Repair an incomplete stiff-kinetics campaign
A kinetics group is replaying a supplied multi-condition Robertson reaction campaign with a historical SciPy source snapshot. The reaction systems, initial compositions, rate constants, observation levels, time horizons, and tolerances are valid. The campaign should produce one finite, mass-conserving observation for every configured protocol and a stable aggregate summary. The supplied implementation completes only part of the campaign, so the aggregate result cannot be released.
Run the offline reproduction, inspect the scientific workflow and source, and repair the implementation so that the complete campaign is scientifically consistent. The repair must generalize beyond the public protocols to other supported differential-equation systems, integration directions, sampling choices, tolerances, and solver configurations. Do not hard-code protocol identifiers, public rates, monitored levels, or report fields, and do not modify the supplied scientific inputs.