ss-disentangled-direct-pretrained-v1

Status: complete - 5/6 runs analyzed; sorting_algorithms failed with CUDA OOM.

This collection is the first direct-pretrained disentangling diagnostic on the synthetic task suite. It establishes the baseline failure mode for the family: final-answer accuracy saturates quickly, but router and posterior probes remain only moderately strategy-aligned under a reconstruction-plus-KL objective.

• design
• results
• analysis

Artifacts

• notebook: path:experiments/synthetic_sequences/disentangled/analysis/notebooks/10_apr13_ss_direct_pretrained_v1_analysis.ipynb
• aggregates: path:experiments/synthetic_sequences/disentangled/analysis/aggregates/ss-disentangled-direct-pretrained-v1
• exploratory figures: path:experiments/synthetic_sequences/disentangled/analysis/aggregates/ss-disentangled-direct-pretrained-v1/figures

Context

• baseline for ss-disentangled-direct-pretrained-pv-rmkl-sweep-v1
• contrasted with synthetic-pretrained-disentangled-from-entangled-v1

Runs

• slurmkit collection: path:.slurmkit/collections/ss-disentangled-direct-pretrained-v1.yaml
• run count: 5/6 complete, 1/6 failed
• date range: 2026-04-13