The most common question after "what do I write" is "how long do I need to do this for."
The unsatisfying answer is that longer is better, indefinitely, and there is no point at which the record stops improving. The useful answer is that a long record is not simply a bigger version of a short one. It answers different questions, and this guide is about which ones.
Three weeks tells you what
Twenty-one entries is not nothing. Kept properly, it establishes what your symptoms actually are, in your own words: how they present, how long they run, what they interrupt.
That is real, and if three weeks is what you have, three weeks is what you bring. Do not let anyone talk you out of starting because it is too late for it to be impressive.
But three weeks cannot tell anybody how often this happens, and frequency is the question almost everything else depends on. Four bad days in three weeks might be a terrible month or a good one. There is nothing in the record to say which.
A year tells you how often, whether it is changing, and what sets it off
Frequency stops being a guess. Ninety-six bad days out of 365 daily entries is a number that came from counting. "Two or three times a week, I think" is an estimate, and estimates are the thing a log exists to replace. This is the single largest difference and it arrives around the point where you have several months of continuous entries.
Direction becomes visible. Whether something is stable, gradually worsening, or improving is invisible over three weeks and obvious over a year. That matters for reasons well beyond paperwork — it is a genuinely useful thing to be able to show a doctor, and it is a thing almost nobody can answer accurately from memory.
Seasonality shows up. Winter, heat, the anniversary of something. Patterns on that scale simply cannot appear in a short record, and several of them are common enough to be worth knowing about.
It stops looking situational. This is the one people do not anticipate. A short record kept during a rough patch is compatible with a rough patch. A year of records that covers a house move, a job change, two holidays and a good stretch in the spring, and still shows the same pattern running through all of it, says something a three-week record structurally cannot.
Gaps stop being fatal. In a three-week log a missing week is a third of the document. In a year it is a fortnight in a long series that continued before and after. Consistency over a long stretch buys tolerance for the occasional lapse.
The one thing you cannot buy back
Everything else on your claim can be improved later. You can write better entries starting tonight. You can reorganise, consolidate, get the format right, add the functional-impact line you have been leaving out.
You cannot add last March.
A log started today has today as its earliest date, permanently, and every month you wait moves that date forward. That is the whole argument for starting tonight rather than at the weekend, and it is the reason the honest answer to "should I bother, my claim might be a year out" is yes, emphatically, because a year out is exactly when a year of records exists.
Do not restart
The corollary, and it comes up constantly.
Somebody keeps a patchy log for eight months, reads a guide like this one, realises their entries have been thin and inconsistent, and decides to start fresh and do it properly.
Do not. You would be trading eight months of imperfect evidence for zero months of good intentions. The patchy record still shows dates, still shows a rough frequency, and still shows this has been going on since well before the claim.
Keep everything. Improve from today. In six months you will have eight months of thin records followed by six months of good ones, which is a better document than six months of good ones alone, and far better than what you would have had if you had started over.
What a year actually costs
Worth being concrete, because "keep a log for a year" sounds enormous and is not.
At ten to sixty seconds an entry, a year is somewhere between one and six hours in total, spread across 365 nights in pieces too small to feel. The hard part is not the time. It is remembering to do it on the days when nothing happened, which is precisely why the ten-second entry matters so much.
If you are choosing between doing this well for a month and adequately for a year, choose adequately for a year. The long thin record wins.
If you have less than a year
Bring what you have, and say plainly how long it covers. A summary line at the front does that in one sentence:
Records from 3 June 2026 to 28 August 2026, 87 daily entries.
Nobody is confused by a short record honestly labelled. What causes trouble is a short record presented as though it were longer, or one whose date range you have to reconstruct by reading it.
Then keep going. The record you hand over is not the last version of it — claims take time, and the log that gets looked at later is longer than the one you started with.
The short version
Three weeks shows what. A year shows how often, whether it is changing, and that it is not just a rough patch. The elapsed time is the one thing you cannot add later, so start tonight rather than at the weekend, never restart a patchy log, and prefer a thin record kept for a year over a thorough one kept for a month.