audit_results

Reads the result files and names anything missing, inconsistent or suspicious.

Group: Check the answer.

Parameters

Parameter	Type	Required	Default
work_dir	string	yes
claimed_order	number	no	0.0

What the model reads

The text below is the tool's own description, exactly as the AI model receives it.

Show the full description

Check your OWN result files for the failures that most often sink a
result set — BEFORE you hand it in. Uses only files you produced; no
reference solution is involved, so a clean audit means self-consistent,
not correct.

What it catches, measured on 94 independently-checked correct
result sets (no false alarm on any) and 102 complete-but-wrong ones
(39 caught, about four in ten — more when the result set states its
claimed convergence order, which the order check needs):

  * NEAR-ZERO FIELD - your finest solution peaks below 1e-8. On a
    driven problem that almost always means the source/load was never
    wired in (a defined function no condition references, a load curve
    never activated), not that the answer is small.
  * FLOOR - successive refinement levels within 5% of each other:
    whatever limits your number, it is not the mesh. Usual cause is a
    solver tolerance (nonlinear/iterative defaults stop near 1e-6).
  * ORDER MISMATCH - your levels improve at a measurably lower rate
    than the order you are about to claim. Usual causes: element
    degree below what the task states, volumetric locking, a
    first-order integrator behind a spatial study.
  * NON-MONOTONE - a refinement made the answer worse.

Call it on the directory holding your per-level outputs (it reads
your summary file and your per-level field files), pass the
convergence order you intend to claim, and treat any finding as a
reason to look BEFORE handing in - each one names where to look.

Args:
    work_dir: directory containing your results (searched recursively)
    claimed_order: the convergence order your result set will claim
        (0 = no order claim, order checks are skipped)