# One fixed M67 patch, two local native model sets

This standalone Python3.12 workflow actually fetches/parses original inputs,
selects Gaia sources, compares native same-band curves, builds the two local
twelve-age browser inputs, and checks a discrete licensed-node population.
It does not rebuild the protected Lab Engine, run stellar evolution, fit an age
posterior, recover a complete IMF or rerun a spectroscopic abundance analysis.

Use the versions in requirements.txt in an environment you manage. The scripts
never install tools. Run from this directory, with fresh output directories:

    python verify_package.py
    python fetch_inputs.py --out inputs
    python prepare_modelsets.py --inputs inputs --out local-modelsets
    python reproduce.py --inputs inputs --out results
    python validate.py --inputs inputs --results results

Then choose parsec-native12.json or mist-native12.json from local-modelsets in
the Lab's local file input. The site receives no upload and initiates no remote
model job. The importer pins source schema, all twelve actual age nodes,
composition, EDR3 passbands, full original numeric-table hashes, native mass
order and phase refusal. It does not interpolate age.

The default public model is different: the CC BY4.0 ISOA slice has one3.95Gyr
age and native GaiaDR2 magnitudes. Its98 selected H-burning nodes are sampled
as a discrete distribution with IMF integrals over midpoint mass cells. We do
not use the failed continuous public photometry interpolation. Source metadata,
attribution, license link and selection changes accompany the included JSON.

What the commands calculate:
- fetch_inputs uses exact upstream recipes. A generated CMD/MIST URL is
  temporary; the fixed POST request is the reproducible entry. MIST uses a
  verified central directory and exact206photometry-member ranges, with CRC
  and decompressed length checks. Ignored ranges stop; no whole-ZIP retry.
- prepare_modelsets parses complete original tables and derives the98 public
  nodes anew from the actual author CSV. Import JSON stays local.
- reproduce computes both actual astrometric selections and source flags,
  all12PARSECage/threeAv-node descriptive comparisons at three distance
  moduli, MIST/PARSEC physical discrepancies, original projection checks,
  the same-input independent audit, public node weights/photometry, and the
  fixed paper Table2 aggregates. Individual Gaia/model points and plots
  are created only under the reader's results directory.
- validate independently rereads the original CSV to recover selected IDs,
  and checks component identities, total actual birth mass, same-band light,
  vector distances and positive normalized native-node weights.

Expected failures remain failures:
1. public continuous ISOA interpolation exceeds its predeclared0.002mag gate;
2. withheld native age interpolation fails and continuous age is refused;
3. the original seed20261104 projection sample exceeds its original99.9%
   DKW band, although independent same-input geometry agrees.
A zero command exit means the workflow preserved and checked these outcomes,
not that all scientific hypotheses passed. A boundary age/distance minimum
is not a posterior or an age hard bound.

All original inputs and individual-source derivatives stay on your computer.
Gaia's dedicated data page specifiesCC BY-NC3.0IGO; no Gaia rows or real-source
scatter is bundled. PARSEC/MIST table redistribution is not assumed from
public access or software licensing. The package contains code, recipes,
hashes and the expresslyCC BY4.0 default slice, not those tables or surrogates.
Check source terms before sharing your local generated files.

For a fully offline cached-body replay, supply each original input by name:

    python fetch_inputs.py --out inputs --cache /path/to/original-cache

Repeat --cache for multiple directories. This validates raw direct-source
hashes and full canonical model numerics before computation; it does not
claim a fresh HTTP acquisition. Future source/service changes fail visibly
and require an explicit reviewed input revision. The original response-body
hash and canonical float64 numerical-table hash are recorded separately.