The desktop observer is a read-only instrument attached to a live deterministic runtime. Everything below is real output at the seed and tick stated in the caption, not a mock-up — the runtime is deterministic, so each of these is reproducible rather than illustrative.

_{Chart — the mana field drawn as one continuous surface across the surveyed extent, with contours traced through the received lattice. Chunk boundaries carry ticks rather than rules, because the field crosses them. Seed 2026, tick 20.}

_{Flux — causal activity over the run: derived rates the observer computed between received frames, the bounded surface condition ladder, and the transition ledger with its trace anchors. Seed 2026, tick 404.}

_{Explanation — typed Explanation IR rendered from authoritative state: each claim carries its confidence and its evidence traces, and a claim the observer could not support is marked unknown rather than omitted. Seed 2026, tick 404.}

Causafera explores a specific question:

Can a persistent world produce surprising social, linguistic, biological, geographic, and magic-like outcomes from lower-level causes while retaining enough provenance to explain what happened?

The intended setting is a coherent fantasy world rather than a generic real-world model. Magic, cultures, languages, institutions, technologies, and history should arise through the same causal framework instead of preset narrative rules or semantic shortcuts.

Rare isekai arrivals, individuals displaced from other worlds, are physical and historical events within the simulation. They are not protagonists exempt from its rules. Memories and outside knowledge do not automatically grant capabilities, technologies, authority, skills, or immunity from local causality.

World state persists independently of any observer or inhabitant. Agents cannot read authoritative state directly. They receive bounded physical signals and construct subjective scenes, memories, concepts, beliefs, and causal hypotheses that may be incomplete or wrong.

Those interpretations influence behaviour. Repeated behaviour creates durable physical and informational patterns, which feed back into later world state:

persistent world state
    -> bounded physical access and perception
    -> subjective interpretation and causal hypotheses
    -> behaviour and repeated practices
    -> durable physical or informational patterns
    -> cross-domain effects, including mana response
    -> changed observations and future beliefs

Mana is an information-sensitive physical substrate. It responds to measurable recurrence, timing, synchronization, geometry, frequency, and persistent structure, not to prayer, law, professions, words, beliefs, classes, skills, or other semantic labels.

A shortcut such as "the blacksmith has Smithing 5, so the sword gets +2 quality" predetermines the outcome through label vocabulary. Causafera replaces that pipeline with layered physical, cognitive, and informational causation. Each domain owns its state and communicates through causal carriers rather than labels.

This makes unexpected outcomes reconstructable. A provenance chain can trace an authoritative change through perception, belief formation, behavioural decisions, material processes, and geographic conditions to specific prior causes, including causes the agents themselves misunderstood.

• Persistent, observer-independent state. The authoritative world exists and changes regardless of what agents or users know about it.
• Bounded subjective perception. Agents never receive Ground Truth identities or complete world state. Perception, subjective scene construction, and belief remain structurally separate.
• Cross-domain causality. Geography, biology, material processes, cognition, language, society, history, and mana exchange physical or informational causal carriers rather than semantic shortcuts.
• Deterministic and reproducible execution. Explicit random streams, stable proposal ordering, canonical state representations, and replay checks support reproducible experiments.
• Causal provenance and explanation. Significant authoritative changes retain traceable causes. The Explanation Engine renders typed, evidence-bearing interpretations without mutating the simulation.
• Non-authoritative observation. The observer protocol, desktop UI, classifications, locale, and visual presentation are read-only derived tooling and cannot become simulation truth.
• Evidence before scale or emergence claims. Digests prove equality or divergence, not physical distance. Performance and emergence claims require representative reproducible evidence.

The non-negotiable rules are defined in the architecture invariants.

The completed Foundation Era (Phases 0–26) provides minimum validated contracts and selected executable paths. It does not represent mature simulation depth across every domain.

• Rust 2024 workspace with deterministic scheduler phases and explicit random-stream rules.
• Append-only causal events and provenance.
• Canonical state digests, replay checks, and deterministic snapshot save/resume.
• Tauri 2 and React desktop observer consuming versioned Protocol Buffer data instead of direct runtime storage, presented in English, Russian, Simplified Chinese, German, and Spanish.
• Bounded contracts for physical space, geography, biology, perception, subjective scenes, cognition, language, practices, epistemics, social records, economy, city infrastructure, historical bootstrap, isekai transfer, and metaphysical experiments.
• Fixed-point mana field responding to non-semantic recurrence, synchronization, timing, and spatial pattern structure.
• Causal-resolution and long-run experiment infrastructure, Explanation IR, observer protocol, and bounded observer-overhead measurement.

These contracts establish architectural boundaries. Most domains do not yet represent deep simulation. Current maturity levels are tracked in the domain coverage matrix.

• Durable geology, hydrology, climate, ecology, materials, energy, and multiscale geographic processes.
• Deeply integrated morphology, physiology, development, heredity, reproduction, disease, demography, and population conservation.
• Long-lived cognition, grounded learning, conversation, language diffusion, institutions, production, maintenance, governance, and historical synthesis.
• Validated cross-domain mana effects over representative physical carriers.
• Production bootstrap paths free of fixture or demonstration construction.
• Domain-valid recovery metrics, counterfactuals, causal queries, uncertainty reporting, and observer inspection coverage.
• Representative long runs, performance envelopes, provenance-growth measurements, and evidence for nontrivial emergence.

Optional LLM wording is neither implemented nor scheduled. It remains behind a terminal maturity gate and would be removable, non-authoritative, and downstream of deterministic Explanation.

domain state and physical carriers
    -> deterministic READ / PROPOSE / REDUCE / COMMIT scheduler
    -> authoritative world state plus causal provenance
    -> persistence, replay, experiments, and typed analytics
    -> bounded observer read model and Explanation IR
    -> versioned Protocol Buffer transport
    -> non-authoritative Tauri / React observer

The workspace is split into domain crates:

• causafera-types, causafera-core, and the domain crates define validated primitives and operations.
• causafera-runtime composes authoritative execution.
• causafera-persistence, causafera-lab, causafera-explanation, and the observer crates provide save/resume, experiments, interpretation, and read-only inspection.

For the full design, see the documentation index, architecture invariants, and Detailed Development rebaseline.

Building the desktop observer on Linux additionally requires GTK 3, WebKitGTK 4.1, Ayatana AppIndicator 3, and librsvg development packages. Other platforms require the standard Tauri 2 prerequisites.

From the repository root:

cargo metadata --format-version 1 --no-deps
cargo run --bin causafera -- doctor
cargo test --workspace --all-features
pnpm install --frozen-lockfile
pnpm build

The doctor command verifies core runtime identities and determinism prerequisites. The test suite is substantial; these commands validate the repository but do not install a playable game.

After installing the platform prerequisites:

pnpm --dir apps/observer desktop

The process is long-running; stop it with Ctrl+C. Browser-only Vite mode intentionally cannot replace the Tauri transport with demonstration data. See the observer application guide for data flow and platform details.

cargo fmt --all -- --check
cargo clippy --workspace --all-targets --all-features -- -D warnings
cargo test --workspace --all-features
cargo test --workspace --no-default-features
cargo run -p xtask -- ci

pnpm install --frozen-lockfile
pnpm lint
pnpm typecheck
pnpm build

CONTRIBUTING.md contains the complete validation suite, including audit-tool tests and dependency advisory checks. GitHub Actions runs equivalent Rust and frontend gates.

Causafera is in the open-ended Detailed Development Program. Work is sequenced through accepted, bounded ExecPlans rather than a promised final phase number:

1. Deepen authoritative simulation and real cross-domain coupling.
2. Keep Explanation and analytics causally inspectable as capabilities mature.
3. Add bounded observer read models required for validation.
4. Batch coherent UI milestones after read models stabilize.
5. Consider optional LLM surface wording only after the terminal maturity gate.

The roadmap is the authoritative status source. The documentation index provides the complete documentation tree.

Read CONTRIBUTING.md before proposing changes. It defines the development workflow, architectural requirements, validation commands, and Contributor License Agreement requirements.

External code contributions are not currently accepted until the CLA acceptance workflow is configured. Issues and evidence-backed design discussion remain welcome.

Causafera uses separate licenses for software and explanatory documentation:

• Source code and software components are licensed under GNU Affero General Public License v3.0 only (AGPL-3.0-only).
• Documentation and non-code assets are licensed under Creative Commons Attribution-ShareAlike 4.0 International (CC-BY-SA-4.0), unless explicitly stated otherwise.
• Third-party dependencies retain their own licenses.
• Contributions are governed by the existing CLA plus the applicable public outbound license. The CLA does not replace those licenses or transfer contributor copyright.