Guides and research notes
Technical tutorials, comparisons, research notes, and change records.
-
Designing a hyper-specialized browser skill pack
Define bounded roles, compatibility, evidence, authority, and retirement rules before choosing model files.
-
Why twelve installed models should not mean twelve active models
Separate installed capacity, active residency, and concurrent execution before estimating a modular tiny-model system.
-
MiRust Teleodynamic AI research expansion 1.2.0
A dedicated Teleodynamic AI guide, deep documentation section, editable evidence matrix, research notes, source boundaries, and new visual components.
-
Auditing the emergent-halt claim in Teleodynamic Learning v1
The paper’s empirical narrative argues for endogenous freeze, while its appendix proof guarantees eventual freeze through explicit structural caps. This note defines the evidence needed to separate the two.
-
Teleodynamic AI and tiny language models: controller, substrate, or metaphor?
A tiny language model may provide predictions inside a teleodynamic system, but size, local inference, and quantization do not satisfy the paradigm by themselves.
-
Designing a teleodynamic event ledger
An append-only decision record makes path dependence, local rationality, resource coupling, and phase claims inspectable and replayable.
-
Reproducing DE11: evidence plan and limitations
The Distinction Engine results are a starting point for reproduction, not proof that a browser runtime or language model exhibits teleodynamic learning.
-
Mapping browser resources without pretending telemetry is thermodynamics
A browser can expose useful application-owned measurements, but it rarely exposes direct device energy, thermal headroom, or complete memory pressure.
-
A timeout is not an emergent halt
Structural freeze is an endogenous action-selection claim; timeouts, memory caps, and manual stops remain external controls even when they are necessary.
-
The five commitments of Teleodynamic Learning
A practical reading of the five conditions that distinguish the published paradigm from generic resource-aware or adaptive software.