analysis

Physics analysis catalogue

1. For table-based analyses (fit_physics_spectrum_peaks, analyze_physics_signal, fit_physics_york_line, check_physics_lab_experiment) first materialize the measurement table with $physics-dataset; inside Agentlas Science pass its dataset_run_id, standalone pass the normalized table. Column names refer to that table and units come from its column declarations.
2. Parameter analyses (compute_physics_significance_limits, propagate_physics_uncertainty, analyze_physics_units, simulate_physics_ode) take explicit numbers; state every input the result depends on (background uncertainty, seed, tolerances, unit strings).
3. Starting values for peak fits and York fits are the caller's responsibility: read them off the data first, then check summary.converged, stopReason, at-bound parameters, and χ²/ndf before quoting any parameter. For a spectrum fit with a mapped sigma_column whose cells are all empty, omit uncertainty_model so the recorded result uses Poisson weighting with a warning. A partly empty measured uncertainty column is not silently filled; explicit column weighting requires complete values.
4. Every result carries boundaries (method limits) and warnings (data-specific cautions). Quote them with the numbers; never promote an asymptotic or first-order result beyond what the boundaries allow.
5. Preserve analysisSha256 and figure.figureSha256 with the publication table and the Vega figure; a replay with the same inputs reproduces both hashes.
6. Cross-check expectations: each analysis has a numpy/scipy oracle under tests/<name>-crosscheck.py; tolerances are documented in docs/science/physics-tools.md.