Complete optimizer matrix qualifies
All 48 generated computational cases, all 24 paired-start comparisons and all four original-input replays pass the combined audit. It checks the 44 completed cases from the interrupted matrix and four separately completed controls. All 241 shared execution sources and the numerical environment match.
The original interrupted study stays unqualified, with its four interrupted starts and their charges preserved. The combined result uses each computational case exactly once; it does not relabel interrupted work or select among competing fitted answers. All three generating candidates, both input formats, both processing settings and the same original starts are covered.
The matrix took 51.0 minutes; the separate four-case completion took 4.6 minutes. Their pre-execution software qualifications passed 352 and 364 checks. The saved-only assembler passes 16 additional checks, including deliberately missing or duplicate cases, failed fits, bad pairings, changed sources and environments. It independently re-audits both source publications without running a fit.
This completes the declared generated-data optimizer coverage. A fresh calibration/validation design still needs disjoint cohorts and a separate finite review. No uncertainty threshold, aircraft confidence level or new Earth-model finding follows. The original 53 aircraft comparisons remain unchanged.
Evidence and reproduction
All 48 completed cases · 24 paired-start comparisons · Preserved interruption records · Audit plan.
The interrupted matrix and completion controls retain their own frozen sources, inputs, outputs, journals and verified publication receipts. The combined receipt binds the auditor, tests, qualification evidence, exact constituent completion hashes and combined report. Shared source snapshots remain in the constituent publications; they are not copied into another worker tree.
With the documented frozen environment and source restoration, re-run the saved-only check from the repository root:
env PYTHONPATH=analysis ~/venv/bin/python analysis/tests/run_numerical.py --workers 1 --threads 1 -- ~/venv/bin/python -c 'from pathlib import Path; from lll.timing_matrix_assembly import inspect; r=inspect(Path.cwd()); print(r["qualified"], r["cases"], r["start_pairs"], r["original_input_replays"])'
Expected output is True 48 24 4. The auditor rejects changed source files or
environments; restore its recorded versions rather than substituting newer code.