Durable experiment record · Public core code · curated results

Relational manifold learning with Geometric Transformers

Triangle-aware relational recovery, dimension diagnostics, Hodge controls, and the registered MovieLens negative result.

12 code surfaces0 frozen runs1 code packagescatagi-gt-manifold

Experimental contract

What this family tests

controlled synthetic recovery and registered real-data calibration

  • Controlled relational recovery
  • Post-confirmatory dimension diagnostic
  • Pooled Hodge baseline
  • MovieLens registered negative calibration

Book-to-code crosswalk

The highlighted Experiment cards covered here

Implementation map

From entry point to source surface

Start here

These concrete files are selected from the complete registered source surface to orient the reader.

Author-declared entry points

  • relational_descent.py
  • run_relational_dimension_diagnostic.py
  • run_hodge_baselines.py
  • run_movielens_experiment.py

Resolved code

implementation · 3
result-record · 4
runner · 5

Evidence record

Artifacts and frozen runs

Result artifacts

  • RELATIONAL_CONFIRMATORY_RESULTS.md
  • RELATIONAL_DIMENSION_DIAGNOSTIC.md
  • HODGE_BASELINE_RESULTS.md
  • MOVIELENS_RESULTS.md

Run records

No run-level packet is registered. The family retains code-level provenance, but public run reproduction still requires curation.

Evidence boundary

The synthetic advantage is specific to the declared two-dimensional bottleneck, and the MovieLens study is a retained negative result rather than evidence of universal relational benefit.

This registry links the experiments reported by the three books to their executable or archived code surfaces. A code link establishes traceability, not independent reproduction, correctness, or support for a claim beyond the experiment's stated boundary.