You have run out of past papers. Now what?

Every senior student hits the same wall around six weeks out. The official past papers are finished — and worse, half-memorised, so redoing them tests recall of last week rather than the subject. Trial papers get swapped around and finished too. Textbook exercises drill one skill at a time, which is not what an exam does. What you actually need is an endless supply of questions in the exact shape of your paper, on the exact topic you are weakest at.

That is what AI should be perfect at. Mostly it isn't, and the reason is worth understanding.

Why generic chatbots write bad exam questions

A general-purpose model has read a lot of chemistry. It has read comparatively little of your syllabus's assessment style. So it writes textbook questions with an exam wrapper. The failure is consistent and easy to spot:

Generic chatbot questionReal exam question
Structure One question, one answer Multi-part — (a) sets up, (b) applies, (c) evaluates, each building on the last
Marks None, or invented Allocated per part, and the allocation tells you how much working is expected
Command verbs "Explain" for everything Explain / justify / evaluate / account for — each demanding a different answer shape
Difficulty Uniformly mid Ramps within the paper; the last parts separate the top band
Solution Printed immediately, under the question Mark scheme, released after you sit it

That last row is the quiet one. A question with its answer visible underneath is not practice — it is reading. The moment your eye catches the working, you switch from recall to recognition, which is exactly the mode that makes you blank in the exam hall.

What a past-paper-trained model does differently

errorboard's practice chat is built on a model trained on real past papers paired with their mark schemes — question papers matched to the marking guide, question by question, across the courses below. Trained on the pairing, not just the questions, so it learns what a 3-mark answer contains and not merely what the topic is.

In practice that means when you type "give me a question on equilibrium and Le Chatelier", what comes back is shaped like the paper:

A sealed vessel contains the equilibrium N₂O₄(g) ⇌ 2NO₂(g), ΔH = +57 kJ mol⁻¹.

(a) The volume of the vessel is halved at constant temperature. Explain the effect on the amount of NO₂ present at equilibrium. (3 marks)

(b) The temperature is then increased. Justify the change in the colour of the mixture. (2 marks)

(c) Explain why adding a catalyst changes neither answer above. (2 marks)

▸ Show solution

Three parts, marks per part, real command verbs, a ramp — and the worked solution collapsed behind a toggle you open only after you've committed to an answer. On Pro, a 1–5 difficulty dial sets how hard the ramp goes, so the same topic can be a warm-up or a band-6 stretch.

Courses covered

Practice chat runs on three senior curricula:

  • HSC (NSW, Australia) — Mathematics Advanced, Mathematics Standard, Physics, Chemistry.
  • VCE (VIC, Australia) — Mathematical Methods, Specialist Mathematics, General Mathematics, Chemistry, Physics, Biology, Psychology, Economics, Business Management, Legal Studies, Health and Human Development, Accounting.
  • A-Levels (UK, AQA) — Mathematics, Physics, Chemistry, Biology.

No English, deliberately: extended-response marking is band descriptors rather than a single correct answer, so there is nothing honest to train a question generator on.

The loop worth running

Unlimited questions are only useful inside a routine. The one that pays:

  • Pick the topic from your error log, not your mood. Students practise what they enjoy, which is what they are already good at. Your error log is the honest list of what costs you marks — start there.
  • Attempt on paper before opening the solution. Full working, as if it were the exam. The toggle exists so you can be disciplined about this; nothing enforces it but you.
  • Mark yourself against the solution, harshly. Half marks for "I knew that really" are how people walk into exams overconfident.
  • Log what you got wrong. A missed practice question is worth exactly as much as a missed past-paper question — it goes in the log with a one-line diagnosis, and comes back around. Snap a photo of your working and errorboard does the diagnosis and filing for you.
  • Re-generate the same topic three days later. Different numbers, different framing, same skill. If you get it clean cold, the gap is closed. That spacing is the entire mechanism.
Why the loop beats more papers. A past paper tests forty things, of which perhaps five are your weaknesses — you spend three hours to drill five. Generated practice targeted from your error log is all five, for twenty minutes, repeatedly.

How practice fits the rest of errorboard

errorboard's job is the clerical half of studying: you make a mistake, it captures it, diagnoses why, categorises it and files it into a per-topic error log. Practice chat closes the loop at the other end — the log tells you what to practise, practice makes fresh questions on exactly that, and anything you get wrong goes straight back into the log. The quiz then re-surfaces old mistakes on a spacing schedule so nothing quietly rots.

It runs in the browser on any device — laptop between classes, phone on the bus, iPad alongside the handwriting notebook so you can write the answer next to the question.

Common questions

Is it free? There's a free daily allowance of practice questions, reset at midnight UTC, no card needed. Pro removes the cap and adds the 1–5 difficulty dial plus marking on work you upload.

Are the questions real past-paper questions? No — they're new questions generated in the style the model learned from past papers and mark schemes. That's the point: real papers run out, and once you've seen one you can't un-see it.

Can it mark my handwritten working? Yes. Photograph your working and errorboard marks it against the solution, then logs whatever went wrong — same pipeline that handles a photo of your paper working from any phone.

Will it just give me the answer? Not unless you open the toggle. That single design choice is most of the difference between practice and reading.

Start with your worst topic

Not your favourite one. Open practice chat, name the topic you have been avoiding, and do one question properly — on paper, solution closed, marked honestly. That single question tells you more about your exam readiness than an afternoon of re-reading notes.