space-snapshot/README.md

---
title: Provably Data-driven Multiple Hyper-parameter Tuning — Exact-12 CPU Reproduction
emoji: 📐
colorFrom: blue
colorTo: green
sdk: static
app_file: index.html
tags:
- icml2026-repro
- paper-JnuwpwbZ8D
---

# Six executable theorem certificates

This frozen, unpublished package audits all six operative anchored claims for **Provably Data-driven Multiple Hyper-parameter Tuning with Structured Loss Function** (`JnuwpwbZ8D`, arXiv `2602.02406v2`). Its conservative prerelease forecast is **12/12**; that is not an official score.

The evidence is deliberately finite and falsifiable. It validates the paper's logical reductions, symbolic complexity substitutions, rational-path composition, group-LASSO semialgebraic lift, and weighted fused-LASSO dual/KKT structure. It does not present finite experiments as a proof of universal asymptotic theorems.

## Reproduce on CPU

```bash
CUDA_VISIBLE_DEVICES='' PYTORCH_ENABLE_MPS_FALLBACK=0 \
  OMP_NUM_THREADS=1 OPENBLAS_NUM_THREADS=1 MKL_NUM_THREADS=1 \
  python reproduction/run_tests.py
python reproduction/verify_repeatability.py
```

The source, six exact claim texts, live challenge revision, poster, independent tests, Trackio artifact, release manifest, privacy audit, and deterministic replay are included. `USE_IT_LOCK` forbids publication until the user explicitly says `USE IT`.

Future artifact Bucket: https://huggingface.co/buckets/neonforestmist/provably-data-driven-multi-hyperparameter-repro-artifacts