Reproducing the sensitivity sweep Download model_sweep.py and sweep.csv into a working folder. The script needs Python 3, NumPy and Numba. It was checked with Python 3.12.3, NumPy 2.4.2 and Numba 0.65.1. For example, in an isolated Python environment: python3 -m pip install numpy==2.4.2 numba==0.65.1 python3 model_sweep.py validate python3 model_sweep.py sweep --output results/sweep.csv The output defaults to sweep.csv in the current folder. Choose a new filename if that file already exists. The script creates missing parent directories and refuses to overwrite an existing result. The sweep's 133 rows use the existing fixed-step implementation, a time step of 0.01 hours, and its original calibration precision. Its numerical equations, parameters and CSV formatting haven't changed. The expected CSV SHA-256 is 9a083b4a2eb737245c26f5c4f3c0678478d76fa15c950ef3ee4e61b9ffb23d69. Validation compares the script's results with the archived reference values. Its absolute tolerances are 0.20 mg CN/kg for chamber results, 1.0 mg CN/kg for R12 and R13, and 0.00005 per hour for the capture coefficient where a reference coefficient is supplied. It checks finite values and returns a nonzero exit status if a comparison fails. These are numerical comparison tolerances, not measurement errors or confidence limits for historical walls. The main fixed-step result rounds to 8.18 / 2,884 / 3,053 mg/kg. Some auxiliary results differ from the archived calculations in their last digits, including 520.33 versus 520.46 and 551.76 versus 551.93 mg/kg for chamber results. Exact last-digit agreement isn't claimed. The reason for every difference hasn't been isolated. variant_results.json preserves the archived adaptive calculations, including their schedules and assumptions. Its values aren't replacements for sweep.csv. variant_definitions.json gives the seven capture exponents and their meaning. concentration_summary.json gives mobile inventories, assumed water volumes, time points and resulting concentrations. Those concentrations aren't measured historical moisture or wall concentrations. Each record retains its original calculation values; local paths and execution-host metadata are omitted. Numerical reproducibility doesn't validate transfer of laboratory inputs to the historical wall materials. The page states the remaining limits.