Module analysis 

Derived series and diagnostics over a SolveReport.

Mirrors the Python analysis helpers in studio/mcp/pounce_studio_mcp/ reports.py so the desktop / VS Code shells and the MCP server can agree on the same notion of “stall window”, “restoration window”, and “common failure modes”. Heuristics are tunable via the parameters on each function; the defaults are the ones the Python diagnose tool ships with.

Structs

AugmentedIterate

Augmented IterRecord returned by get_iterate: the raw row plus derived log10 values handy for tooltip / LLM rendering.

CompareRow

One row in a side-by-side comparison.

ConvergenceTrace

Per-iteration trajectory in column-oriented form. More compact than a Vec<IterRecord> when serialised, since the column names are emitted once.

Finding

One finding from diagnose.

RestorationWindow

Contiguous runs of iters tagged 'r' in the alpha-primal char column — one entry per restoration entry → exit cycle.

Stall

One stalled-progress window: consecutive iterations whose log10-residual moved by less than the configured threshold.

Summary

Compact view-model derived from a SolveReport. Suitable as the “headline summary” that an LLM or dashboard reads first.

Enums

Severity

Functions

compare_runs

convergence_trace

diagnose

Run common Ipopt-failure heuristics and return all findings.

find_stalls

Default stall detection: 5+ consecutive iters with <0.3 orders of magnitude movement in either inf_pr or inf_du.

find_stalls_with

get_iterate

restoration_windows

summarize