AC19 standalone scientific materials

Python 3.9+ standard library:
python3 verify-observations.py --input .
python3 verify-inference.py --input .
python3 reproduce-orbits.py --out new-orbit
python3 analyze-frozen-posterior.py --directory .
python3 reproduce-timing.py --input . --out new-posterior --iterations 16000 --ensembles 4
python3 run-frozen-inference.py --input . --out new-all

Use spaces between option names and numbers, for example --iterations 16000 --ensembles 4.
The full inference has four independent 30-walker ensembles per model, 16000 iterations, initial third discarded, retained every 20 iterations. Seeds 19019+101*ensembleIndex. The six experiments keep actual 25 measurements, wider prior, and synthetic analytic injection separate. Saved CSV.gz chains retain the correlated draws; effective ensemble means are a different count.

Core inputs: original ajac8553t1_mrt.txt; fifty-one measured rows in kepler289-observed.csv. Only the same fifteen middle and ten outer Kepler times enter each actual likelihood. No forecast time or forecast error enters it. Source global row 430 has no observed 2021 WIRC value.

Nine physical parameters: middle/outer mass in Earth units, real/imaginary Zfree, central BJD epochs 10/5, periods in days, common white jitter in days. Source time scale is unspecified beyond BJD-2454000, and the fixed adopted stellar mass is 1.08 solar. Full prior bounds and sampler metadata are in each timing-results.json.

Optional independent numerical reference: Python 3.12 + REBOUND 5.2.2.
python3 -m pip install rebound==5.2.2
python3 verify-ias15.py --input reference-input.json --out new-reference.json

Optional harmonic check: install NumPy 2.3.5, then
python3 validate-harmonics.py --input default-model-check.json --out new-harmonic-check.json

Optional figures: Python 3.10+ with matplotlib 3.10.7 and NumPy 2.3.5.
python3 -m pip install matplotlib==3.10.7 numpy==2.3.5
python3 plot-timing.py --directory . --out new-figures

Model hierarchy: synthetic full Newtonian orbit -> geometric event -> fitted linear clock -> timing residual. Independent IAS15 validation tests numerical event accuracy; the real-data reduced harmonic likelihood has substantial remaining discrepancy. The analytic injection checks inference closure only, not interval coverage or independent harmonic physics. No complete published photodynamic fit is reproduced.

Outputs across Python/platform math libraries need not be byte-identical MCMC chains. Input/source/download bytes are hash-checked. Compare derived numerical/statistical quantities with their numerical and Monte Carlo precision, and inspect all diagnostic flags. A diagnostic pass is not scientific certification.