Master Answer WritingSign in

How the marking works

An AI examiner, calibrated to the real one

'AI evaluation' usually means a language model told to be encouraging. Ours is told the truth instead: that the All-India topper scored 86/250, that 5/10 is a good answer, and that its job is to mark like the examiner you will actually face.

Under the hood, every mark is awarded against a frozen yardstick memo with pre-scored anchor answers — the same calibration a real examining board gives its markers. Same question, same standard, every time.

50 free credits on signup · No card required

Fact-checked feedback, or no feedback

The fastest way an AI evaluator loses your trust is inventing a 'missing point'. Every missing-value item we show is verified against a reference before display; anything unverifiable is dropped. You will never revise from a hallucination — and awards above 65% need a second confirmation pass before you see them.

Honest failure, automatic refunds

If an evaluation fails or takes longer than our stated limit, the credits refund themselves — you do not ask, and there is no support ticket. Illegible pages are blocked before any credit is spent. The ledger behind your balance is append-only and visible to you, entry by entry.

How strictly we mark

5 out of 10 is a good day. That is the point.

Bands are anchored to real outcomes — the All-India topper scored 86/250 in a GS paper. A platform that hands you 8/10 is selling comfort and costing you a year.

  • 70%+ExcellentRare. Topper territory.
  • 55–70%GoodInterview-call standard.
  • 40–55%AverageWhere most good answers land.
  • 25–40%WeakKnows the facts, misses the argument.

Questions asked straight

Frequently asked

Which AI does the marking?

Frontier Claude models from Anthropic, with the marking model pinned and monitored for reproducibility — the same copy re-marked should score within half a mark. Cheaper models are used only for things that cannot change your score, like illustrative model answers.

Can an AI really judge handwriting and diagrams?

It reads your pages as images, so margins, headings, underlining and diagrams are part of what it judges — not a flattened transcription. Genuinely illegible passages are marked as unread rather than guessed at, exactly as an examiner would.

Why should I trust a strict mark more than a kind one?

Because the kind one costs you a year. A platform that inflates your marks is optimising for your renewal, not your rank. Calibrated strictness means the number tracks the real exam — which is the only reason to check answers at all.

Write. Check. Improve.