Synthetic draws · target-rule checks

upg-strings

ML-guided ranking for rare Calabi-Yau targets in the string landscape

Rank synthetic Hodge-number draws against dataset target rules, with reproducible seeds and shareable candidate dossiers — not a claim of experimental physics verification.

Verified means the geometry passes that dataset’s target rule on synthetic labels (for Kreuzer–Skarke: |χ| < 100). It is not lab confirmation of a vacuum.

Durable product: Hall of Fame + shareable dossiers, plus Lookup for known Hodge triples.

Textbook example: the quintic mirror (h¹¹=1, h²¹=101, χ=−200) — kreuzer-skarke-66d611d18a9d.

Explore first

Primary entry points are curated candidates and lookup — not a single sync search button.

Hall of Fame

Persistent board of geometries that passed dataset target rules, each with a shareable dossier page.

Lookup

Identify a Hodge triple and open its dossier without waiting on a live ranking job.

Quintic candidate

Open the classic mirror-quintic-style example dossier (synthetic / literature Hodge pair).

Optional: run a live ranking

Secondary demo: rank ~5,000 synthetic draws. Metrics are synthetic retrieval vs random baseline — target-defining features are held out of the model. Runtime is typically a few seconds.

For Scientists

Use upg-strings as a reproducible filter before deeper analysis — with clear limits on what “verified” means.

Outputs

results_topk.csv

Top-ranked candidates with scores and target-rule status

metrics.json

Synthetic retrieval metrics, baseline, and honesty / leakage notes

repro.md

Complete reproducibility report with environment details

Compatible exports: CYTools, cymetric, Sage, Mathematica, plus JSON/CSV.

CYTools cymetric Sage Mathematica JSON CSV

Metrics (honest framing)

Precision@k

Fraction of top-k that pass the synthetic target rule — vs random baseline in the same draw

Recall@k

Fraction of all rule-positive rows found in top-k

Hold-out

Target-defining columns (e.g. |χ|) are withheld from the RandomForest

Links

Disclaimer: Demo corpora are synthetic Hodge draws. This tool ranks and packages candidate dossiers for exploration — Verified means a dataset target-rule match on those labels.