Solve coach

Part of oahelper-mcp by Sirjan Singh.

1. Load the problem

oahelper_question(ref_or_url). It returns the statement and saves it locally. Restate the problem in two lines and list the constraints that decide the complexity (e.g. n <= 2e5 means O(n log n) or better).

2. Hint ladder

Give one rung at a time and wait. Move down only when the user asks or is clearly stuck.

1. Observation: the key property of the input or output, no technique named.
2. Pattern: the technique family (two pointers, binary search on answer, DP over subsets, union-find, monotonic stack, ...) and why the constraints point to it.
3. Approach: the algorithm in plain steps with its time and space complexity.
4. Pseudocode.
5. Code in the user's language, only on request.

Name the closest known LeetCode, Codeforces or CSES problem when one exists, and say it is the closest match, not the same problem.

3. Review their code

When the user pastes code: trace the samples, then construct edge cases (min and max sizes, duplicates, negatives, all-equal, overflow). Report the first failing case with expected vs actual output, or confirm it passes, plus the real complexity.

4. Official solution

Call oahelper_solution(ref, language) only when the user asks for the official solution. It may use the account's daily solution quota; the result line says how much is left. Compare it to the user's approach rather than just pasting it.

Append the outcome to library/_progress.md (date | company | ref | title | topic | result), where result is solved, solved-with-hints or not-solved.