When a parameter sweep cannot identify its own model

The analysis pipeline records non-identifiability instead of filling missing parameters by interpolation or by sharing a fit across thicknesses. That negative result constrains what later calibration work is allowed to claim.

Run from the repository root:

E:\SOFTWARE\annaconda\annaconda_evn\python.exe analysis\run_all.py

The pipeline is read-only with respect to source data. It writes only below analysis/outputs/. Configuration, mappings, units, thresholds, bootstrap count, and the stiffness window are in config.yaml. Data inventory is limited to tracked files plus non-ignored untracked files, so local solver runs excluded by .gitignore cannot silently change the formal inventory. run_all.py writes output_manifest.csv last; every listed size and SHA-256 must verify after a successful run.

Important design limitation: only the 1.25 mm eyelid has a full multi-IOP curve. Other thicknesses have independent 0 and 20 mmHg endpoints. The code therefore records their rational-model parameters as non-identifiable; it does not fill them using shared parameters or interpolation.