Primary-photon fit form quality across 13 materials

Sibling of the neutron and secondary-photon harnesses. Validates the shipped 1D primary-photon fit (Shin-Ishii double exponential) against photon-source MC anchors.

Setup

Same cache file as the other two harnesses, but loads only the photon-source materials (*_photon keys: polyethylene_photon, concrete_photon, ..., tungsten_photon). These come from compute_buildup runs with a 1 MeV photon source — dose-AP is then the primary-photon dose (no secondary radiation, no coupled-photon component to worry about).

The fit form is Shin-Ishii:

\[ B(\tau) = A \cdot e^{-\alpha_1 \tau} + (1 - A) \cdot e^{-\alpha_2 \tau} \]

with \(B(0) = 1\) by construction.

Result (13 photon-source materials)

Metric	Value
Worst holdout \(z\)	5.50 (polyethylene)
Mean holdout \(z\)	2.08
Median holdout \(z\)	1.07
Worst training \(z\)	1.69

Cleaner than the neutron case because primary-photon physics has no late-stage thermalisation tail; Shin-Ishii's two decaying exponentials match the underlying physics well across all materials. Mean holdout \(z = 2.1\) vs \(4.9\) for neutron and \(6.0\) for secondary photon.

Running

python verification_and_validation/primary_photon_fit_test_suite.py

Same machinery for flux

Photon flux uses the same Shin-Ishii form. To validate it:

Regenerate the photon-source materials in the cache with compute_buildup(quantities=["flux-photon"], source=photon_src, ...).
Swap the field read in _load() from r["mc"]["dose-AP"] to r["mc"]["flux-photon"] (and the matching pk, mc_std_dev).

The harness code is identical otherwise.