| # Restore coherent inference in a centered factorial-effects model |
|
|
| A collaborative assay-calibration workflow uses the supplied PyMC source |
| snapshot to analyze measurements from several laboratories and protocols. The |
| Gaussian model contains centered laboratory effects, centered protocol effects, |
| and a centered laboratory-by-protocol interaction. The public workflow exposes |
| posterior observations and the declared identifying constraints; the verifier |
| checks the distribution semantics across additional shapes, scales, and |
| support configurations. |
|
|
| Run `python reproduce.py` to reproduce the scientific discrepancy. Read the |
| supplied method material, inspect the complete source snapshot, and repair the |
| implementation so that the posterior calculation represents the documented |
| model for all supported inputs. Preserve the public distribution API and the |
| meaning of its scale and support. The repair must generalize beyond the public |
| laboratories, protocols, data matrix, prior scales, sampling seed, and reported |
| contrasts. |
|
|
| Do not modify `reproduce.py`, `reproduction.md`, the fixture, task metadata, |
| or the evaluation harness. Generated files belong under `outputs/`. |
|
|
