How Backsolved works

The case itself

Each case is a written interview. There is a named interviewer with a house style, a company with a real problem, and a set of facts held back until you ask for them. You type; the interviewer answers, pushes back, and asks you for a recommendation at the end. It takes about 30–40 minutes and you can stop and resume.

The interviewer will not do your arithmetic for you and will not volunteer the answer. If you ask a vague question you get a vague answer. If you ask for the cost breakdown by segment, you get the cost breakdown by segment — and nothing else. That constraint is deliberate: it is the difference between practising hypothesis-driven inquiry and practising “ask for everything and hope”.

The answer key

Before you ever open a case, that case already has a fixed key: a written account of what is actually going on, the specific figures that matter, and a tolerance for each one — how close you have to be for a citation to count as right.

This is the part that makes the report worth reading. The key does not change between candidates, it does not change between your first attempt and your third, and it is not generated in the moment alongside your reasoning. It sits on our server, where you cannot read it and neither can the interviewer’s conversational model until the case is over.

What actually gets checked

Three things, all by ordinary code rather than by judgment:

• Equations you typed. If you write “450 − 380 = 70”, that is recomputed. If you write “(500 − 450) / 500 = 1.0%”, the two-step version is recomputed too. Being wrong by 3% is treated as rounding; being wrong by more is treated as wrong.
• Figures you cited. When you state a number near a fact the case has an answer for — the default rate, the breakeven volume, the synergy value — that number is matched to the key within the tolerance set for it.
• Numbers you contradicted. If you give one value for something on turn 6 and a materially different value on turn 14, both are shown to you side by side.

When a number is wrong, the report tries to say how it is wrong. A miss that is a clean factor of ten, a hundred or a thousand is almost never a reasoning failure — it is a misplaced decimal, a percentage written as a fraction, or thousands confused with millions. Those get named as such, because “off by 100×, likely a percentage-as-fraction mix-up” is a fixable habit and “incorrect” is not.

What the report contains

• Scores on four dimensions — structure, quantitative logic, business intuition, and synthesis — each with the quotes from your own transcript that produced them.
• A framework checklist: which of the case’s drivers you reached, and on which turn.
• The full arithmetic table, every check, right and wrong.
• Pacing: when you stated a structure, when you first engaged with the numbers, how long the whole thing took.
• The ground truth, revealed in full once you are finished — the real root cause and the recommendation the case was built around.
• Your transcript, with an alternative phrasing offered turn by turn.

What it does not do