Method

One model, many governed projections.

This site is a small instance of a larger idea: fundamental-engine, a representation-independent substrate that separates what a system is from how it is shown. The vocabulary below comes from Foundations of Computational Interaction; the point is discipline, not jargon.

The separations

Most software binds the domain model, its state, its rendering, its available actions, and its telemetry together — so a visual change quietly alters the model, the action set, or the instrumentation. This site keeps them apart:

model commitments ≠ state ≠ operational semantics ≠ projection ≠ host representation ≠ interpretation ≠ behavior

Model & state live in the archive (the record of what exists), not in the page.
Semantics — the meaning of a section, the accent of a piece — is a declared layer (the Semantic Article Field), separate from both the store and the rendering.
Projection & representation are this client's job. The same content is projected many ways; each declares what it keeps and drops in /projections.json.
Interpretation & behavior are not asserted from the runtime — see the boundary below.

FCI Audit §12 · Empirical Ablation Benchmark

Interactive Architecture Ablation

Toggle system layers to measure what each separation buys against a simpler, coupled architecture.

Decoupled Model & Archive SSOT record of facts (`zach-core`) separate from view rendering

Governed Projections Explicit preserve/omit contracts (`/projections.json`, `/claims.json`)

Semantic Article Field Argument roles, receipts, concept threading, and claims verification

Reciprocal Physical Coupling Two-way feedback: DOM elements bend the field; field bends them back

Claim C6 Support Score

0.78 Supported

Agent Token Efficiency 94%

Confound Isolation 100%

Complexity/Value ($U$) 0.45

System at Full Integrity: Model commitments, projection governance, and reciprocal physics operate without confounding presentation with ground truth.

Governed projections

The reading page, the particle field, the structured data, /llms.txt, /rss.xml, the sitemap, and the REST API are all projections of one model. A projection is replaceable but never neutral — "shows the same world" is not enough — so each one declares its target, mapping, and what it preserves, transforms, omits, and introduces at /projections.json. Where the meaning layer is machine-generated, the entry carries its own evidence chain (data.generated: which fields, by what, when).

Professional claims have a separate source of record. The cleared public projection at /evidence.json carries claim provenance, observation dates, and limitations. Editorial selection ranks that evidence without changing the underlying facts.

What this site does not claim

The honest boundary. None of the following is inferred here:

That you were shown a thing means you understood it. Display is not comprehension.
That a control is visible means an action is feasible or believed. Exposure is not affordance, permission, or belief.
That a number is a truth. A metric (reading time, field density) is not a validated construct.
Anything about who you are. Aggregate analytics and bounded navigation events describe routes taken; they do not establish comprehension, belief, identity, or cultural or demographic characteristics.
That the motion means what the text means. Motion is decorative-but-meaningful and never the sole carrier of information; it is omitted entirely under prefers-reduced-motion.

These are the model's falsification lines, kept deliberately. This site is graded against a scientific-audit standard as an M2–M3 conceptual/infrastructure instantiation — its conceptual claims hold, its scientific-value claim is explicitly undemonstrated. The machine-readable records: /claims.json (claim inventory + support scores), /measurements.json (what the numbers do and don't mean), /projections.json (projection contracts). Full audit in the repo at docs/computational-interaction-audit.md.