How does a research group keep working knowledge when students graduate?

What actually leaves when a student graduates?

Not the thesis. The reasoning that produced it.

The written outputs stay: papers, the dissertation, the repository, the protocol in the shared drive. What goes is the layer underneath. Why that parameter and not the obvious one. Which of the two failure modes you are looking at when the run dies at hour six. The three approaches tried in the first year that did not work, and the specific reason each was abandoned. Who at which institution actually answers email.

The knowledge management literature has a precise distinction for this. Explicit knowledge is codified and transfers relatively easily; tacit knowledge is personal, bound to experience and context, and covers intuition, problem-solving habits, and relationship networks. The consistent finding across that literature is that tacit loss does more damage than explicit loss, because explicit knowledge stays in the organisation as documents while tacit knowledge walks out with the person.

A research group is an unusually severe case of this, because the turnover is scheduled. In most organisations departure is an event; in a lab it is the calendar. The group loses a substantial share of its working expertise on a known cycle and then rebuilds it, and each cohort re-derives things the last cohort already learned.

Why does documentation not solve this?

Because it captures the procedure and drops the judgement, and because nobody reads it at the moment they need it.

Studies of research settings describe both halves of the problem. A great deal of what a group knows is never written down at all and lives in the memory of the supervisor and the current members. Where central documents do exist, students routinely struggle to read them end to end or to locate the specific thing they need at the moment they need it.

There is also a structural reason documentation under-performs here. Documentation is written in the order the writer understands the work. Questions arrive in the order the reader is stuck. A protocol answers "how do I run this"; the new student's actual question is "it failed like this, is that normal", and that question is answered nowhere in the document because it was obvious to whoever wrote it.

None of this is an argument against writing things down. It is an argument that a written record is a necessary artefact and an insufficient one, and that the gap is specifically the reasoning a document does not think to include.

When should we capture it?

In the middle of the work, not in the last fortnight.

The exit interview in the final week is the standard approach and it is close to the worst possible timing. The person is finishing a thesis, already partly gone, and being asked to summarise several years of accumulated judgement into a document nobody has time to review. What comes back is a tidy account of the final method, which is the part that was already written down.

Capture works better when it happens while the work is live and the reasoning is still contested. A recorded conversation in the second year, when a student can still tell you why they rejected the alternative, holds more than a summary written after they stopped caring about the alternative.

A workable cadence for most groups:

• At the halfway point of each member's time, one recorded conversation about decisions and dead ends rather than results.
• When something fails interestingly, a short note on what it looked like and what it turned out to be. These are the notes with the highest value per minute and they are almost never written.
• At handover, a conversation between the person leaving and the person arriving, recorded, rather than a document written alone.

What does a usable handover contain?

Five things, in rough order of how often they are missing.

• Decisions and their reasons. What was chosen, what was rejected, and why. This is the layer that disappears most completely and is most expensive to re-derive.
• Failure signatures. What going wrong looks like, and which distinct cause each appearance points to. New members lose weeks to failures the previous cohort could have named in a sentence.
• The state of the work. What is finished, what is provisional, what is known to be wrong but not yet fixed, and what is held together by something temporary.
• The people map. Who to ask about what, inside and outside the group, and which relationships are personal rather than institutional.
• The unwritten conventions. Naming, storage, the order things are done in, the reason the folder structure is the way it is.

The test is not whether a document exists. It is whether a person arriving eighteen months later can ask a specific question and get the answer the departed member would have given, without needing to know which document to open.

How does Knowledge Transfer work here?

An interview produces the material, the material becomes sources, and the sources answer questions.

The working shape is deliberately unglamorous. Existing material is gathered first: papers, protocols, notes, repositories, slide decks, whatever already exists. Then a structured interview covers the layer that was never written down, aimed at decisions, dead ends and failure signatures rather than at restating the method. Both go in as approved sources.

What sits on top is a Mentor the group can ask in the way a new member would ask a person. When it answers, it names the source and the passage it drew on, so an answer can be checked against the material rather than trusted. When it has nothing to answer from, it says so, which matters more here than in most settings: a system that fills the gaps in a departed student's reasoning with a plausible guess is actively worse than an empty folder.

Ownership stays with the group. The sources are exportable, the interview recordings are the group's own material, and the knowledge map and handover deliverables are documents you keep whether or not the relationship continues.

What if we have no budget for this?

Then do the part that costs nothing, because it is also the part that matters most.

Record one conversation with each member at the midpoint of their time, ask about decisions rather than results, and keep the audio somewhere the group can reach. That single habit captures more of what is at risk than most formal documentation efforts, and its cost is an hour per person per year.

Two additions, both free. Keep a running file of failure signatures, one line each, written at the moment of the failure rather than afterwards. And when someone leaves, make the handover a recorded conversation with their successor instead of a document written alone, because the successor's questions are the ones a document never anticipates.

If that material later becomes something more organised, it will be built from recordings that already exist. If it does not, the group still has the recordings, and they are the irreplaceable part.