Add real-input regression validator for subchunk max-azimuth rev9/rev10

[nicklasorte]

Motivation

• Provide a trustworthy regression harness that runs subchunk_agg_check_maxazi_rev9 and subchunk_agg_check_maxazi_rev10 on the exact same production inputs to detect behavioral drift.
• Validate structural invariants and numeric agreement while measuring runtime to catch performance regressions.
• Fail closed on any meaningful mismatch so CI or a developer run cannot silently accept regressions.

Description

• Added validate_subchunk_agg_check_maxazi_rev9_rev10_real.m which exposes the exact production signature and accepts the real inputs used by the existing codebase.
• The validator calls both versions with identical arguments (via local function handles), captures outputs, and performs structural checks: size match, column-vector shape, NaN pattern match, Inf pattern match, and class compatibility.
• It performs numeric comparisons including diff_vals, abs_diff, rel_diff, max_abs_diff, max_rel_diff, mean_abs_diff, an exact_match via isequaln, worst-index reporting, and a finite-element mismatch count versus tight tolerances tol_abs = 1e-10 and tol_rel = 1e-10.
• It measures runtimes using timeit, computes speedup = runtime_rev9 / runtime_rev10, prints a concise --- VALIDATION SUMMARY ---, returns a results struct with clear fields, and raises error('Validation failed: rev10 does not match rev9 within tolerance.') when checks fail.

Testing

• Verified the presence and signature of the existing implementation by inspecting subchunk_agg_check_maxazi_rev9.m, which succeeded.
• Created and committed validate_subchunk_agg_check_maxazi_rev9_rev10_real.m to the repository, and confirmed the new file content with nl/sed.
• Attempted to execute an Octave runtime check but the environment lacked Octave/MATLAB (octave: command not found), so runtime timeit execution was not performed here.
• To fully exercise the validator run it in a MATLAB or Octave environment with the real production inputs; the validator will print the summary and return the results struct or raise an error on failure.

---

Codex Task

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 72492f3afc

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Guard missing rev10 implementation before executing validator

The validator calls rev10_fun() unconditionally, but this repository does not contain a subchunk_agg_check_maxazi_rev10 definition (repo-wide search only finds this reference), so execution fails immediately with an undefined-function error. That prevents the new regression harness from running at all in the current codebase and blocks the intended rev9/rev10 comparison workflow.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Evaluate numeric pass/fail with per-element tolerance logic

The numeric_match check uses global maxima with ||, which can incorrectly fail even when every finite element satisfies the intended (abs<=tol_abs) OR (rel<=tol_rel) rule. For example, one element can pass only relative tolerance while another passes only absolute tolerance; finite_mismatch_count is zero, but both global maxima exceed their thresholds so this expression marks the run as a failure. This can produce false regression failures and make CI results unreliable.

Useful? React with 👍 / 👎.