The Ideas Behind Spillway
Spillway is not just an email program with AI added to it. It is built around a view of how people deal with information, attention, uncertainty, relationships, and decisions over time.
This page is a map of those ideas. The deeper arguments live on the linked pages.
Email is a social system
An inbox is a record of people asking, promising, coordinating, persuading, sharing, and changing their minds. The meaning of a message depends on who sent it, what came before it, what project it belongs to, and what happened afterward.
That is why Spillway treats messages as parts of changing human systems rather than isolated pieces of text.
Knowledge should retain its source
Spillway tries to preserve the path from source material to observation, evidence, inference, and decision. That makes it possible to distinguish what was directly observed from what was inferred, what a person decided from what a model suggested, and what remains uncertain.
From Data to Knowledge is the canonical explanation of those layers. Explainable AI describes how retaining that structure can make an opaque model part of an inspectable system.
Human systems change
Relationships, projects, plans, and meanings evolve. Old evidence may remain historically true while becoming less useful for a current decision. New evidence can strengthen, weaken, or overturn an earlier interpretation.
Spillway therefore aims to preserve history without treating every stored conclusion as permanently current.
Attention is limited, and opportunity matters
“What matters?” and “what is useful to do right now?” are different questions.
A task can remain important while being a poor fit for the user's current time, device, location, or cognitive bandwidth. Conversely, a smaller task may be a sensible use of a short opportunity without becoming more important in the underlying model.
Different devices also create different opportunity structures. A phone, iPad, and Mac are not merely different screen sizes; they make different kinds of work practical. From Evidence to Action develops the social-science implications of attention, latent motives, constraints, and device affordances.
Behavior is evidence, not preference by definition
Opening, clicking, replying, deferring, or completing something can be informative. Those behaviors can also be produced by habit, friction, urgency, interruption, difficulty, or the opportunities available at the moment.
Spillway therefore tries not to optimize observable behavior as though it were the user's underlying goal. Computational Fallacies catalogs this and related failure modes.
AI is one inference mechanism, not the whole system
Some questions are best answered deterministically. Others require interpretation. Human decisions carry a different authority from model proposals. Reference Sources can supply private context a base model does not know. Different evidence can deserve different weight.
The goal is not to choose between rules and AI, but to use each where it is strong while preserving provenance and the ability to correct mistakes.
Why these fields belong together
Communication research studies how information moves among people and how context changes meaning. Political science studies institutions, incentives, uncertainty, and decision-making. Psychology examines attention, memory, motivation, and judgment. Survey methodology asks whether indicators measure the constructs they are supposed to measure. Statistics and data science help reason from imperfect evidence. Human-computer interaction asks how systems shape what actions are possible and how people experience them.
Spillway sits naturally at their intersection.
Why this project comes from a communication researcher and political scientist
Josh Pasek wanted Spillway because ordinary email was consuming too much attention and hiding too much work. His academic background shapes a second question: how should software reason about people, evidence, uncertainty, measurement, context, and change without becoming confidently wrong about what matters?
That is the intellectual thread connecting the rest of the Ideas & Science section.
From Evidence to Action · Computational Fallacies · From Data to Knowledge · Explainable AI · References and Further Reading
Documentation provenance: Iterative human–AI construction. See Documentation Provenance.