Citations and Thread Score

Cabdell counts the citations its articles receive from other Cabdell articles, and turns them into one measure, the Thread Score. Both are computed by the Cabdell indexer from the blockchain and the articles’ texts, so anyone can compute them again (see Verify without trusting cabdell.press). A third number, Citations (all works), comes from outside Cabdell (OpenAlex) and is explained at the end.

The rules come from DECISIONS D-126 to D-129 and the indexer’s src/citations.ts.

What counts as a citation

An article cites another Cabdell article when its current version points to it in one of four ways:

How	What the citing article contains	Shown as
Link	a link to the other article’s page on the same network (on TestNet, any form of https://cabdell.press/testnet/a/<number>, also deeper pages such as /a/12/citations)	“Links to this article.”
DOI	a DOI that the other article declared and whose record does not contradict it (verified, not checkable, or not checked yet)	“Cites this article by its DOI.”
CID	the CID (content fingerprint) of one of the other article’s versions	“Cites a version of this article by its CID.”
Amendment	it is an amendment whose parent, recorded on the blockchain, is the other article, and its submitter is an author of that article	“Amends this article.”
Note	it is an “amendment” of the other article published by someone who is not one of its authors: a third-party note, never shown as the article’s own amendment. Contract v4 (D-163) accepts an amendment only from an author of its parent, so new ones cannot appear	(a note about this article)

The indexer looks for web addresses, DOIs and CIDs everywhere in the text (links included, so a References section with DOI links counts) and in the references list of the article header (see Write the article). A link only counts within the same network: a TestNet article is cited by TestNet articles.

Further rules:

• An article never cites itself, and each pair of articles counts once.
• Citations from retracted articles are not counted.
• Only the current version counts: a reference removed in a new version stops counting.
• Citations are computed again whenever content, versions, DOIs or their checks change.
• A CID or a DOI belongs to one article (D-162): a CID to the article that used it first; a DOI to the article whose declaration of it was verified, else to the one that declared it first. Another article that publishes the same CID or declares the same DOI receives none of its citations, and its page names it as a duplicate.

To cite in a way that counts, see Cite an article.

When a cited article is retracted

When an article is retracted, every Cabdell article that cites it (and is not retracted itself) shows a warning naming the retracted article and the date, and its confirmed authors receive a notification once (D-140). The correction reaches the work that relied on it, without anyone having to track it.

Where citations are shown

• On an article: “cited by N on Cabdell · cites M”, linking to its citations page with two tabs, Cited by and References on Cabdell; each article in them says how it cites.
• In search: “cited by N” on each result, and the sort Most cited on Cabdell (see Search).
• On a profile: Cited on Cabdell, the citations that the person’s articles received, linking to a page that lists each citing article, which article it cites, how, and whether it is a self-citation (an article the person also wrote). Self-citations are included in these counts and marked; the Thread Score leaves them out.

Thread Score

The Thread Score is Cabdell’s own citation measure, named after Ariadne’s thread (DECISIONS D-127 to D-129). It goes from 0 to 100, and 50 is typical. An article’s Thread Score is its percentile; a person’s is a weighted mean of their articles’ percentiles. Cabdell shows no h-index.

An article’s Thread Score

1. Count the citations from other people. Only citations from Cabdell articles that are not retracted and share no author with the cited article count. “Author” means the publishing author and the co-authors who confirmed. Each citation is dated by the citing article’s current version. Since version 4 of the Thread Score (D-162), two more rules keep cheap articles from inflating it:

   • the citing article must be vouched for: it has at least one review that counts toward the seal, or one vote from someone who, when voting, was an expert in its primary or secondary field by the seal’s rules: reputation 10 from at least 3 people other than the article’s authors, earned at least 90 days before, each with a verified ORCID iD or expertise of their own (version 5, D-163; version 4 asked only for reputation 100, which accounts voting for one another could reach);
   • one citation per citing submitter, cited article and year: many articles by the same submitter citing the same article in the same calendar year count once.

   The list of citations and “cited by N” still show every citation; only the Thread Score applies these rules.

2. Choose the comparison group. The article is compared with every article of its primary field. When the field has fewer than 20 articles, it is compared with the articles of its area (the field’s OECD group); when the area also has fewer than 20, with every article on the network. Retracted articles and amendments (corrections of one’s own work) take no part and have no Thread Score.

3. Compare two at a time, at the same age. For each article B of the group, take the age of the younger of the two articles, and count the citations each had received when it was that old. The article “beats” B if it had more, ties with B if they had the same number. It also ties with itself.

4. Compute the percentile (mid-rank): 100 × (articles beaten + half the ties) ÷ size of the group.

Because ages grow, the percentiles are computed again every day.

A worked example

Imagine a field with four articles. (A real comparison needs a group of at least 20; with fewer, the indexer uses the area or the whole network, as above. The arithmetic is the same.) Ages are in years; each list gives when each external citation arrived, counted from the article’s publication.

Article	Age	Citations arrived at	Total
A	6	0.5, 1.5, 3, 5	4
B	4	0.3, 0.8, 1.6	3
C	2	0.4, 1.2	2
D	1	none	0

Article A, compared with each article at the age of the younger one:

A against	Common age	A had	The other had	Result
A (itself)	6	4	4	tie
B	4	3	3	tie
C	2	2	2	tie
D	1	1	0	beats

A’s percentile is 100 × (1 + ½ × 3) ÷ 4 = 62.5. Done the same way, B beats C and D and ties with A and itself: 100 × (2 + ½ × 2) ÷ 4 = 75. C beats D and ties with A and itself: 50. D only ties with itself: 12.5.

A has the most citations in total, yet B ranks higher: at four years B had as many as A, and at two years B had more than C, which A only matched. The extra time A has had does not count in its favour.

One year later, suppose nobody receives a new citation. A (now 7) and B (now 5) are compared at 5 years, so A’s citation that arrived at 5 years now counts: 4 against 3. A beats B, C (at 3 years: 3 against 2) and D (at 2 years: 2 against 0), and rises to 100 × (3 + ½) ÷ 4 = 87.5. An old article keeps climbing while it is cited.

A person’s Thread Score

A person’s Thread Score is the mean of the percentiles of their articles (as publishing author or confirmed co-author), each weighted by their share of it: 1 divided by the number of authors. One typical article, at 50 and with weight 1, is always added:

Thread Score = (w1 × p1 + w2 × p2 + ... + 50) ÷ (w1 + w2 + ... + 1)

p = the percentile of each article
w = 1 ÷ the number of its authors (publishing author and confirmed co-authors)

Continuing the example (before the year passes): if you wrote A alone (62.5, weight 1) and C with one co-author (50, weight ½), your Thread Score is (62.5 + 25 + 50) ÷ (1 + ½ + 1) = 137.5 ÷ 2.5 = 55.

The added typical article keeps small samples honest: one article at 100 gives (100 + 50) ÷ 2 = 75, not 100. Its effect fades as articles accumulate. Amendments and retracted articles do not count.

Why it is built this way

• Across fields: each article is ranked within its own field, so a field that cites a lot and one that cites little can be compared.
• Across ages: every pair is compared at the same age and every year is in the comparison, so a recent article is not judged against years it has not lived, a strong year weighs little, and an old article keeps climbing while it is cited.
• Across careers: a mean, not a sum, so a short career is not outweighed by a long one.
• Across team sizes: each author counts only their share of an article.
• Honest with little data: the typical article added keeps one lucky article from making 100.
• Reproducible: everything comes from the blockchain and the articles, so anyone can recompute it.

Thread Score is not reputation: it gives no vote weight and changes nothing on the blockchain (see Reputation and vote weight).

Where it is shown

• On a profile, Thread Score (rounded), linking to “How the Thread Score works” on the person’s citations page, which also says how many articles it rests on.
• On an article’s citations page: “Thread Score N: cited by others more than N% of the articles of its field, each compared at the same age”, with the size of the group, whether it is the field, the area or the whole network, and the number of citations from other people.

Citations (all works), from OpenAlex

For a person with a verified ORCID iD (see Link your ORCID iD), the profile also shows Citations (all works): the citations of all the works of that iD, not only those on Cabdell, as OpenAlex, an open index of scholarly works, counts them. A note gives the number of works and the citations of each of the last five years, with a link to OpenAlex. The count is read again with the ORCID record, weekly; if a lookup fails the last count is kept, and an iD that OpenAlex does not know shows none (DECISIONS D-125).

This number is OpenAlex’s, not Cabdell’s: it does not enter the Thread Score, and like everything from ORCID it never changes reputation.