01Why the elaborate systems fail.
Because the effort is paid now and the benefit arrives in eight months, and no amount of resolve reliably bridges that gap during a bad week. A system with tags, ratings, summaries and cross-links is genuinely better in principle and is functionally worthless in practice, because it is not the system you will use at eleven at night after a shift.
The failure is also invisible while it happens. Nobody decides to stop; entries simply get thinner, then intermittent, then absent, and by the time you notice, the record covers October and stops. A partial record is worse than none, because you trust it and it is incomplete.
So the design constraint is not completeness. It is that a tired person will still do it, which puts a hard ceiling of about a minute on any single entry.
02The four fields.
This is the whole system, and it fits in one line per source in any document, spreadsheet or reference tool you already have.
- The full citation, in the style your programme uses, so that it never has to be rebuilt from a half-remembered title.
- Where you got it and when. A database name and a date, which takes four seconds and is what makes a retraction check possible later.
- One sentence on what it says, in your words rather than the abstract's. This is the field that does the real work.
- One sentence on why you kept it, which is different from what it says and is the field everybody omits. What claim of yours it might support, or which argument it complicates.
- That fourth field is the reason the whole thing works, because in eight months you will remember topics and not sources, and it is the only field that answers the question you will actually be asking.
03The question you will ask in eight months.
Never what did that paper say. Always something like: what did I read that supported this point, or what was the thing that contradicted the assumption I have been making. Those are queries about your own argument, not about the literature.
A record organised by source cannot answer them, which is why alphabetical bibliographies are useless for the purpose people build them for. A record where every entry says why it was kept can be searched for the claim, and that turns a list into an instrument.
It also makes writing dramatically faster at the point where it matters. Somebody assembling a review from a record with that fourth field is arranging arguments they already identified; somebody working from a stack of files is rereading forty papers to remember which was which, which is the cost reading properly the first time was meant to avoid.
04Where it lives.
Anywhere consistent. A reference manager is genuinely useful, particularly for generating a bibliography, and it is not required. A single document with four fields per entry works, and so does a spreadsheet, which has the advantage of being sortable.
What does not work is distribution: some in a manager, some in a notebook, some in file names, some in the margins of printouts. The retrieval cost of a scattered record rises fast enough that people stop consulting it, and at that point the recording was pure loss.
Where a reference manager is used, one caution worth knowing: it will format faithfully to whatever style setting it holds, uniformly, so a wrong setting produces a whole bibliography that is wrong in the same way and therefore looks consistent. Check its output against your programme's edition rather than assuming, which is the same trap described from another angle in the feedback issue, where the same deduction keeps reappearing.
05Starting mid-degree.
Do not go back. The instinct on adopting a system is to reconstruct everything already read, which takes days and is nearly always abandoned partway, leaving exactly the partial record this issue warns about.
Start with the next thing you read. The record becomes genuinely useful once it holds a term's worth, and reconstructing the past adds material you have already used while costing the time that would have built the future.
The one exception is anything currently load-bearing for a project in progress. Those few sources are worth entering properly today, because they are the ones you will need to find again, and they are the ones a supervisor or a marker may ask about. Everything else can start from now. It costs a minute per source and it is the single highest-return minute in a research week.
Questions at this desk.
Do I need a reference manager?
No, though one is useful for generating bibliographies and for handling large numbers of sources. What matters far more is that everything is in one place and that each entry says why you kept it. A spreadsheet with four columns outperforms an elaborate manager used inconsistently.
How much should I write per source?
Two sentences and two facts, which is about a minute. Anything longer is a note rather than a record, and notes belong wherever your notes live. The record exists to help you find things again; keeping the two functions separate is what stops either from being abandoned.
What about sources I decided not to use?
Keep those too, with one line saying why you rejected them. It stops you opening the same paper a second time in three weeks, and where a marker asks whether you considered something, having a dated line explaining why you set it aside is a far better answer than a vague recollection.
Is this necessary for undergraduate work?
Less critical, and still worth about a minute per source, because the habit is what carries into anything longer. For a dissertation or a doctoral project it stops being optional: you will be asked where things came from, and reconstructing a search from memory two years later is not realistic.
Notes in the margin, from the people at the desks.
READ THE LETTER