Reusable decision strategies for FlossWare.
This repository contains interchangeable strategy implementations. It is not a model router and does not define the Loom execution model.
stable strategy contract
|
v
strategy
|
v
decision + provenance + outcome
Strategies choose among feasible alternatives. They do not own authorization, policy, credentials, billing, quota enforcement, or safety constraints. Those constraints must be applied independently and cannot be overridden by a learned strategy.
- Thompson Sampling
- epsilon-greedy
- deterministic baselines
- future UCB and contextual-bandit strategies
Thompson Sampling can support adaptive model or resource selection, but it remains a reusable strategy rather than a special Loom primitive.
Loom defines execution semantics and the contracts through which capabilities are invoked. This repository supplies strategy implementations that a Loom Worker, Arbiter, or other compatible caller may use. It does not own orchestration or provider invocation.
Optimization algorithms may be implemented in optimization and used by strategies where appropriate. Strategy selection and optimization algorithms are related but distinct responsibilities.
The repository is named strategy. The current distribution/import identifiers remain strategy-ai and strategy_ai; repository naming does not automatically rename a published package or Python imports. Change those identifiers only through an explicit compatibility and release decision.