Where every link comes from

Every link in the record carries the kind of evidence behind it: cited by a repository's record, stated by Wikidata, found in a thesis's own text, inferred from a convention or from shared papers, or claimed by the person on their ORCID record. The kinds are never merged, and only some are counted. Each cited link can be traced to the record it rests on and the day it was read, and where the sources stop, the page says so.

The evidence ladder: six kinds of evidence, each with its mark and an example. Cited, stated and extracted links are drawn and walked as the lineage; cited and extracted are counted among a person's advisors and students, and Wikidata's are counted apart; inferred, claimed and indexed links are drawn and never counted.

What provenance means here

Provenance is the history of a statement: where it came from, who made it, and how it reached the page in front of you. In this record the statements are links between people and theses: this person wrote that thesis, that person supervised it, a third examined it. Every link is kept with three things beside it: the source that gives it, such as a university's repository, a national register or Wikidata; the kind of evidence it is; and the read of that source it came from.

A graph whose every statement carries its source and the act that produced it is often called a provenance knowledge graph. The W3C's provenance model, PROV-O, names its parts. An entity is a thing: a repository's record of a thesis, or a link drawn from it. An activity is something done over a span of time: a read of a source. An agent is who answers for an entity or an activity: the repository that keeps the record, and the site that read it. The diagram puts one link in those terms.

One link in the terms of the W3C provenance model: the link is an entity generated by an activity, a read of the source, which used the repository's record, another entity; the link was derived from that record; the record is attributed to the repository and the read is associated with Academic Ancestry, the two agents.

The site does not publish its record in PROV-O; the model is named here as a way to read the record, not as its vocabulary. What a program reads is the schema.org data each person's and each thesis's page carries, in which every role on a thesis is given with its kind of evidence in words, so a program can tell one kind from another as a reader can.

Two rules follow from keeping provenance this way. Every number the site shows was counted from a read the record can name: the record keeps which read of each source its links came from, and a page shows the day that read was made, not the read's own name. And every link says which kind of evidence it rests on, so a reader can weigh it and need not take the site's word for it.

The kinds of evidence

Six kinds of evidence stand in the record. Each has its mark, the one a person's page uses: a filled dot for cited, a ring for stated or indexed, a dashed ring for claimed or inferred. A supervision found in a thesis's text has no mark of its own on a person's page and is said in words there; these pages draw it as a square.

The marks: a filled dot for cited, a ring for stated or indexed, a dashed ring for claimed or inferred, and here a square for a supervision found in a thesis's own text.

Cited: a repository's record: A repository's record of a thesis names a person in a role: the author, the supervisor and co-supervisor, a committee member, an examiner, a referee, an opponent, as the repository labels them. This is the record's foundation, and most of what it holds. Most often the role is in the record's own fields. Sometimes it is on a page the repository serves beside the record: the thesis's own page on its site, the text of its title page, or a national register's entry for the same thesis. A reading rule written for that one source finds it there, and such a rule is checked on a sample against the theses' own pages before it is used. Where a record names a committee chair and no supervisor, the chair stands as the supervisor.; It proves that the repository says so. It cannot prove the repository is right: a name typed into the wrong field stays there until the source mends it. Nor can it say that a name on one record is the same person as a name on another; that is the subject of who is who.
Stated: Wikidata: Wikidata, the open knowledge base, states who was whose doctoral advisor. The site keeps a copy of those statements and walks them, with the repositories' records, for a person's line up and down. Anyone may edit Wikidata, and a statement may or may not cite a source of its own.; It proves that Wikidata says so. A person's page lists Wikidata's advisors and students in parts of their own, with a count of their own, never added to the record's.
Extracted: found in a thesis's own text: A language model reads the text of a thesis for the supervisor it names, and each finding is kept with the file, the page and the line it was found on, quoted. No finding enters the record until at least 100 of them have been read by a person against the theses themselves and a share of at least 0.90 found right. A person's page says such a link was found in the text of the thesis, and its card quotes the line.; It proves that the thesis's text names that person in those words. It cannot prove more than the words say, and it is never counted as cited.
Inferred: suggested, not recorded: A link a rule suggests and no record states. At a few German repositories, whose theses name referees and often no supervisor, the first referee is drawn as a likely supervisor, by that convention, at the rate it was found to hold at each source. People who wrote papers together, as OpenAlex lists them, are drawn beside a person as coauthors. An inferred link is drawn dotted and said to be suggested. It is never counted, and a climb up a line never steps across one.
Claimed: the person's own word: Where the record links a person to an ORCID iD, their page can show, in a fold of its own, the works their ORCID record lists, asked of ORCID only when a reader opens it. It shows what the person says they wrote. It is never joined to the record's works, never counted and never used to tell who someone is.
Indexed: what an index lists: OpenAlex, an open index of scholarly works, holds author records with their papers. Where the record links a person to an OpenAlex author record by an identifier, a person's card shows the papers OpenAlex lists there, as OpenAlex holds them. The record keeps none of them and counts none.

Some links are cited and still stand beside the lineage. A committee member, an examiner, a referee, an opponent, a custos or a jury named on a thesis record is drawn on the student's row and never counted as an advisor or a student. The teachers and mentors Wikidata states where no thesis stands between two people, and the authors a person cites most or is cited by most, are kinds of their own, drawn beside a person and never counted with the lineage.

Why a new kind is never folded into cited

It would be easy to let a good new source of evidence join an old one. A supervisor found in a thesis's text and checked on a sample is right most of the time, and counting it as cited would make the record look fuller. It would also make the record harder to read: the share of the record a repository stands behind could no longer be told, and a reader who trusts a repository's catalogue and not a language model would have no way to set one aside.

So a new kind of evidence is given a name of its own, a mark or words of its own and its own place in the counts before any link of that kind is drawn. The kinds also decide what may replace what. Where a repository's record names a supervisor, the recorded link takes the place of a suggested one on the same pair; nothing suggested ever takes the place of something recorded.

The totals of the record count the kinds apart. The coverage page counts the cited doctoral supervisions, and says Wikidata's statements apart from them; a person's page counts their advisors and students from the record and lists Wikidata's beside them, with a count of their own.

Where the record stops

The record holds what its sources hold. A line that ends may end because the sources end, not because no one came before or after, and the site says so where it happens. On the sheet, a short bar over or under a person's mark says the record holds no one beyond them on that side, which does not mean there is no one, and the person's card says the same in words. A climb up a line ends with the words that the records stop there, which is not to say the line began there.

A missing source is not an absent student. Many repositories are not read yet, a few refuse to be read and are left alone, and many records name no supervisor at all. The coverage page shows, by country and by decade, where the record holds supervisions, and leaves a blank where it holds none rather than writing a zero, since a blank and a zero say different things.

A thesis whose record names no supervisor keeps the reason, and its page says which of these holds:

the record names the author and no one else;
it names people besides the author, none of them as a supervisor;
it names a committee, examiners or a jury, and no supervisor;
it names no author, so no student can be linked to a supervisor;
the supervisor it names is the author again, and no one is drawn beside themselves;
the only supervisor it points to is a first referee, a link suggested and not recorded;
its supervisor is already linked from an earlier record in the same repository, so the link is drawn once.

None of these is filled by a guess. If you know an advisor or a work the record lacks, report it: a report is checked against the record and against Wikidata and kept, and it is never drawn or counted by itself. What the record leaves unknown is gathered on what the record does not know.

Tracing a link back

A cited link can be followed to the record it rests on. A person's page lists each advisor with the title of the thesis that links them, and the person's own theses each lead to a page of their own. A thesis's page holds every step: its title, which is a link to the thesis's page at its repository; the year, the school and the level; everyone the record names, by role, each with the kind of evidence their link rests on; where the record stops, when it names no supervisor; the file and the licence where the record gives them; and a citation that names the day the record was read.

Tracing a cited link back, left to right: the link on a person's page, the page of the thesis it rests on with every role and its kind, the repository's own page for the record, and the day the site read it, named in the page's citation.

The repository's own page is the record as its source keeps it now. Where the two differ, the site has not yet read the change: each source is read again, some every night and some at the longer pause their hosts ask for, and its links are replaced whole, so a page never shows half of one read and half of another.

A program can follow the same steps. A thesis's page has its schema.org data at a second address, in which each person the record names is a role carrying its kind of evidence in words; the page for developers lists the addresses. Behind both, the record's answers carry the read each figure was counted from, so any number can be traced; the pages show its day.