Why “the Last 500”? Recency Rings, Explained Honestly
Every ATPL student is eventually told to “do the last 500”. Almost nobody is told what the last 500 is, who decided which 500, or when the advice stops working. It is a genuinely good instinct wrapped around a number that is doing far less work than people assume — and the cases where it fails are predictable, which means they can be planned around. This is how recency actually works, including the maths we use and the three situations where we would tell you to ignore it.
Last updated
What “last 500” actually refers to
There is no official last 500. EASA does not publish one, and neither does any national authority. Every “L500” you have been recommended was built by a provider out of student feedback, which means the number describes that provider’s reporting pool, not the exam. Two providers’ L500 sets for the same subject can differ a great deal and both be honestly produced.
Ours works like this. Every time a student tells us which of our own published questions looked familiar after their exam, that becomes a dated, authority-tagged signal. Each question accumulates a recency score, questions are ranked within their authority and subject, and the top 100, 300, 500, 800 form nested rings — L100 sits inside L300, which sits inside L500.
The decay, in plain numbers
A report from last month should count for more than one from last year, but not infinitely more, and the drop-off should be smooth rather than a cliff at some arbitrary cut-off. So each sighting is weighted by exp(−age ÷ 240), giving a half-life of about 166 days. The reasoning for that length is simply that the ECQB refreshes on roughly an annual cycle: a sighting from six months ago is still substantially informative, one from two years ago is mostly noise.
| A student who… | Days ago | Counts for |
|---|---|---|
| Sat last week | 7 | 0.97 |
| Sat 3 months ago | 90 | 0.69 |
| Sat 166 days ago | 166 | 0.50 |
| Sat a year ago | 365 | 0.22 |
Two more weights sit on top. How sure the student was — a question that looked similar counts 0.4 where an identical one counts 1.4 — and how reliable that reporter has proven, which is the next section. The whole table is rebuilt from scratch every night at 02:40 UTC, nightly, so nothing lingers.
Why we have to weight the reporter, not just the report
Recognition is a soft signal. Ask anyone which of twenty questions they saw in an exam they sat under pressure and some will tick generously, some will tick nothing, and a few will tick everything on the way past. Averaging that produces confident nonsense.
So every grid contains five questions the student cannot have seen. They are not marked, and the grid never reveals which they are. Recognising none of them is a clean report. Recognising them as readily as the real candidates is indistinguishable from guessing, and those sightings are discarded outright rather than counted weakly — three unreliable reporters must not clear a three-reporter threshold between them.
One honest caveat, because it is the case that applies while the rings are still filling: when a grid has no high-recency candidates to compare against, the controls have nothing to measure, so that report is given a neutral weight rather than a computed one. It would be tidier to claim unreliable reporters are always down-weighted. It would not be true.
Three times studying by ring will let you down
Everyone selling recency has an incentive to tell you it always works. It does not, and the cases where it fails are predictable enough to plan around.
Straight after an edition change
This is the big one, and it is happening right now. When your authority moves to a new ECQB edition, every sighting collected before the switch describes the old bank. The maths does not know an edition changed — it only knows those reports are getting older, and at a 166-day half-life they stay influential for months after they stopped being true. A ring is at its least trustworthy in the weeks after your authority migrates, which is exactly when students reach for it hardest.
A low-volume authority
Recency is computed per authority, because it has to be — students sitting the same edition at different authorities report very different experiences. But that splits the reporting pool. An authority with few candidates in your subject may never reach the thresholds, and we would rather publish nothing than publish a ring built on two people.
An under-reported subject
Reporting is voluntary and uneven. Air Law and Meteorology get reported far more than the smaller subjects simply because more people sit them. A thin subject can sit below the threshold indefinitely while its neighbours are well covered, so "there is no ring here" often means "not enough people told us", not "nothing repeats".
The first of those is worth reading next to which ECQB version your exam uses. A ring means something quite different either side of an edition change, and because authorities migrate on their own timetables, your ring and another student’s can be describing two different question banks on the same day.
What we never collect, and why that limit is the point
We never ask what was on your paper, and there is nowhere to type it. The prompt shows questions from our own published bank and asks only which ones look familiar; what gets stored is a pointer to our own content, plus the authority and the date. ECQB questions are confidential and candidates sign a non-disclosure undertaking. A recognition signal over our own material is a different thing entirely — and it is the only version of this that can honestly exist, which is why we would rather have a slower-growing ring than a set of recalled questions.
Common questions
What does "the last 500" actually mean?
In community shorthand — L800, L500, L300, L100 — it means the questions most recently reported as seen in real exams, cut at the top N. It is not a list published by EASA or by any authority; there is no official "last 500". Every provider that offers one has built it from student feedback, which means the number describes their reporting pool rather than the exam. Two providers' L500 sets for the same subject can differ substantially and both be honestly produced.
How do you decide which questions are in a ring?
Each question gets a recency score for each authority: every report of it decays exponentially with a 240-day time constant, is weighted by how confident the student was, and is weighted again by how reliable that reporter has proven. Questions are then ranked within their authority and subject, and the top 100, 300, 500 and 800 form the nested rings. A question needs at least 3 independent reporters before it is eligible at all.
Why 166 days and not something shorter?
The decay constant is 240 days, which works out to a half-life of about 166. The reasoning is that the ECQB refreshes on roughly an annual cycle, so a sighting from six months ago is still substantially informative while one from two years ago is mostly noise. A much shorter half-life would make the rings twitchy — they would swing on whoever happened to report last month — and a much longer one would keep describing a bank that has moved on.
How do you stop people from just ticking everything?
Every recognition grid contains five questions the student cannot have seen. They are not marked, and the grid never tells anyone which they are. Recognising none of them is a clean report; recognising them as readily as the real candidates is indistinguishable from guessing, and that reporter's sightings are discarded outright rather than counted weakly — three unreliable reporters must not clear a three-reporter threshold between them. The honest exception: when a grid had no high-recency candidates to compare against, the controls could not measure anything, so that report is given a neutral weight rather than a computed one.
Is studying the last 500 enough to pass?
No, and anyone selling you that is selling you a risk. A ring tells you where reported repetition is concentrated; it does not cover the syllabus, it says nothing about the questions nobody happened to report, and it degrades exactly when an edition changes. It is a sensible thing to do last, over a bank you already know. It is a bad thing to do instead of learning the subject — which is why the students who post that they "did L300 three times" and still met an exam full of unfamiliar wording are not unlucky, they are describing the predictable failure case.
Do you store the questions people saw in their exam?
No. We never ask what was on your paper, and there is nowhere to type it — the prompt shows questions from our own published bank and asks only which look familiar. What is stored is a pointer to our own content plus the authority and date. ECQB questions are confidential and candidates sign a non-disclosure undertaking; a recognition signal over our own material is a different thing entirely, and it is the only version of this that can honestly exist.