How persistent software agents could become operating firms, what lowers their viable scale, and where legal accountability still requires humans.
How reader-centered calibration and probabilistic programming solve complementary parts of uncertainty-aware language generation.
Expressing generative models as programs and separating model structure from inference machinery.
How an agent preserves useful state across conversations without treating the entire past as equally relevant.
How trading rules, information, inventory, and strategic behavior produce observed prices and liquidity.
Durable, inspectable, versioned context that survives individual chats and can be mounted by agents and tools.
Why publishing a knowledge base requires synthesis and declassification rather than exposing a private graph.
Updating uncertainty with evidence while keeping assumptions and prior information visible.
What can remain continuous when an agent's model, context window, tools, and runtime all change.
Tracing how prompts, memories, model calls, tools, verification, and external effects form one causal agent run.
Replacing part of a long agent history with a smaller working representation while preserving durable evidence outside the prompt.
Constraining language-model generation so downstream systems receive valid, schema-shaped data.
How agent work resumes after retries, interruption, duplicate delivery, and partial failure.
What an ATProto application should do when a user edits a record after the application has approved or admitted a specific version.
A proposed ATProto record pattern for admitting member-owned content into a forum category without transferring ownership.
A practical guide to finding an ATProto record, identifying its publisher, and preserving the exact version you saw.
A map of this wiki's notes on ATProto identity, repositories, records, strong references, permissioned data, and application-level authority.