How it works

There is no lobby to browse. You press one button and the server finds someone, starting with players rated within 50 points of you and widening the search the longer you wait — 100 after three seconds, 200 after six, 400 after ten, and eventually anyone at all.

If nobody suitable has appeared after fifteen seconds you are offered a practice bot instead, clearly labelled and unrated. You are never told an opponent is human when it is not, and you are never left watching a spinner with no way out.

A statement-based Prelims question is several sentences of reading before recall even begins, so a single fixed clock measures reading speed rather than knowledge. You pick the pace you want to be tested at:

• Brisk — quick decisions, six questions.
• Standard — the default, four questions.
• Steady — room to read, three questions.
• Exam pace — closest to the real paper, three questions and a longer round.

Each pace sets a baseline for a medium question, and difficulty scales it: an easy question gets less, a hard one more. So at Standard an easy question is 27 seconds and a hard one is 67. Slower paces hold fewer questions, so a round takes roughly the same few minutes whichever you choose.

You can accept more than one pace at a time, and most people should: accepting more finds an opponent sooner, at the cost of controlling which pace you get. Where two players overlap on several, whoever has been waiting longer gets their first choice.

Exactly as UPSC marks Prelims, with no variation and no bonus for speed:

• Correct — +2
• Wrong — −0.67 (one third of two marks)
• Skipped, or no answer — 0

Answering faster earns nothing. Rewarding speed would train fast guessing, which is the single habit that destroys a Prelims score under negative marking — so speed is recorded and shown to you, and it decides nothing.

Skip is a full-sized button beside Submit, not a small link. Above roughly one answer in four, attempting is worth it; below that, a skip is the better decision even though it scores nothing. Knowing which is which is a real skill and one of the clearest separators between a middling score and a good one.

The server owns it. Your countdown is drawn from a deadline the server sets and is corrected for your device’s clock, so a skewed clock cannot cost you time. A short grace absorbs network jitter, and a question closes early the moment both players have answered — there is no waiting out a timer on a decision you have already made.

Both players see the same questions in the same order. Neither of you can see the answer key before the question closes; it is not sent to the browser at all until then.

Chosen after you are matched, from the difficulty band that suits the average of your two ratings, and spread across easy, medium and hard rather than being all of one kind. Anything either of you has seen recently is excluded — fairness requires that neither player has an advantage.

The question bank is finite and growing. If it cannot fill a round without relaxing one of those rules, it relaxes them in a fixed order rather than serving you a silently worse round.

The higher total wins the round. If the marks are level, the round is a draw — it is not broken on who answered faster, because that would put a rated result on exactly the thing the marking refuses to reward.

A round is the unit that counts. Rating settles at the end of each one rather than at the end of a session, so leaving after a round costs nothing and there is never an advantage in quitting while ahead.

Ratings use Glicko-2, the system Lichess and Chess.com are built on. Everyone starts at 1200. What you gain or lose depends on who you played: beating a much stronger player moves you a long way, losing to a much weaker one moves you a long way in the other direction, and an even result between close players moves almost nothing.

A draw is priced the same way. Against someone far below you it costs rating; against someone far above you it gains rating. It is never a free result.

Alongside the rating the system tracks how confident it is about you. Early on it is unsure, so your first rounds move you further. Your first ten rated rounds are marked as calibration and no rank is shown until they are done — a rank assigned on two rounds would be a guess presented as a fact.

There are thirty ranks. They exist so that progress is visible on the scale of a week rather than a season, and your rank never falls more than one below the highest you have reached — though the underlying number is always shown to you, unrounded and unprotected, because you can work it out from your own results anyway.

The review opens by itself. It shows every question, what you answered, what your opponent answered, the official key, and an explanation where one has been reviewed by a human.

It also judges the decision, separately from whether you got it right — a good skip, a timid skip in a topic you usually get right, a risky attempt where skipping was worth more. Those verdicts come from your own record in that topic, not from an opinion about the question.

Everything is kept. Your match archive, your accuracy by subject, and your rating over time are all on your profile. With Pro, the questions you got wrong or bookmarked come back in Revision.

No assistance of any kind during a rated duel. No lives, no coins for sale, no way to pay for an advantage, and no paying to remove a penalty. Rated duels are free and unlimited, and nothing about the ladder is for sale.

Open CollectorX and press Find an opponent.