mg7 1 -> N processes: decaying mass without parameter_dict, decay-aware observables, no histograms in MadSpin, 1->3 CI test

[oliviermattelaer]

Problem

With madspace built, MadSpin (spinmode none) generates its decays through output mg7, and test_hepmc_decay / test_lhe_none_decay failed (no *_decayed.lhe.gz). Without madspace they take the madevent fallback, which is why nobody noticed. Two bugs, both in the default [histograms] of a 1 -> N process (81e9a13):

1. RunCardMG7._decaying_mass read model.get('parameter_dict'), and only a ModelReader has that property. A LoopModel raised PhysicsObjectError, which the except didn't catch, so output crashed. A plain Model raised KeyError and silently gave mass 0.
2. madspace Observable assumes two beams (ignore_incoming drops the first 2 momenta, and the sqrt_s kernel is p[0]+p[1]). For a decay, the first decay product dropped out of every selection. z > mu+ mu- asked for lepton_2 and the launcher died with "absolute value of order index must be smaller or equal to number of selected PIDs". In t > b e+ ve, every b histogram sat at 0. This happens with plain output mg7 of any decay, not only under MadSpin.

Fix

• _parameter_value: if the model has a parameter_dict, use it. Otherwise evaluate from the default external parameter values in a scratch namespace (internal masses such as mdl_MW in sm-full included). The model's own parameters are not touched.
• build_observable in the mg7 launcher: with one incoming particle, the selection runs over all momenta (ignore_incoming=False) and the incoming pid is set to 0, which belongs to no group. Used for cuts, integration histograms and event histograms.
• No default sqrt_s histogram for a decay: madspace's kernel adds two beams, and for a decay the value would be a constant anyway.
• MadSpin's mg7 decay runs now drop [histograms] altogether (like set histograms OFF). Nobody looks at the plots of a decay pool, and each one costs an observable evaluation per event.

Tests

• New test_decay_1to3_mg7 (added to acceptancetest_mg7_bundle_1): runs t > b e+ ve through mg7 and pins the width to madevent (0.1636 ± 0.0002 GeV). It also checks the default histograms: no sqrt_s, every momentum histogram holds the full width in range, and the e+ ve pair mass peaks in the bin containing MW. On origin/main it fails: bottom-pt has everything in its first bin and sqrt_s is all overflow.
• New unit test test_decaying_mass_without_parameter_dict (plain Model and LoopModel built from sm-full, where MW is internal). test_decay_process_uses_the_decaying_mass is updated for the dropped sqrt_s.
• Locally, with madspace built:
  • test_hepmc_decay, test_lhe_none_decay, test_decay_1to3_mg7, test_w_production_with_ms_decay_mg7, test_madspin_mixed_flavor_decay_log_summary_mg7: Ran 5, OK. On origin/main the first two fail with the PhysicsObjectError above.
  • All test_banner.py + test_mg7_launch.py classes: Ran 157, OK.
• Physics check on MadSpin's z > mu+ mu- decay histograms, before histograms were switched off there: the pair mass falls in the [90, 92] GeV bin, and the muon pT edge is at m_Z/2.

🤖 Generated with Claude Code