aps-problem-solving-lifecycle

Use the 16-phase lifecycle model to map each user action from problem discovery through review, then apply the stuck-point taxonomy to classify failures as parsing, modeling, idea generation, proof, implementation, or verdict issues. Each stuck type routes to a different product intervention.

An Attempt is the solver's full problem-solving effort including reading, thinking, and local testing, while a Submission is only code sent to the judge. Tracking only submissions misses abandonment, hint usage, and learning state.

Yes, abandoned attempts are a high-value learning signal showing exactly which phase users give up in. The Skill recommends capturing abandonment with low-burden signals like one-tap self-reports rather than treating submission history as the whole journey.

Use the verdict-to-cause map: WA typically stems from misread statements, edge cases, overflow, or wrong algorithms, while TLE indicates complexity too high for the constraints, slow I/O, or infinite loops. Each verdict requires a different diagnostic routing.

Revealing tags too early harms pattern discovery, and full solutions eliminate germane cognitive load so users learn nothing. The Skill recommends graduated hints matched to the user's stuck type and logging assistance as a weaker learning signal.

It does not solve algorithm problems, write competitive-programming code, or debug submissions. It also avoids jumping to UI, database schema, or DDD artifacts unless explicitly requested, staying focused on domain framing, requirements, and scope.