Post hoc channel-availability extension dated 4 October 2026
The initial paper results were already known before this extension was specified.
The numerical extension was executed using verified local replay cohorts, original fold-membership outputs, and the explicitly authorized vitalPeriodic table.
The stable primary clinical family and stable common-risk APACHE family each completed 100 full-refit draws. The stable common-risk clinical family has conditional intervals only.
The portable entry was prepared and compiled separately; numerical execution used the frozen local script. No claim is made that every original replay stage and this extension were run as one command in a fresh environment.

After producing the original corrected replay outputs, use your own authorized local paths:
python run_stability_extension_local.py --verified-replay-dir YOUR_REPLAY_DIRECTORY --raw-vital-file YOUR_AUTHORIZED_VITALPERIODIC_CSV_GZ --output-dir NEW_EMPTY_OUTPUT_DIRECTORY
Add --prepare-only to compile/configure without patient data reads or statistical fits.
The frozen extension fingerprints the same database-v2 input version documented in the study; it is not a generic tool for unverified data versions.
All row-level outputs and raw execution logs must remain local. Only reviewed nonidentifying aggregates may be distributed.
The graph uses aggregate outputs from the original and stability analyses; its plotting dependencies include matplotlib.
