Abstract
Researchers build understanding over time through reading, discussions, seminars, and experiments. Most digital tools store information such as papers, notes, or datasets, or represent curated knowledge through knowledge graphs, but they do not record how a researcher connected these elements while learning. Here we describe a simple methodological approach implemented in Pilus. Scientific entities such as genes, papers, researchers, organisms, processes, and conferences are represented as nodes, and the links created by the user act as records of how understanding was formed. Notes are attached to these entities as contextual traces of specific moments of learning. The result is not a knowledge graph and not a notebook, but a structured record of how a researcher progressively connects scientific elements. We describe this model, compare it to existing tools, and outline how it could be evaluated in practice. We also emphasize that the graph is independent from its visual rendering, so the method remains analyzable and portable across different interfaces.
Introduction
Scientific work involves continuously connecting pieces of information encountered in different contexts: a paper read months ago, a seminar just attended, a researcher met at a conference, a gene studied in the lab. These connections are rarely recorded explicitly. Notes capture fragments of information, reference managers store papers, and knowledge graphs encode established relationships, but none of these systems represent how a researcher personally linked these elements over time.
Pilus is based on a simple idea: if scientific entities are represented explicitly, and if the user creates links between them while reading, attending conferences, or reflecting, then the resulting structure becomes a record of how their understanding was built. It treats cognition as a process that can be externalized, not only a collection of facts.
Core Model
Pilus relies on three simple principles.
Real scientific entities as anchors
Nodes correspond to identifiable entities such as genes, papers, researchers, organisms, processes, and conferences. These are not abstract notes or concepts; they refer to real elements of scientific practice and literature. This anchoring keeps the graph tied to empirical referents rather than purely personal abstractions.
User-created links as traces of understanding
Links are created deliberately by the user. For example, linking a paper to a gene because the paper studies it, linking a researcher to a process because they work on it, or linking a conference to a topic because it was discussed there. These links are not imported assertions from databases. They reflect what the user noticed, interpreted, or considered relevant at a given time. This gives the graph a trace of epistemic decisions rather than a catalogue of objective claims.
Notes attached to entities as contextual records
Notes are not standalone documents. They are attached to entities and capture context such as what was learned during a seminar, why a link was created, or what interpretation was made at that moment. Together, entities, links, and notes form a structured trace of how understanding evolved. Over time, the graph becomes a record of cognitive history, not just a snapshot of knowledge.
What This Is Not
Not a knowledge graph
Knowledge graphs aim to represent relationships that are considered true independently of any individual. In Pilus, links represent what a specific researcher chose to connect. The graph is therefore personal, situated, and temporally contingent.
Not an electronic lab notebook
Lab notebooks record experiments and results in chronological order. Pilus records conceptual connections across entities, independently of experimental timelines. It captures how understanding is organized, not only what was done.
Not a personal knowledge manager
Tools like Obsidian or Roam allow free linking between notes. In Pilus, links occur between predefined scientific entity types, which keeps the structure interpretable in scientific terms and reduces ambiguity about what a link signifies.
Not a network visualization tool
Visualization tools display existing networks. In Pilus, the network is built progressively by the user as part of their learning process. The visualization is secondary; the underlying structure is the method.
Practical Implications
This approach enables keeping track of how different scientific elements became connected in the researcher's mind, revisiting past notes in the context of the entities they concern, seeing which areas of a topic are densely connected and which remain isolated, and making one's understanding more explicit and easier to revisit after months or years. Because links are authored, the graph can also highlight uncertainty, revision, and shifts in interpretation as the researcher updates or removes connections. The structure itself can act as a scaffold for future reading and discussion.
How This Could Be Evaluated
This model could be evaluated by observing whether users recall relationships between entities more easily after several weeks, identify new potential connections when viewing their graph, reconstruct the context in which they learned specific information, and share their graph with collaborators to explain their understanding of a topic. Additional evaluations could test whether the method improves hypothesis generation by revealing under-linked areas or bridges between clusters, and whether it reduces redundancy in reading by making prior conceptual paths visible.
Conclusion
Pilus implements a simple methodological idea. It represents scientific entities explicitly and treats the links created by the user as records of how understanding is built over time. Conferences, articles, and discussions generate ideas. Pilus connects them before you forget by turning those links into a structured, revisitable trace of learning.