What Temper is building toward

Attention is the teleological anchor

Everything here follows from a single commitment: attention is our most precious resource, and a system that touches it owes it an ethic. What attention is, why wasting it is not merely inefficient, and the principles that fall out — that is the attention manifesto.

The pages under /theory are the other half: the semantic model that commitment requires. Temper is built around what falls out of taking attention seriously, and in that frame I am committed to a few key principles.

Forgetting is what makes memory meaningful

Memory in an information system is only as effective as its mechanics for forgetting. Without decay (time-and-distance fading), active deformation (the revision and topological change that follows from new decisions or new data), and scarification (the trace of past mistakes carried into future engagement), signal washes out in noise — and attention can no longer land on what matters now.

A system that lets everything stay equally findable, equally true-sounding, equally clean has not committed to attention; it has only committed to retention.

The system must be capable of deforming itself

An information system must be able to change its own shape — actively, when decisions are codified into the substrate, and in aggregate, when thresholds are crossed or patterns arrived at in practice. The information landscape is not static. Decisions supersede earlier ones; concepts merge or split; tolerances shift; what was once peripheral becomes central.

A system that requires every change to come from an explicit author act has externalized work the system itself should do.

Data is only information when it has a frame of reference

Data is not information. Data becomes information when it is engaged from a particular perspective, at a particular time, at a particular resolution, for a particular purpose. What matters, to whom, at what level of granularity, on what time horizon — these are not secondary concerns to be layered on top of "the data." They are what makes data legible as anything at all.

Projections and summaries are different acts

A projection is what happens when a perspective's intention queries the substrate and renders what is relevant from a position in perspective-space. It is lossy by necessity, perspective-laden by construction, but it does not claim to be a new resource.

A summary is something else: a deliberately crafted artifact that selects what to include, weights what to emphasize, and manages the boundary of what gets disclosed to inform a decision. Summaries carry intention and implicit analysis; they are not value-neutral views of the substrate.

Both projections and summaries are necessary. But if the system cannot distinguish them, every summary becomes an unmarked reweighting of how the underlying information will be read — and decision-making compounds on top of views that look neutral but are not.

Events and observers weave through every kind of work

Customer discovery, research, requests, analytical insights, recommendations, prototypes, tickets, pull requests — every kind of work artifact carries topics, scopes, and relationships that make it more or less useful to different consumers at different times. The operational pattern connecting them is not a separate workflow layer bolted on top; it is the events-and-observers structure of the substrate itself.

Resources are not free-floating. They are situated, scoped, related — and only when those properties are first-class can the system route attention to what matters now, for whom, and at what resolution.

The system does not store knowledge

These are not features I want to ship. They are commitments I want to keep. There is a separate semantic model where the architectural detail lives — what the manifold is, what fields are, how forgetting works geometrically, how perspectives are characterized. The pages under /theory introduce that model.

One commitment in the model runs through everything that follows: the system stores data and traces of past intentional acts — recorded questions, notes, decisions, which themselves become further data. It does not store knowledge. Knowledge is produced in the generative space defined by a perspective-as-intention and the information-from-data that this perspective describes, transformed into a cognitive or operational artifact - a tool, in effect, that can be used.

The label "knowledge base" is a misnomer in light of this. Knowledge is always potential, never actual, until activated by a perspective. The system's job is never to be right about what something means — meaning is not in the system. Its job is to faithfully represent data, faithfully record intentions, and faithfully compute projections such that perspectives engaging with those projections are well-equipped to produce knowledge.

The shape of the model

The pages here introduce the model in the order the source document does. Each is short; each can be read alone; the sequence is the most coherent path through.

• Ontology — Data, intention, information, knowledge. The stratified layers.
• Manifold — Positions, fields, streams. The geometry.
• Time — Time as a primary axis. Events-as-primary. Why this is a substrate commitment.
• Deformation — Forming and forgetting. Strong vs. weak. Scarification. Self-cohesion and relaxation.
• Perspectives — Trajectories, not points. Role-perspective vs. individual. Access vs. expertise.
• Translation — Why translation is irreducible. Bridges. Knowledge as relationship.

Two reference surfaces sit alongside:

• Schema — The structural codification: entity types, event structure, topic taxonomy, resolved stances. Work in progress.
• Open questions — What is not yet settled. Updates over time as items resolve.

The model is provisional. It captures a mental picture clearly enough to be argued with.

There is an interesting symmetry that emerges - the perspective-side has roughly the same shape as the data-side: discrete deformations, continuous trajectories, characteristic decay rates, prior-and-likelihood structure. Resources and perspectives are both on the manifold; both have positions, trajectories, and decay; both can be deformed strongly or weakly; both have spatial profiles for their effects. This is some evidence the primitives are well-chosen rather than ad-hoc — a model that requires fewer primitives to describe more phenomena is more likely to be sound.

These pages are the model. What it is ultimately for — and why attention deserves an ethic rather than an optimization — is the attention manifesto. The model is held lightly; the commitment underneath it is not.