Check the program,
then read the evidence.
By Alvin
python3 check_programs.py my-notebook-fixed/notebook.py my-publication/publication.py
After inspecting your code, use this checker with invented inputs. It prints each result and a total. It checks exact quotes, sorted IDs, preserved input files, help, empty lists, invalid files and records, duplicate IDs, the two queues and ready-only export. A pass covers these examples.
| NOVEMBER 17 / M1 MAX | RECORDED RESULT |
|---|---|
| Quote Desk / repaired | 7/7 behavior checks; repair request 58.43s |
| Publication Desk | 9/9 behavior checks; request 39.24s |
| Engine preparation | Existing source and weights reused; build replay and two numerical checks completed. |
We ran one request per build brief and one focused notebook repair, then executed the saved programs on the M1. Both also passed the same checks on the primary Mac. This is a small real-program demonstration, not a general model reliability score or a controlled speed comparison.
The kit keeps the two exact prompts, sample data, generated source, raw replies and check results. Run python3 check_programs.py with no filenames to check the supplied tested builds. No model download is needed to run those saved Python programs.
The November 14 notebook/triage function experiments on the following pages remain separately dated. The first-time operating-system installers were not rerun; the M1 already had its development tools and model files.