Activation geometry / training governance / alignment evaluation
Arca Futura develops systems for reading, evaluating, and governing model behavior at the structural level - from depthwise activation geometry to diagnostic alignment suites.
Training-state governance built around sensing, authorization, and bounded response.
A 77-model preprint study of recurring depthwise activation landmarks in decoder-only LLMs.
Landmark-aware ablation work testing whether depth coordinates reveal load-bearing attention heads.
Founder, Arca Futura LLC
Earlier projects explored memory, tutoring, and adversarial evaluation. The current work is narrower and stronger: measuring the internal geometry of model training, turning those signals into governance tools, and publishing diagnostic benchmarks with careful evidence labels.
Career transitions
Üppward helps people explore possible career paths against real opportunities and organize the work of a transition. It keeps people in control of their direction and next steps.
Visit ÜppwardCurrent focus
AEGIS is a training-state governance system with three separable functions: observe model-internal change, decide whether the evidence warrants action, and route a qualifying event into a bounded response selected by the training team.
Repository-audited studies support prospective transition localization, zero actions across 18 stationary, clean, and difficult-but-valid negative controls, and matched containment benefit in three Qwen and Llama cohorts through 4B parameters. The demonstrated response is useful and bounded; it is not claimed to be optimal.
Each layer is tested separately so a signal is not mistaken for permission.
5 / 5 Qwen, 5 / 5 Llama, and 3 / 3 Qwen3-4B pairs benefited at their own registered endpoints.
Timing optimality, low-loss harm, recurrent protection, and distributed efficacy remain open.
Public research
The Hau Curve v2 paper identifies a recurring tri-phasic activation geometry across decoder-only transformer models, with early-training convergence and stable depthwise landmarks. This work is the foundation for AEGIS and the later functional-anatomy studies.
This follow-on study asks whether landmark-relative depth coordinates help identify attention heads whose causal importance is not explained by salience alone. It frames the Hau Curve as a practical coordinate system for transformer functional anatomy.
StratEval is a diagnostic evaluation suite for instrumental convergence and strategic misalignment under structured pressure. It is meant to characterize elicited behavior under authored scenario frames, not rank models for deployment.
Pipeline context
Source refinement and motivational signal scoring, developed by ZeroThreeSixNine Ventures.
Activation-geometry monitoring and training governance, developed by Arca Futura.
Motivational alignment architecture under active research and validation.
PYPER3 is best described publicly as a collaboration and research pipeline. Some materials are investor-facing or confidential; this site names the architecture without importing private pricing, customer, or distribution claims.
Earlier prototypes