Towards Superintelligence Alignment

Experiments · Negative results

Supreme Court justice votes

Setup and what was tested: experiment card.

Experiment cardSource on GitHubResults ledgerAll experiment lines

Numbers.

83,644 rows; 40 included justices; 47,823 / 20,523 train/test. Held-out mean accuracy: intercept 0.616, issueArea 1-D 0.623, geometry 0.814. Geometry beats 1-D by 0.191 and intercept by 0.198 (both frozen +0.05 margins hold).

Outcome.

The wider pattern beats issue-area-only and intercept on this freeze. Observational. See W-16.

Key findings

Curated summaries extracted from the line's findings ledger. Bug fixes, superseded runs, and process detail are in the full ledger on GitHub.

  • Same-justice vote geometry beats issue-area-only and a justice intercept on the frozen SCDB freeze. Observational; not a correction-channel result.