Agents suggest. The engine decides. The research asks what it takes to make that boundary hold, and how reasoning, evidence, memory, and learning can work within it.
How a proposal becomes a commitment, how plans earn authority, and where policy participates.
Deterministic convergence
Start with the foundation: a shared context, an explicit promotion gate, and nine invariants. The paper sets out the conditions under which a run reaches a unique result, with a trace of how it got there.
Read at Reflective LabsPlanning and review
A plan earns authority through admission, decomposition, collaborative reasoning, adversarial review, and simulation. This paper examines how those stages connect human intent to a governed commitment, while keeping learning separate from authority.
Read at Reflective LabsAuthorization and confluence
Policy changes as an organization changes. The paper places authorization alongside other contributors to the decision loop and examines the conditions policy must satisfy for the final result to remain independent of execution order.
Read at Reflective LabsWhat a result establishes, how constraints are checked, and which calculations can be replayed.
Constraint solving
A delivery plan needs to fit the available time, capacity, and resources. This paper brings constraint solvers into the same loop as language models, carrying a clear distinction between a proven optimum, a feasible solution, and a heuristic.
Read at Reflective LabsSymbolic assurance
An assurance claim becomes more useful when its limits are explicit. The paper uses symbolic solvers to search for violations of encoded rules and distinguishes that evidence from a proof validated by an independent checker.
Read at Reflective LabsReplayable analytics
Means, trends, rankings, and other direct calculations can contribute evidence to a governed decision. This paper explores analytics whose results can be reproduced from the same facts, with promotion authority held by the engine.
Read at Reflective LabsHow an organization brings past experience into a decision and learns from what follows.
Governed memory
Past experience should arrive with its source and a clear path into the current decision. This paper treats recalled material as a proposal subject to the promotion gate and examines how different retrieval rankings can be combined.
Read at Reflective LabsWorking with vague concepts
Organizations use words such as large, material, and strategic. Fuzzy inference makes those judgments explicit through membership functions and expert-authored rules, while preserving the distinction between a degree of membership and a probability.
Read at Reflective LabsModel provenance
A fitted model needs a traceable identity: its data, parameters, seed, and evaluation. This paper describes the implemented provenance boundary and a proposed training process within a governed Formation, clearly separating the two.
Read at Reflective LabsLearning about coordination
What would a model need to learn about the trajectory of an initiative? This paper develops the question, the data requirements, and a staged evaluation plan. It presents a research formulation, with training still ahead.
Read at Reflective LabsShared judgment
Good collective reasoning depends on people checking proposals and bringing relevant knowledge into shared consideration. The paper examines failures in that process and what an organization can learn from them, keeping the focus on the decision process rather than scoring individuals.
Read at Reflective Labs