name	cumcm-problem-reading
description	Parse and decompose 2026 CUMCM problem statements into subquestions, data requirements, deliverables, hidden constraints, scoring signals, assumptions to verify, and first-day action plans. Use when reading a new CUMCM problem, extracting tasks, clarifying what must be modeled, or building the initial problem brief before choosing methods.

CUMCM Problem Reading

Use this skill at the start of a contest problem or when a team is confused about what the problem is actually asking.

Split the statement into subquestions, required outputs, given data, constraints, and evaluation hints.
Mark ambiguous wording and assumptions that must be decided by the team.
Build a task matrix: subproblem -> model need -> data need -> result artifact -> validation need.
Identify which parts can start immediately and which depend on data cleaning or external definitions.
Produce a first-day plan with baseline model, data audit, and paper skeleton actions.
Read references/checklist.md for the full reading pass.
Read references/task-matrix.md when converting a problem statement into work items.
Read references/examples.md when deciding how much detail to include.

Do not solve from a vague reading; list missing definitions first.
Do not invent hidden data or performance numbers.
During the active contest window, do not browse public discussion of current problems.

name	cumcm-problem-reading
description	Parse and decompose 2026 CUMCM problem statements into subquestions, data requirements, deliverables, hidden constraints, scoring signals, assumptions to verify, and first-day action plans. Use when reading a new CUMCM problem, extracting tasks, clarifying what must be modeled, or building the initial problem brief before choosing methods.

Use this skill at the start of a contest problem or when a team is confused about what the problem is actually asking.

Split the statement into subquestions, required outputs, given data, constraints, and evaluation hints.
Mark ambiguous wording and assumptions that must be decided by the team.
Build a task matrix: subproblem -> model need -> data need -> result artifact -> validation need.
Identify which parts can start immediately and which depend on data cleaning or external definitions.
Produce a first-day plan with baseline model, data audit, and paper skeleton actions.
Read references/checklist.md for the full reading pass.
Read references/task-matrix.md when converting a problem statement into work items.
Read references/examples.md when deciding how much detail to include.

Do not solve from a vague reading; list missing definitions first.
Do not invent hidden data or performance numbers.
During the active contest window, do not browse public discussion of current problems.

cumcm-problem-reading