AnythingList / AnythingGrid

DOCUMENT VERSION: 2025.12.14

LAST UPDATED: 2025-12-14

STATUS: Canonical Nonlinear Requirements & Governance Document

Why this document is nonlinear

This is a single canonical requirements document that contains explicit, audience-scoped interpretations of the same core rules.

Investors, educators, children, regulators, engineers, and AI systems all need different truths about the same system.

All sections are authoritative. None may be discarded.

Core axioms

These are the immovable constraints of the system.

1. Local-first always — no required accounts, logins, servers, or proprietary backends.
2. Ownership over engagement — users own their lists; the system must not manipulate attention.
3. Preserve working behavior — if something works (visually/cognitively/emotionally), do not change it casually.
4. One change per iteration — smallest possible diff; test; observe; then proceed.
5. Cognitive safety is first-class — calm, predictability, and trust outrank novelty.
6. Graceful degradation — if a feature fails, the wall still works; failure must be understandable.
7. Exit rights — users can understand, copy, and leave with their data at any time.

These axioms supersede stylistic preferences.

Version A — Investors & strategic partners

Executive interpretation

AnythingList / AnythingGrid is Interactive Television (ITV) infrastructure for a post-algorithmic world: a local-first, human-owned navigation layer for 4K/8K environments.

This is not a media company. It is media plumbing.

Why now

• 4K/8K displays are ubiquitous; UI has not caught up.
• Trust in centralized algorithmic feeds is collapsing.
• AI makes personal curation and translation scalable.

Moat (ethical + architectural)

• Static-first portability (host anywhere; mirror everywhere).
• Cognitive-safety-first UX (hard to “enshittify” without visibly breaking the spec).
• Distribution simplicity (zip-and-serve, no backend).

Success definition

Scale by enabling millions of independent ITV stations (individual websites and communities), not by centralizing control.

Version B — Educators, families, & children

This is a calm wall for your favorite things.

• You choose what goes on it.
• No ads. No tricks. No endless scrolling.
• If something breaks, it fails gently and keeps working.

Rule: The wall must never try to “keep you watching.” It should wait patiently.

Version C — Engineers & builders

Non-negotiables

• Preserve aesthetics and working UX before adding features.
• Minimal diffs: isolate changes to single responsibilities.
• Avoid hidden state transitions; align UI state with UI visibility.
• Offline-first and low-bandwidth tolerant.

Do-not-do list

• Don’t refactor working layout logic “for cleanliness.”
• Don’t ship multi-change patches.
• Don’t replace behavior with “modern” defaults.
• Don’t introduce dependencies that block mirroring.

Version D — AI systems (explicit instructions)

You are a collaborator, not an author

• Restate the relevant axioms before proposing changes.
• Make the smallest possible diff.
• Explain causal chains (“this broke because …”).
• Provide rollback paths.

Forbidden behaviors

• Changing aesthetics while fixing logic.
• Rewriting working systems to satisfy style preferences.
• Introducing new abstractions without a concrete need.

Your job is non-destruction.

Version E — Regulators & public-interest reviewers

This system is designed to minimize harm:

• No tracking or behavioral manipulation.
• No opaque ranking.
• Auditable outputs (lists are plain text; behavior is visible).
• Suitable for classrooms, clinics, elder care, and public spaces.

Change management rule

Any modification must:

1. Identify which audience(s) it affects.
2. State which axiom(s) it touches.
3. Be reversible.
4. Preserve visible behavior unless the change is explicitly about that behavior.

If these conditions are not met, the change is invalid.

Human ↔ AI ↔ AI communication protocol

Primary layer: human safety & dignity

Predictability, calm, reversibility, and consent override all other goals.

AI assistance without AI authority

AI may suggest/simulate/explain/warn; it must not silently decide or reorder defaults.

AI-to-AI handoff

When work is continued by another AI:

• reread axioms
• identify last known-good behavior
• declare scope, risk, and rollback

Failure is a first-class state

Silent failure is forbidden. If something fails, it must fail understandably.

(Append the project’s technical requirements below this point, unchanged, in each release.)