PYQGraph
Sign in

Method

How an answer earns its label.

Six steps between a printed paper and a question on this site. Anything that fails a step stays off the site until it passes.

A fountain pen making a tick in the margin of a printed page.
  1. Start from the official paper.

    Every question starts from the commission’s own paper.

    Every question begins as the commission’s own question paper, exactly as candidates sat it — not a coaching institute’s retyped version of it.

  2. Read it, then read it again.

    Each scanned page is read, then checked by an independent second reading.

    Scanned pages are read by a model, then checked against a second, independent reading of the same page. Any question where the two readings disagree is held back until it is fixed.

  3. Match it to the commission’s key.

    Where the commission published a key, that is the answer.

    Where the commission published a key, that is the answer. We record whether it is the final key or the exam-day key, and which booklet series it belongs to — a detail that, if missed, silently shifts every answer on the paper.

  4. Where we hold no key, solve it twice.

    No key? Answered blind, and labelled for exactly what it is.

    A model answers the question blind — it never sees a key. If a second model, from a different family, independently agrees, the answer carries the Two models, agreed label. If only one high-confidence answer exists so far, it is shown as Our answer, unofficial, without an explanation, until the second check agrees — or it is withdrawn if it does not.

  5. Check the explanation.

    An explanation is shown only when its answer has been confirmed.

    An explanation is published only when the answer it argues for either matched the commission’s key — reached blind, without seeing the key — or was confirmed by a second model. Where neither has happened yet, the question shows its answer and no explanation.

  6. Hold back whatever has not settled.

    Anything that has not settled stays off the site.

    If the readings disagree, the answers disagree, or a check fails, the question stays off the site. Holding back is the resting state; publishing is earned.

i.The labels

What each label means, exactly.

Commission’s final keyPublished by the commission after candidates’ objections were heard and decided. Where it exists, nothing overrides it — including us.
Commission’s exam-day keyPublished by the commission on or near exam day, before objections. Some answers change between the exam-day key and the final one. If a final key appears, it replaces this one.
Two models, agreedThe commission never published a key, or we have not found one. Two models solved the question separately and agreed. This is our answer, not the commission’s, and it is never styled as one.
Our answer, unofficialNo commission key and no second opinion yet: one solver answered with high confidence. Shown so the question is usable, labelled so it is never mistaken for a key, and its explanation is withheld until a second check agrees — or the answer is withdrawn if it disagrees.
Question deletedThe commission struck the question from the paper. We show that, instead of inventing an answer for it.
ii.What we do not claim

We would rather tell you less, and have all of it be true.

We do not claim a person has checked every question. We tell you exactly what did check it, on the question itself.

We do not publish an accuracy percentage. The only papers we could measure one against are also the most widely republished, which means a model may simply remember them. A number measured that way would flatter us and mislead you.

We do not say a topic “will come”. We can say how often it has come, and where.

Found something wrong? Write to hello@pyqgraph.in. Corrections are made on the question and dated.