You are analyzing a failed benchmark submission for a scientific computing task.

Given the scoring breakdown, agent-generated code, and execution log,
identify the root cause(s) of failure.

Classify the error using this taxonomy:
  algorithm_error:        wrong scientific understanding
    wrong_method           used an inappropriate method entirely
    wrong_formula          formula / equation error
    wrong_discretization   wrong numerical scheme
    missing_step           critical algorithmic step absent

  implementation_bug:     code is scientifically correct but has bugs
    off_by_one             boundary / index error
    shape_mismatch         array shape incompatibility
    type_error             data type mismatch
    import_error           missing dependency
    runtime_crash          unhandled exception

  misunderstanding:       prompt was misread
    wrong_output_format    output file format / shape wrong
    wrong_input_parsing    input data parsed incorrectly
    incomplete_reading     prompt not fully read

  quality_issue:          correct method, poor execution
    numerical_instability  scheme unstable under given parameters
    convergence_failure    iterative solver did not converge
    insufficient_precision tolerance too loose

  resource_issue:
    timeout                exceeded time limit
    memory_overflow        OOM
    gpu_required           code assumes GPU not available

Output a structured JSON diagnosis with:
  error_category, error_subcategory, root_cause, evidence, fix_suggestions, confidence
