Patch broken DarkNews ModelContainer

[austinschneider]

Stack: PR 9 of 14

This PR is part of a 14-PR stack decomposing dev/HNL_DIS into reviewable chunks.

• Base: feat/detector-sector-by-name
• Head: fix/darknews-modelcontainer

Merge order: feat/detector-sector-by-name should land before this PR.

Squashed contents

Squashed diff for paths:

• python/SIREN_DarkNews.py

Source commits on dev/HNL_DIS_clean that contributed:
4745193 (Austin Schneider): Patch broken DarkNews ModelContainer
521695f (Nicholas Kamp): fix issues with 2D tabular integral calculation, sampling errors with very short decay lengths, DarkNews errors, and kinematic erros in Dipole portal HNL DIS
ac0ce24 (Nicholas Kamp): changes to facilitate hadronic and W/Z HNL decays
f183191 (Nicholas Kamp): more complex logic for adding darknews interactions, fix some errors in SIREN_DarkNews

Notes

• This branch is the squashed cumulative diff of the listed source commits. Per-commit history is browsable on dev/HNL_DIS_clean.
• Large .fits data files have been removed from the branch and are packaged separately.

[Copilot]

Pull request overview

Note

Copilot was unable to run its full agentic suite in this review.

Patches DarkNews decay serialization/integration handling to fix broken ModelContainer behavior in the stacked HNL DIS refactor.

Changes:

• Adds decay_dict to stored decay state and serialization output.
• Updates pickle deserialization to load both decay_dict and decay_integrator.
• Adjusts decay-width computation and DarkNews sampling input shape.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

pickle.load(ifile) is now assumed to return exactly a 2-tuple (decay_dict, decay_integrator). If existing tables were written by older code that pickled (_, decay_integrator) (or only the integrator), this will raise a ValueError at runtime and break users with cached tables. Consider supporting both formats (e.g., branch on tuple length / type) and initialize self.decay_dict appropriately when loading legacy pickles.

Suggested change

	self.decay_dict, self.decay_integrator = pickle.load(ifile)
	decay_data = pickle.load(ifile)
	if isinstance(decay_data, tuple) and len(decay_data) == 2:
	self.decay_dict, self.decay_integrator = decay_data
	else:
	# Backward compatibility with legacy caches that stored
	# only the integrator object.
	self.decay_dict = None
	self.decay_integrator = decay_data

[Copilot]

Adding the new "decay_dict" field changes the serialized representation schema. If there is any corresponding deserialization path (e.g., from_representation / model container loader) that expects a fixed set of keys, it may break when reading older saved state (missing key) or when older code reads newer state (unexpected key). A tangible mitigation is to make (de)serialization tolerant: treat "decay_dict" as optional with a default, and/or include a simple schema/version tag in the representation.

Suggested change

"decay_dict":self.decay_dict,

[Copilot]

Commenting out the print leaves dead code in a runtime branch. Prefer either removing the line entirely, or switching to a proper logger call (at an appropriate level) so behavior can be controlled without editing code.

[Copilot]

np.expand_dims(PS, 0).T changes the shape from (1, N) to (N, 1) (and does nothing for 1D inputs beyond swapping axes convention). If get_decay_momenta_from_vegas_samples expects the original orientation, this will silently change semantics and can produce incorrect momenta. A more robust approach is to reshape explicitly to the exact expected shape (e.g., (1, -1) or (-1, 1)) and optionally validate the resulting dimensions before calling DarkNews.

[austinschneider]

Closing to reopen from the commit author's account so they can be added as a reviewer. Branch unchanged; will be re-linked from the new PR shortly.