A result
A measured effect, numerical estimate, observed relationship or experimental outcome.
THE LINEAGE PROJECT
What does a paper inherit?
A citation tells us that one paper points to another. It does not tell us what crossed the gap. The Lineage Project follows the actual inheritance: findings, methods, measuring tools, categories, datasets, errors, corrections, arguments and even copied references.
01
A citation graph is a skeleton.
The lineage begins when we put the tissue back.
Suppose a 2026 paper cites a paper from 2006. The database records one link. A human reader can see many different events hiding inside that link.
The later paper may repeat a numerical result. It may use a questionnaire introduced in the older paper. It may borrow a definition. It may reproduce a method but reject the conclusion. It may cite the paper only because the older work became a famous error. It may even have copied the reference from a review without closely reading the original.
Bibliographic databases usually count all of these as the same thing: one citation.
The Lineage Project treats the citation as an invitation to reconstruct transmission.
That turns the literature from a pile of documents into something closer to an evolving family tree — except that scientific descendants can inherit from hundreds of ancestors, reject their parents, revive forgotten methods and carry an error while discarding the result that originally made the paper famous.
02
The first experiment: retractions as a natural marker.
A retraction gives us something rare: a known date when an ancestor’s formal status changes.
That makes retracted papers unusually useful for studying scientific inheritance. Before the notice, a paper may be cited as ordinary evidence. After the notice, the same bibliographic object can continue to circulate — but the meaning of that circulation becomes testable.
Our first working case took the ten most-cited retracted papers highlighted by Retraction Watch and rebuilt the open citation graph around them. Then, where text was available, we recovered the words surrounding later citations.
The result immediately broke the simple story that “a citation after retraction is a bad citation.” Some later papers criticized the old work. Some discussed it historically. Some cited an old version when a corrected replacement existed. Some appeared to keep using a substantive claim or tool. These are different events.
The public case study is live at retractions.transductive.science.
WHY RETRACTIONS?Not because retracted science is the whole lineage problem. Because the retraction date acts like a dye injected into a river: it helps reveal where information continues to flow.
03
The larger project goes backward as well as forward.
We want to reconstruct long, continuous chains of scientific inheritance.
The ambitious version asks for the longest defensible chains that connect contemporary research to earlier scientific practice through actual citations and identifiable inheritances.
That is not the same as finding the oldest paper somebody happens to cite. A meaningful lineage should be able to say what each generation received from the previous one. The object may change over time: a measurement becomes a standard, a standard becomes a dataset field, a concept becomes a textbook category, a method becomes software.
The nineteenth-century research university matters here because seminar papers, dissertations and specialized research communities helped make written work into a repeatable training and inheritance system. Citation indexing in the twentieth century then made parts of that ancestry computationally visible. Digital full text lets us begin asking what actually moved along the links.
04
What can travel down a scientific lineage?
The working vocabulary is deliberately concrete.
A way of doing
A protocol, statistical test, experimental procedure, software workflow or analytical trick.
A tool that becomes infrastructure
A scale, questionnaire, benchmark, sensor, reference material or calibration procedure.
A way of dividing reality
A diagnosis, taxonomy, named effect, variable, construct or contrast that later work keeps using.
Something reusable
A dataset, specimen, model, codebase, image collection, reference sequence or corpus.
A controversy or canonical example
A paper can survive because it becomes the famous thing later researchers criticize, teach or use as a warning.
05
A paper can die in one way and live in another.
That is the conceptual bridge from the Lineage Project to The Paper format.
If scientific knowledge were simply a collection of validated propositions, a withdrawn proposition should be relatively easy to remove. But research practice is not built only from propositions.
A paper may introduce a measuring scale that thousands of later studies adopt. It may define a category that becomes ordinary vocabulary. It may popularize a method whose validity does not depend on the paper’s headline result. It may become a negative example that keeps generating research after its claim collapses.
This is why publication design matters. A linear paper hides these different objects inside one container. The Transductive format should let readers move among claims, methods, data, interpretations, versions and public explanations while preserving their relationships.
THE CONNECTION
The Lineage Project asks what survives between papers. The Paper asks how one publication should expose those parts clearly enough that future readers can tell.
06
Current research program
The project is exploratory, but the work can be made concrete.
Trace long chains
Genealogy
Build citation paths that connect modern papers to earlier work, then distinguish a merely bibliographic link from a defensible inherited object.
Mark status changes
Intervention points
Retractions, corrections, replacements, disputed measurements and changed standards provide dates at which an ancestor’s status changes.
Recover nearby text
Context
Read the sentences around each citation and, where necessary, the surrounding section. Machine classification can create review queues; human inspection decides ambiguous cases.
Follow research objects
Transmission
Track findings, methods, tools, datasets and categories separately instead of assigning one truth value to an entire paper.
Publish the graph
Public inspection
Every aggregate claim should be explorable down to the papers and citation excerpts that produced it, with downloadable data and explicit uncertainty.