1

What Trackademia is

Trackademia counts the publications and citations of university departments so they can be compared on equal terms. It starts from each academic's CV, adds what a verified Google Scholar profile and the OpenAlex and CrossRef databases know about the same works, and counts every work by one set of written rules.

The Trackademia team collects the CVs, runs the reading and the enrichment, and reviews whatever the checks flag. Members of an institution read the results: Database holds search, academics, publications, departments, journals and reliability; Analytics holds the charts, in four tabs; and this guide explains where each number comes from and what it means.

Demo University today. A publication two colleagues wrote together is one of the counted publications here, and counts for each of them on their own pages.

1. 1What Trackademia is
2. 2Where the data comes from
3. 3What counts as a publication
4. 4Whose work it is
5. 5Citations
6. 6Comparing departments
7. 7Professor rank
8. 8Career charts
9. 9How reliable it is
10. 10Glossary

2

Where the data comes from

Four sources describe an academic's work. They are not equally trustworthy, so each has a fixed role.

Why the CV comes first

OpenAlex and Google Scholar both build author profiles by machine, and both sometimes put a namesake's work on the wrong person or split one person across several profiles. A CV is the academic's own claim and is never mixed up with someone else's. So the CV decides what exists, a Scholar profile is used only after it has been checked against the CV (section 4), and OpenAlex only ever adds detail and citations to works one of those two already backs. A record only OpenAlex has stays in the database and is never shown or counted.

How a CV is read

A CV is turned into numbered lines. An AI model reads those lines and returns, for every entry, the line numbers it sits on together with its title, year, kind, journal and status. The model never supplies the text of an entry: code rebuilds each entry from the numbered lines and checks every answer against them.

• The title must match the entry's own lines at a similarity of at least 85 out of 100.
• The year must be printed in the entry, or in a year heading above it.
• A DOI must be printed in the entry; one that is not is removed.
• Lines no entry covers are read again, and a section that still has more than 10% of its lines unread is read again with more care.

An entry that fails a check goes to the Trackademia team for review; nothing on a CV is dropped silently. Other facts read from a CV, such as the PhD year and the academic's positions, are kept only when they are found on the lines the model points to.

For a CV entry without a printed DOI, CrossRef is searched with the entry's own text. A DOI is accepted only when its title agrees at 90 out of 100 or better, its year is within 1 year of the CV's, and the academic's surname is among its authors.

One row per work

The same article can be listed on the CV, on Scholar and in OpenAlex. These records are joined into one work for each academic when:

• they carry the same DOI and are the same sort of output (a book and a review of it can share a title, never a DOI); or
• their titles are the same apart from punctuation, capital letters or a subtitle added or dropped, their years are at most 1 apart, and nothing contradicts it: two different DOIs of published versions, different sorts of output, or different numbers in the titles, such as a survey of 2012 and a survey of 2016.

Clear pairs are joined by code. Unclear pairs, with titles that agree at between 80 and 92 out of 100, are judged by an AI model, and code checks the judgment before anything is joined.

When the records of one work disagree, each detail is taken from the source most likely to be right:

3

What counts as a publication

A work several colleagues wrote counts in full for each of them, and once for a department that has several of them. Undated works count in totals and are left out of charts with a year axis, which say how many they left out. Every page and every download count exactly the same works.

Kinds

The kind says what sort of output a work is. It comes from the CV's own section and wording first, then from CrossRef or OpenAlex, and for a record only Scholar has, from markers in its journal and title.

Statuses

Working paper is a status, not a kind: a working paper of an article is still an article, and once the article is published the working paper becomes its preprint. Works that are not counted are still listed on each academic's page, under In progress or Not counted, and no count, chart or comparison includes them.

Related outputs, counted once

One piece of writing can leave several distinct outputs. Each keeps its own row, but together they form a group that counts once, under its lead: the published original.

These are separate works, each counted: a chapter in someone else's edited volume; a review the academic wrote of someone else's book; a response to a roundtable on their book; an editor's own chapter in a volume they edited; and an article and a later book on the same topic under a different title.

Settled questions

• A book review the academic wrote counts as a publication.
• Reports, policy briefs and white papers count.
• Letters, comments and replies in journals count, as articles.
• An article that came before a book of the same title is a separate work, unless the CV lists them as one.
• A record only Scholar has, with no year, no journal and no citations, is not counted: about nine in ten such records are not real publications.
• Co-authored work gives full credit to each author.
• A paper in the NBER or a similar series is a working paper, and the preprint of its published version.
• A CV entry with no status marked is published.
• Working paper is a status, not a kind.
• A work is either the academic's and shown, or not shown; there is no state in between.
• Downloads count exactly what the pages count.

The settled questions on Scholar profiles, the h-index and ranks are in section 4, section 5 and section 7.

The works switch

Pages that count works carry a switch: CV and Scholar (the default) counts works the CV or the verified Scholar profile backs, CV only the works the CV backs, and Scholar only the works the profile backs. The two lists disagree, and the switch shows by how much rather than hiding it.

4

Whose work it is

Verifying a Scholar profile

Scholar sometimes adds a namesake's work to a profile, or offers a namesake's profile in a search. So a profile is checked as a whole, by code, against the CV, and it is the academic's only when it passes all three tests:

1. At least 3 of the CV's titles, or 30% of them, are on the profile. A title matches at 90 out of 100 or better.
2. If the profile names another institution or email domain, or a name that differs by more than initials, accents and short forms, it needs more: 10 CV titles under a name that agrees, or at least 50% of its articles being CV titles.
3. It is not a much larger profile holding only a small part of the CV. A profile at least 2 times the size of the CV, carrying under 25% of the CV's titles, with under 15% of its own articles on the CV, is refused: that is the pattern of a namesake's profile seeded with a few of the academic's works.

A profile that fails is refused outright, and no AI model is asked to overrule the tests. Institutions are compared with the academic's own department, university and email domain, never with a generic name.

Each article on a verified profile is then checked on its own. It is not the academic's when its full author list does not include their surname, when it is a review of their own book, or when it is another contributor's chapter in a volume they edited.

OpenAlex

The OpenAlex author is found through the DOIs on the CV, and through an ORCID when there is one. It is accepted when at least 5 of the CV's titles, or 30% of them, are among its works. It is used only to add identifiers, journals and citations to the academic's own works, never to add works: OpenAlex splits people across profiles and merges namesakes too often to be trusted with that.

Why Scholar coverage varies

294 of the 480 academics with a CV at Demo University have a verified Scholar profile. The share differs a great deal between departments, for three reasons.

• No profile exists. Academics make their own Scholar profiles, and many never made one. Nothing can recover these, so a department with more such people looks weaker on any Scholar figure for reasons that have nothing to do with what it publishes.
• The profile could not be verified. A profile that fails the tests above is refused, and someone with no CV has nothing to check a profile against.
• The profile is very large. Scholar sends 100 articles at a time and each academic gets 4 requests per run, so the longest profiles are read in part and completed on a later run.

Read any Scholar figure alongside how many of a department's academics have a profile: the Reliability page in Database shows it for each department.

5

Citations

Citations count how often other work cites an academic's work. There are two measures, and they answer different questions.

A department's profile total is the sum of its academics' totals, so a paper two colleagues wrote together is in both of their totals, as it is in the field's own reporting. A profile total covers a whole career, so the year window does not change it; when a window is set, the page says so beside it (section 6).

Scholar and OpenAlex

Google Scholar usually counts more citations than OpenAlex for the same work, because it indexes citing documents OpenAlex does not see, such as theses, working papers, books and reports. Neither is wrong; they count different things.

Under the works measure each work takes Scholar's count when any of its records has one, and otherwise OpenAlex's, for works the CV lists. The two are never added together. A citation source switch reads Scholar first and OpenAlex where Scholar has no count (the default), or one source alone. Because a total can mix the two, every total says how many of its values came from each source.

A work's citations are its group's: a book's count includes the citations of its separately listed chapters and reprints, and a record several rows point to is counted once. Journal weights (section 6) apply to the works measure only; a profile total is one number with no list of works behind it, so there is nothing to weigh.

The h-index

The h-index is the largest number h such that h of a person's works have at least h citations each. It is the h-index on the academic's verified Google Scholar profile. For someone without one, it is computed from the OpenAlex citations of the works their CV lists, each work once with its group. A work with no count adds nothing, and someone none of whose CV works has a count has no h-index, which is not the same as an h-index of 0. OpenAlex's own author h-index is never used, because it can include namesakes' works. Filters do not move it.

No figure from OpenAlex's author record (its works count, citations, h-index or i10-index) is shown or downloaded.

6

Comparing departments

Analytics compares departments: named rosters of academics at the same institution, also called cohorts. A large department publishes more in total than a small one, so the charts let you choose what a department is judged by.

Each department belongs to a field, such as Political Science, Chemistry or English. Publication and citation norms differ by field, so Analytics compares the departments of one field at a time (Political Science unless the Field control says otherwise), and every number, the journal weights' inputs included, is that field's alone.

Analytics has five tabs that share one rail of controls at the left: the departments and the filters (the kind, the year window, the works and the rank, each any combination of its choices). Under them the rail states How to rank: what counts (the citation measure and source), how departments are judged, and how journals are weighted. Those settings are changed on the Journals tab and apply to every tab. Each tab counts only its own charts:

A department's publication total counts a paper two of its members wrote together once; the mean and the median count it for each of them. For publications every academic counts, one with none as 0. For citations under the works measure every academic counts too; under the profile total only academics who have a total count, because a missing profile is not zero.

A score equation

An equation can use these names, with + - * / ^, parentheses, numbers and log, ln, log10, sqrt, exp, abs, min, max:

Equations such as mean * log(1 + people) or median * sqrt(people) say a bigger department is better, but not in proportion to its size.

B leads on the total, A on the mean and the median, and B again on the equation. None of these is the right answer; each answers a different question, so choose the one you mean to ask.

The year window

The year from and year to fields narrow every chart to a span of years; a blank field leaves that end open.

• Publication counts, the works citation measure, the top ten percent share, the journals table, the rank bars, the output by rank and every chart on the Data tab count only the works published inside the window. An undated work is still counted in totals, and is on no chart with a year axis.
• A chart with a calendar-year axis shows only the window's years.
• Citations over time shows the citations received in the window's years. From Google Scholar that is each academic's profile history for those years; from OpenAlex it is the citations received in those years by the works the kind and works switches select, whatever year the works were published. Both then mean the same thing.
• Career charts count the works published inside the window, and an academic only in the career years that fall inside it (section 8).
• Promotion timing counts the promotions that happened inside the window.
• The profile total and the h-index cover a whole career and do not follow the window. When a window is set, the page says so beside them and links to the works measure, which does. A rank is current, so the department makeup by rank does not follow the window either.

Journal weights

Some readers value a citation in one journal more than one in another. A journal weight multiplies the citations of every work published in that journal, under the works measure: a weight of 1 leaves them as they are, 0 drops them and 2 doubles them. Weights are a matter of judgment, so each reader can keep their own version. There are two ways to set them, and they combine.

• By hand: a weight from 0 to 10 typed against a journal. It overrides the formula for that journal.
• By a formula: an equation over four facts about each journal, such as 0.5*A + 0.5*B.

The equation reads each fact as a percentile: the journal's position among the journals, from 0 for the lowest to 1 for the highest. The four then share one scale, so 0.7*A + 0.3*B counts A for more than twice as much as B, and one famous paper cannot dominate the weighting. Only journals with at least a minimum number of works (by default 20) are ranked; a smaller journal has too little to measure and weighs 1 unless a weight is typed. The results are then scaled in a straight line so the lowest journal weighs 1 and the highest the top weight (by default 3). A journal the equation gives no number for weighs 1.

A, B, N and P are counted over every academic of the field's departments, never only the departments on the chart, so a journal's weight does not change with what is being compared.

The journal averaging 40 citations is one step above the one averaging 9, not several times its weight.

7

Professor rank

Each academic has one current rank: the highest current appointment at their own department's university. The CV lists positions, each with its title exactly as printed, the institution and the years. A position is kept only when its title and years are found on the CV lines it comes from. The AI model copies positions and never classifies them; code turns a title into a rank.

• A named chair counts as distinguished, not full.
• A joint appointment, such as "Professor of Law and Political Science", is one rank, from the highest current title at the department's university. A courtesy or affiliated appointment never sets the rank.
• Flags are kept beside the rank: tenured or untenured when the CV says so, acting, visiting, courtesy.
• Administrative roles (dean, chair, director, provost, president) are kept separately and never set the rank.
• The rank is as of the CV's date, or its latest dated position, and is shown with that year.
• The CV decides. A verified Scholar profile's affiliation line is a cross-check: when it names a different rank, the academic goes to review. A rank set by hand overrides both and is kept through every re-run.
• Everyone counts in Analytics by default, whatever their rank; the rank filter narrows every chart.

The rank filter

The Rank control on the Analytics rail keeps the academics whose current rank is among the ranks ticked, any combination of them, and every chart and table on every tab is then counted over them alone: the summary, the charts over time, the career charts, the faculty charts, the journals and the Data tab. Nothing ticked, or every rank, is everyone. Three shortcuts tick a group:

The journal weight formula ignores it: a journal's weight is a fact about the journal over every academic of the field.

Rank charts

• Faculty ranked against faculty names every academic with their rank, such as "Jane Doe · Associate".
• Department makeup by rank shows each department's academics by current rank, as counts or as shares, from the most senior. A rank is not a work, so the year window and the kind and works switches do not change it.
• Output by rank shows, per department and rank, the median publications of an academic, or the median citations under the citation measure. A rank off the tenure track that fewer than 10 academics of the departments shown have is pooled into Other.
• Promotion timing shows, per department, the median years from the PhD to the first associate professorship and to the first full or distinguished professorship, over the academics whose CV states a PhD year and lists that position at any institution. Courtesy and visiting positions are not read. A position with no start year is not read, and a first start before the PhD year is left out. With a year window set, only promotions inside it count.

8

Career charts

A career chart lines academics up by the years since their PhD instead of the calendar, so a department of recent PhDs can be set beside one of senior scholars.

• Career year 0 is the PhD year the CV states. The axis runs from 5 years before it, where graduate-school work sits, to 45 years after.
• Each point is the mean over the academics who had reached that career year by last year (2025), not over everyone. Someone whose PhD was in 2020 has reached career year 5 and not year 20; counting them as 0 there would bend every curve down.
• A career year that fewer than 5 of the academics count in is left as a gap, so one long career cannot draw the end of a curve on its own.
• Academics whose CV gives no PhD year are left out. 457 of the 774 academics at Demo University have a PhD year from their CV.
• Citations on a career chart are those of the works published in that career year, not the citations received during it, so recent work has had less time to be cited.
• The kind and works switches apply, and so does the year window: only works published inside it count, and an academic counts at a career year only when that year, their PhD year plus the career year, falls inside the window. With a window from 2015 to 2020, someone whose PhD was in 2005 counts at career years 10 to 15 and at no other, so they are never a zero in years the window leaves out. The current year is incomplete and left out.

9

How reliable it is

What is checked

• Every entry read from a CV is checked against the CV's own lines, and one that fails goes to review instead of being dropped.
• A DOI from CrossRef is accepted only when its title, year and author agree with the CV.
• A Scholar profile is verified as a whole against the CV, and each of its articles on its own.
• OpenAlex never adds a work.
• Records are joined by written rules; an AI model judges only the unclear pairs, and code checks its judgment.
• The Trackademia team works through everything the checks flag.

What can still be wrong

An error is a work wrongly counted, and just as much a real work that is missing.

• Missing works. A CV can be out of date, or not a full CV at all (an annual report listing only work in progress, for example), and a scanned CV with no text cannot be read. That academic's count is then below their real record. A person with no CV has no publications here, which is not the same as zero publications.
• One work counted twice. When the CV and Scholar describe one work too differently to join, often a title reworded between the working paper and the article, CV and Scholar counts it twice. Pages say how many titles look shared.
• No work types from Scholar. Most works only Scholar has are of kind other, so a kind filter such as articles only leaves them out.
• Unrecognised translations. A translation the CV does not mention is not recognised as one and counts as a separate work.
• A DOI for another version. An accepted DOI occasionally belongs to a different version of the work, most often a book chapter reprinting a journal article.
• Uneven Scholar coverage. Scholar figures partly measure who has a profile (section 4).

The Reliability page in Database shows, for each department, how many academics have a CV, an OpenAlex identity and a verified Scholar profile, and how far each source agrees with the CV.

10

Glossary

Academic

A person on a department roster; every figure is about academics.

Department

A named roster of academics at one institution, compared as a group; also called a cohort.

Source

A place a list of someone's work comes from: the CV, Google Scholar, OpenAlex or CrossRef.

Record

One entry in one source.

DOI

The permanent identifier of a published work.

Work

One piece of writing, joined from all its records into one row per academic.

Group

The related outputs of one work, such as a preprint, a reprint or a translation, counted once.

Lead

The published original a group is counted under.

Kind

What sort of output a work is: article, book, chapter, thesis and so on.

Status

How far along a work is: published, forthcoming, under review, working paper or in progress.

In progress

Works not yet published; listed on the academic's page and not counted.

Not counted

Works of a kind that is not a publication, such as theses, talks and media; listed and not counted.

Counted

A work that is the academic's, of a publication kind, published, and counted once with its group.

Shown

A work the CV or a verified Scholar profile backs; a record only OpenAlex has is never shown.

Verified profile

A Google Scholar profile that passed the tests against the CV.

OpenAlex identity

The OpenAlex author the CV's DOIs point to, used only for detail and citations.

Works switch

The control that counts the works the CV backs, the Scholar profile backs, or either.

Profile total

The citation total on a verified Scholar profile, or else the OpenAlex citations of the CV's works.

Works measure

The citations of the works matching the filters, each work once with its group.

Citation source

Which count a work takes: Scholar first and OpenAlex where Scholar has none, or one alone.

h-index

The largest h such that h of a person's works have at least h citations each.

Total, mean, median

A department's figure as a whole, for its average academic, and for its middle academic.

Score equation

An equation over the total, mean, median and number of people that a department is judged by.

Journal weight

A number that multiplies the citations of the works in a journal.

Percentile

A journal's position among the journals, from 0 for the lowest to 1 for the highest.

Career year

Years since the PhD the CV states; year 0 is the PhD year.

Rank

An academic's highest current appointment at their department's university, such as Full or Teaching.

Flag

A note kept beside a rank: tenured, untenured, acting, visiting or courtesy.

Administrative role

Dean, chair, director, provost or president; kept apart and never sets the rank.

Review

The Trackademia team's look at a case the checks flag.