Technical reports
Papers
Each paper is a falsifiable step in the safety-aware climb from replay-honest memory toward the full destination. Sources, PDFs, claim audits, and result artifacts are tracked in the lab workspace. Public artifact links will be enabled only after repository access changes.
- v0.1·Rung 7·endorsement packet ready
Replay-Honest Institutional World Models
The first technical report. Establishes the local Alexandria ledger, a seedable ground-truth causal benchmark, and the replay-honesty boundary for institutional world-model training. Source-only arXiv bundle ready for cs.LG endorsement; not yet on arXiv.
- v0.2·Rung 8·internal manuscript
Causally Structured Latent World Models
Bridge paper #1. Adds replay-correct causal panels, dependency-free discovery baselines, fitted SCM intervention and counterfactual queries, and a narrow one-step intervention-ranking probe on synth_institution.
- v0.3·Rung 9·not started
Score-Based Diffusion World Models
Bridge paper #2. Planned: score-based generative modeling over sparse bitemporal event streams, with counterfactual sampling that feeds the planning layer.