Deep Tech Research
11 reports and 4 notes on decisions, memory, and what a proof can establish.
- 01
Proposal, gate, commitment
Deterministic convergence as a substrate for organizational decisions
- 02
Institutionalized disagreement
How a plan earns the right to be run
- 03
Confidence is a certificate
Provable optimization inside a probabilistic loop
- 04
Recall is a proposal
Governed memory for multi-agent organizations
- 05
Searched, not verified
Bounded symbolic assurance for governed decisions
- 06
Nothing to fit
Closed-form analytics as governed Suggestors
- 07
Degrees that aren't probabilities
Fuzzy inference for organizational vagueness
- 08
Training as a Formation
Fitted models under governed provenance
- 09
Policy that participates
Monotone authorization as a precondition for confluence
- 10
What to learn before you train
Formulating an Initiative Trajectory Model, and why training waits
- 11
No room for slop
Individual and group failure modes an ever-learning organization is built to close
Notes
What is JEPA?
A world model that predicts in representation space instead of pixel or token space — and an open research question for us, not a shipped capability.
Why RAG, LoRA and MCP are not the right approach for organizations
Three popular patterns for extending a model — and why each one, taken as the whole architecture, quietly erases the boundary that makes a decision accountable.
Why context is more important than ever
When agents talk to each other directly, the conversation becomes the architecture. When they only read and write a shared context, the context becomes provable.
What is SMT, OR-Tools, and CVC5?
Three tools for three different kinds of certainty: exhaustive search for counterexamples, provable optimization, and the industrial solver underneath both.