Motivation

• Speed up the hot aggregation path while keeping strict per-iteration RNG identity unchanged so rev9 vs rev10 comparisons remain deterministic and auditable.
• Remove inefficient azimuth chunking and repeated partial max scans while preserving function contract, units, and numerical behavior.
• Provide a strong, reproducible validation harness to prove rev10 matches rev9 within a tight tolerance before any RNG/vectorization changes in a later revision.

Description

• Added subchunk_agg_check_maxazi_rev10.m, a refactor of rev9 that preserves the MC seeding pattern (rng(rand_seed1+mc_iter), +1, +2) and keeps the MC loop to maintain RNG stream semantics while removing the azimuth chunk loop and repeated partial max scans; aggregation now computes dBm -> watts -> sum -> dBm across the full azimuth set and takes a single max per MC iteration.
• Precomputes the off-axis gain matrix once (same nearest-neighbor behavior as rev9) and gates optional expensive NaN checks behind DEBUG_CHECKS=false to avoid hot-path cost.
• Reduces temporary allocation overhead in the hot path by avoiding chunk slicing and redundant max operations while keeping numeric operations and shapes identical to rev9.
• Added validate_subchunk_agg_check_maxazi_rev10.m, a deterministic regression harness that builds fixed synthetic cases (small, medium, large), runs both subchunk_agg_check_maxazi_rev9 and subchunk_agg_check_maxazi_rev10 on identical inputs, checks shape/NaN/Inf patterns, reproducibility, computes max_abs_diff and max_rel_diff, enforces a tight tolerance (tol = 1e-10), measures runtime (using timeit where available or tic/toc), and fails fast on any mismatch.

Testing

• Created and inspected both new files (subchunk_agg_check_maxazi_rev10.m and validate_subchunk_agg_check_maxazi_rev10.m) using file-printing/smoke commands (nl -ba), which succeeded in this environment.
• The validation harness is fully automated: it runs both rev9 and rev10 on three deterministic cases, checks size, column-shape [num_mc_idx x 1], NaN/Inf pattern equality, reproducibility, computes max_abs_diff and max_rel_diff (denominator max(abs(out_rev9),1e-12)), uses tol = 1e-10, measures runtime with timeit fallback to tic/toc, prints a structured summary, and raises an error on failure.
• Note: the harness was not executed to completion inside this CI environment because no MATLAB/Octave interpreter was available here; the harness is ready to run in a MATLAB/Octave environment and will produce PASS/FAIL and runtime speedup numbers when executed there.

Codex Task