### 1. generar los datos con la receta
Project:  CelsiusToKelvin
Seed:     20260825  mode: coherent  rows: 300
Train:    datos/celsiustokelvin-synthetic-train.csv  (240 rows)
Eval:     datos/celsiustokelvin-synthetic-eval.csv  (60 rows)
Manifest: datos/celsiustokelvin-synthetic-manifest.json

### 2. entrenar
Training OK: v1
Best epoch: 25
Best validation loss: 0.000000
R2: 1.000000
MAE: 6.50521e-17
Artifacts: runs/v1

### 3. exportar el paquete
Bundle OK: paquete
Files: README.md, data_recipe.txt, example_input.json, expected_output.json, export_manifest.json, inference_spec.json, model.mxai, model.mxtrain, model.onnx, model_manifest.json, params.best.json, predict.py, reproduce.json, requirements.txt, space/README.md, space/app.py, space/requirements.txt
Equivalence PASS: max_abs_diff=3.57e-08
Self-usable: yes (predict.py + inference_spec.json included)
Reproducible: yes (reproduce.json)

### 4. predecir SIN MatrixAI, con valores crudos
  0 °C -> 273.15000146627426
  100 °C -> 373.1500029563904
  -40 °C -> 233.15000236034393

### 5. verificar el paquete descargado (rehace el dataset y reentrena)
manifest  PASS
R1        PASS
training  PASS
R3        PASS