited or (once ORCID registration opens) registered with its iD, posts results, contests and appeals through the same screening and review; there is no extra hold for new accounts. An account made with an email address reads, saves and follows, and links an iD to write (accounts).

Timing. Editors' and referees' decisions have no deadline. The only timers are the screening's (retried for up to six hours; a result still unscreened after seven hours is held), the 14-day rule for held results, the 14 days within which an editor triages a contest, and the window for an appeal. Authors, editors, referees and followers are notified in the site at each stage (by e-mail only in an email digest, if they choose one): who is told what.

The life of a result, and who acts at each stage.

When results are open

During PubPhys's launch, results, contests and comments open in a field once it has one editor (the chief editor counts). With one editor, a decision cannot yet be contested or appealed to anyone else: contests and appeals are published and wait for a second editor, with no deadline. Referees are appointed as the site grows; a solution is verified only after two referees recommend it.

Results open field by field. After the launch, a field takes them once it has at least 2 editors of its own, or the chief editor and one of its own, and 2 referees other than the chief editor, so that every result can be decided by someone other than its author and checked by two independent referees. Results accepted during the launch can still be contested after it ends. An editor is the field's own when their earliest editor appointment still in force, not counting one lent from another field, is in this field; editors lent to the field and suspended editors and referees do not count. PubPhys can also pause results everywhere. A rejected problem takes no results.

Kinds of result

Progress: a partial result: a special case, a bound, a reduction. It is accepted or rejected, never refereed.
Solution: answers the precise question. Accepted, it is refereed; two recommendations make the problem verified.
Disproof: shows the question is ill-posed or its premise false. Accepted, it is refereed; two recommendations make the problem disproven.

The stages of a result

Stage	Starts when	Who acts, and what they can do	What they see	Recorded	Who sees it	Next	How to challenge
Signed	The author signs with ORCID on the problem's page.	Nobody; the screening starts by itself. PubPhys checks once more, inside the signature, that the problem has not changed in substance, that the field is open, and that the author's ORCID iD has signed fewer than 5 results this UTC day.	The author sees the result's page and its signed record.	The `claim` record in the public log, with the fingerprint of the problem's revision it answers.	The log entry is public; the content only to the author and the field's editors.	Screening.	Nothing to challenge; the author can withdraw it.
Screening	Right after signing.	With the author's consent, an AI reads only the result's kind, text and references (what is sent) and asks whether it is a serious, intelligible piece of physics work. It does not judge correctness and never rejects anything.	Nobody looks at it yet.	Its report, for the field's editors only (it may be manipulated by the result, so it is shown to them as untrusted).	The author and the field's editors.	Published if confident; otherwise held. Held at once without consent, when AI is switched off or its daily spending limit is reached, or when the provider cannot be reached. A failed answer is retried for up to six hours; an hourly check holds any result still unscreened seven hours after signing.	The author can withdraw it.
Held	For one of the hold reasons below.	An editor of the field releases it (it is published and counts) or signs that it is not accepted at screening, with reasons, under one of the categories off topic, not physics, unintelligible.	The editor sees the text, references, declared AI systems, cited results, the hold reason and the screening report. The author sees the hold reason only.	Release: PubPhys's audit log, which is internal (only the people who operate PubPhys see it), not signed. Not accepted: a review record signed by the editor.	The author and the field's editors only.	Published, or not accepted at screening. A held result blocks substantive changes to the problem for 14 days; after that it no longer does, it is listed in the chief editor's section of the queue, and PubPhys's administrator is alerted. Held again after a granted appeal, it counts from the grant.	The author can withdraw it.
Published	The screening passed it, or an editor released it.	Anyone can read it. It counts for the problem's status at once. An editor of the field signs accept (a serious contribution of the stated kind that specialists should look at) or reject, with reasons. Accepting does not mean it is correct.	The editor sees the result, its history and, for a result the screening published, its report.	The decision is a review record signed by the editor.	Everyone.	Accepted or rejected.	The author can withdraw it.
Accepted	An editor signed accept.	For a solution or a disproof, referees of the field read it in full and each signs recommend or do not recommend, with reasons (referees). Progress is not refereed.	A referee sees the public result, its history and earlier recommendations.	The signed decision.	Everyone, with the editor's ORCID iD and reasons.	Refereed, or stays accepted until withdrawn or removed.	A contest against it, or against the acceptance.
Refereed	Two different referees of the field signed recommend on an accepted solution or disproof.	Nobody needs to act. Other referees may still give their recommendation, for or against; a recommendation against is shown and blocks nothing by itself.	As above.	Each recommendation is a `review` record signed by its referee.	Everyone, with each referee's ORCID iD and reasons.	The problem is verified (solution) or disproven (disproof) while the result counts.	A contest against the result or a recommendation.
Rejected	An editor signed reject.	Nobody; it stops counting.	Anyone sees the result, marked rejected, with the reasons.	The signed decision.	Everyone: the result stays readable, marked rejected, with the reasons.	Final unless appealed.	Appeal by the author within the 30 days (appeals); a granted appeal returns it to waiting for an editor.
Not accepted at screening	An editor signed it on a held result.	Nobody.	The author sees the decision and its reasons.	The signed decision.	The author and the field's editors only; it was never published.	Final unless appealed.	Appeal by the author within the 30 days (appeals); a granted appeal returns it to held.
Withdrawn	The author signs a withdrawal: once, at any stage until it is rejected, not accepted at screening or removed.	Nobody; it stops counting at once.	Anyone who could see the result sees it marked withdrawn, with the reason.	A `withdrawal` record signed by the author, with an optional reason (at most 500 characters).	Public if the result was published: both stay readable, the result marked withdrawn. Otherwise the author and the field's editors only.	Final.	It cannot be undone.
Removed	An editor removes it for copyright, personal data, abuse or spam, at any stage.	Nobody; it stops counting. Not the result's author, nor the author of another live result on the problem, may remove it.	Nothing of the text.	A review record written by PubPhys on the editor's instruction; the editor is in the audit log.	The text is withheld for good; the record's fingerprint stays in the log. A result that was public keeps its page, which says why; otherwise it is not found.	Final.	Through the administrator only.
Abandoned	A sealed result was revealed, and its author deleted the account while it was still screening or held.	Nobody; it never counts.	Nobody, not even an editor of the field.	Nothing new: the sealed record and its reveal stay in the log.	The text and its files are withheld for good; the record's fingerprint stays in the log.	Final.	Nothing to challenge.

Hold reasons

doubtful: the screening doubted that it is a serious piece of physics work.
likely reject: the screening found it likely to be spam, advertising, a crank claim, unintelligible, abusive, or to contain personal data about others.
low confidence: the screening passed it, but without enough confidence.
injection suspected: its text seems to address the AI (an attempt to steer the screening).
no consent: the author did not allow AI to read it.
ai unavailable: AI is switched off, its daily spending limit is reached, or the provider could not be reached.
appeal granted: an editor granted the author's appeal of its rejection at screening, so it waits for a fresh decision.
file unreadable: a source file could not be read as text (for example a scanned PDF), so the screening could not read it.
file too long: a source file is longer than the screening reads (40,000 characters, or more than 30 pages of a PDF).
quote withheld: a contest only: the text it quotes was withheld before the screening read it, so nothing was sent.

How a problem's status is computed

Every problem shows two statuses. In the literature is editorial information about published research. On PubPhys follows only from signed results and decisions on this site, by the rules below. It is recomputed whenever a result or a decision changes, and anyone can recompute it from the public records. A problem that is not public (held, not accepted at screening or removed) has no results that count, so its status is open for whoever can still see it.

A result counts when it is published, not rejected, withdrawn, refuted or removed, and answers a revision with the same core statement as the one the problem shows now. A new revision that changes only wording keeps every result counting; one that changes the substance cannot be made while a result is live (screening, published, or held for up to 14 days; revising a problem).

A result's standing changes only through upheld contests.

Intact: no upheld contest on error, gap or irreproducibility against it.
Incomplete: an upheld contest found a gap: a solution or disproof then counts as progress.
Refuted: an upheld contest found an error, or that it cannot be reproduced: it no longer counts.

The first rule that applies gives the status:

Contested: an accepted intact solution and an accepted intact disproof both count, or an open contest (triaged as substantive, not yet decided) stands against a counting solution or disproof or an effective decision on one.
Disproven: an accepted intact disproof has two recommendations from different referees of the field.
Verified: an accepted intact solution has two recommendations from different referees of the field. Both must have been referees of the problem's field when they signed; a recommendation against does not count against them.
Claimed disproven: an intact disproof counts, accepted or not yet decided.
Claimed solved: an intact solution counts, accepted or not yet decided.
Claimed progress: progress counts, or a solution or disproof found incomplete.
Open: nothing counts.

Only results on the problem itself count. A problem built on other problems shows its parents' results for reference, but no status passes from a parent to a child or back: a verified special case leaves its general problem open.

Worked examples

A solution accepted and verified

The records in order, as the result's history shows them:

A claim record (kind solution), signed by its author. The screening passes it; it is published and the problem is claimed solved.
A review record, decision accept, signed by an editor of the field. Still claimed solved: accepting is not checking.
A review record, decision recommend, signed by referee A. One recommendation: still claimed solved.
A review record, decision do not recommend, signed by referee B, naming a step they doubt. Shown; nothing changes.
A review record, decision recommend, signed by referee C. Two different referees recommend it: the problem is verified. If the author later withdraws it, it stops counting and the problem is no longer verified.

A held result released

A claim record signed without AI consent: nothing is sent, it is held (no consent); the problem stays open.
An editor of the field releases it (audit log, no record): it is published and the problem is claimed solved.

A withdrawal

A published claim record (kind solution): claimed solved.
A withdrawal record signed by the same author: the result stops counting, the problem is open again, or claimed progress if progress still counts. A second withdrawal is refused.

More examples

A solution passes the screening: the problem is at once claimed solved. An editor rejects it: it stops counting and the problem is open again.
A solution and a disproof are both published: claimed disproven (a disproof counts first). Once an editor accepts both, contested.
An editor corrects a typo in the problem: results keep counting. A change to the precise question waits until no result is live.

A solution contested for a gap and made incomplete

A published claim record (kind solution), accepted by editor A: claimed solved.
A contest record (grounds gap), signed by another physicist, quoting the step it disputes. The screening passes it; it waits for triage and nothing changes yet.
A review record, decision substantive, on the contest, signed by editor B (not editor A, who decided the result): the contest is open and the problem is contested.
A review record, decision upheld, signed by editor C (neither A nor B): the result is incomplete and counts as progress; the problem is claimed progress.

A rejected solution, appealed

A published solution, rejected by editor A: it stops counting and the problem is open again.
Within the 30 days its author signs an appeal record against A's decision. No AI reads it; it waits in the editors' queue.
Editor B (who decided nothing in the case) signs appeal granted: A's rejection stops counting, the solution waits for an editor again and counts, and the problem is claimed solved again.
Editor C (neither A nor B) accepts or rejects it. That decision is final: it cannot be appealed.

Who decides on a result

Editors of the problem's field, and the chief editor, accept or reject it by signed records (how editors sign).
Referees of the field, never the chief editor, recommend an accepted solution or disproof or not; never the editor who decided it (who may referee).
Nobody decides on, removes, or withholds a text about their own result, by account or by ORCID iD; nor about another result on a problem where they have a live result themselves.
Only one accept or reject stands on a result; it cannot be signed on a withdrawn or removed result. A contest that is upheld against the acceptance reverses it: the result waits for an editor again, and the editor whose acceptance was reversed cannot decide it again.
Contests are triaged and decided by editors with no stake in them: not the result's author or competitors, the filer, the signer of the contested decision or of any acceptance of the result, its referees (for a contest of the result), anyone who decided a contest of the result before, nor (for the outcome) the editor who triaged it (contests).
An editor or referee whose account is suspended is paused: not counted, offered work or notified, and signs nothing until the admin reinstates the account (suspension).

Removal and withheld texts

A result, or the free text of a withdrawal or of an editor's or referee's decision, can be withheld for copyright, personal data, abuse or spam. The record's fingerprint and its effect stay (a withdrawn result stays withdrawn, a decision stays in force); only the text goes, for good. Each withholding is a review record written by PubPhys on an editor's instruction, entered in the audit log and in a ledger that is re-applied after any restore from backup. The editor must not be the result's author nor the author of another live result on the same problem (for a decision on a problem: not the problem's author; for a contest or a decision on one, also not the contest's filer; for PubPhys's own review of a seed problem's or a topic's revision, any editor of the field who did not write, sign or approve that revision, from the record's page; never the decision's own signer). A contest is removed the same way; while it is undecided or an appeal on it waits, only by an editor who could decide it and did not sign the appealed decision (removal would close it). Removing a result or a contest also withholds its source files; one source file can be withheld on its own the same way, by an editor who could remove the item (for a contest, not its filer), and the item stays. A contest quoting a text that is withheld or removed shows its quote as withheld on its pages.

What the records show

A result is a claim record: its author is the ORCID iD that signed it, its target the fingerprint of the problem's revision it answers, and its content the problem's reference, the kind, the text, the references and the AI systems the author declares. It may cite up to 10 earlier public results on the problem or its parents. A withdrawal is a withdrawal record whose target is the result. Editors' and referees' decisions on a result are review records whose target is the result's claim record. A contest is a contest record whose target is the result or the contested decision; decisions on it target the contest. Anyone can check them with their bundles, without trusting PubPhys (the protocol).