Simulator Specification

Status: implemented conceptual specification. This is a buildable pattern, not a result of the completed tests. It does not claim that the physical universe, a mind, or consciousness uses this architecture.

The browser simulator explicitly reports a RETURN_WRITE handshake. It is not Test 05 evidence (and must not be described as such). The display is an executable toy model, not a physical-time, cosmology, consciousness, or resource-advantage claim.

RETURN_WRITE handshake

The fixture separates seed-supplied facts, beginning requirements, return-supplied requirements, temporal obligations, endpoint requirements, and writer policies. A finite frozen candidate domain is enumerated in canonical order. Each candidate is applied hypothetically at the beginning, forward execution derives an ending proposal from that returned content and explicit writer-eligibility predicates, and an independent sealed-history validator checks typed events. The implemented ablation demonstrates that the beginning fact determines the returned content; it does not establish a general theory of motive or causation. Closure and uniqueness are separate: histories are classified NONE, UNIQUE, or MULTIPLE; uniqueness is required only when the fixture says so.

Events carry an effective tick, authored-at tick, provenance, and append identity. Authored-at metadata is provenance, never proof of ending authorship. Each reconstruction mode enumerates candidates independently. Reciprocal closure is reported separately from information loss and semantic continuity through loss. Observer projection is an explicit whitelist of at most two frames and does not expose ticks, phases, endpoints, provenance, contracts, ledgers, candidate data, or handshake metadata. Byte/resource values shown in the UI are illustrative serialized sizes, never benchmark evidence.

Purpose and boundary

The simulator studies whether a bounded executable can preserve a specified world payload while it splits work and reunites work. An optional mode discards lower-level detail to test continuity through loss. The payload is operational: identities, relationships, obligations, unresolved events, and plot phase. “Narrative” means state and dependency bookkeeping, not subjective experience.

The Animus Omega Cycle is the public name for the stronger proposed branch-reconvergence version; its technical subtype is the Animus Omega branch-reconvergence cycle. The browser includes a separate deterministic teaching trace that illustrates this proposed architecture; this built-in trace is not experimental evidence and does not show that an independent run achieved reconvergence. The cycle begins with one authoritative history, permits declared represented branches during expansion, and reconciles them under a frozen narrative-equivalence projection during contraction until one canonical history remains. “Narrative collapse” means this deterministic semantic reconciliation; it is not a physical, quantum, or consciousness claim. Candidate enumeration and observer threads do not count as represented timelines.

The teaching trace is separate from the trunk closure run. Its stages show one authoritative history expanding to four declared branches, grouping under a frozen narrative-equivalence projection, contracting within equivalence classes, and ending at one canonical history. Every merge audit lists retained invariants, discarded distinctions, conflicts, and provenance. Its scorecard reports branch reduction, information loss, continuity through loss, and reciprocal closure as four separate fields; reciprocal closure is deliberately not evaluated by this trace.

The proposed narrative-closure boundary is an untested internal constraint: an ending-to-beginning pathway must supply the conditions required by the beginning, while the boundary remains an event in one continuous simulated history. This is not a physical time loop, backwards causation, consciousness, or evidence about the universe.

State model

Trunk

The trunk is the sole authoritative state:

  • a monotonic world clock and configuration;
  • entity and identity store;
  • relation ledger (edges, obligations, ownership, and provenance);
  • unresolved-event and causal-dependency ledger;
  • plot-state ledger, including open arcs and required next-scene conditions;
  • resource accounting and versioned schema.

Threads never commit directly to the trunk. Every mutation is an event with source thread, logical time, input version, and deterministic conflict metadata.

Observer and region threads

At split, the scheduler creates observer, region, or inhabitant threads. Each receives a versioned partial view and a bounded working buffer. A thread may read only its view, emit proposed events, and maintain local microstate. It does not receive an implicit full copy of the world. Views must identify their omissions and stale reads.

Protected narrative ledger

The protected ledger is the minimum state that may not be silently dropped: stable identity mappings; relation edges and their provenance; obligations and promises; unresolved events; plot phase and open dependencies; and the conditions marked as required for a future scene or closure boundary. Each entry has a schema version, owner, last authoritative update, and retention reason. A dump may reject an entry only through an explicit, recorded policy.

Dump

dump(thread_state, policy) produces a residual plus an audit record. It may either preserve state losslessly or unload inactive microstate or summarize it. Information loss is not required for cyclic operation. A lossy mode must report discarded fields, retained fields, transforms, and reconstruction assumptions. Dump cost includes serialization, summary computation, storage, I/O, and information loss. A successful dump is not proof of fidelity.

Merge and still

merge(residuals, trunk_version) synchronizes threads at a barrier, validates versions, orders events deterministically, applies the declared conflict rule, updates the trunk, and records discarded detail. The default conflict rule is “reject ambiguous writes”; alternatives require a versioned specification.

still is a merge with new event intake paused: threads drain or cancel according to policy, residuals are committed, and the trunk emits a quiet checkpoint. Stillness is an engineering state, not a claim about meditation, love, or physics.

Animus Omega Cycle (optional branch reconvergence)

When the Animus Omega Cycle extension is enabled, every branch has a declared origin and unique identity. The frozen narrative-equivalence projection identifies the protected identities, relations, obligations, causal dependencies, permissions, ownership, and observer-visible consequences that must agree. Reconciliation proceeds in declared stages, records conflicts and provenance, and must end with one authoritative history before reconstruction and return.

The run reports branch counts and equivalence classes at every stage. Mapping distinct branch states to one retained state is information loss even when all protected narrative invariants survive. Branch-state reduction, continuity through loss, and reciprocal closure are therefore separate outcomes.

The Animus Omega Cycle teaching trace illustrates this contract, but a reserved branch-reconvergence experiment requires a separate, prospectively locked protocol package. Before execution that package freezes:

  • canonical fixtures and digests, branch origins and immutable identities;
  • branch-local initial state, permitted and forbidden transition tables, and arm-blind transition code;
  • the total narrative-equivalence projection, including protected and ignored fields;
  • conflict classes and their deterministic, content-independent dispositions;
  • merge ticks, resource limits, observation horizon, and the exact one-history termination rule;
  • matched positive and negative controls, thresholds, replay count, exclusions, and fail-closed rules.

The minimum control set includes: no cut, to prevent credit for an always-single history; a valid equivalent pair; an undeclared hidden trunk/full-state copy; scripted convergence that reads an arm label or expected output; arbitrary branch deletion; and an unresolvable protected-field conflict. Control arms use the same scheduler, budgets, validator, and scoring implementation as the positive arm except for their named intervention. A failed control invalidates the run family.

The experimental report must score branch creation, branch-set reduction, information loss (YES or NO), continuity through loss, deterministic replay, and reciprocal closure independently. Reconvergence additionally requires no hidden executable trunk copy, arm-blind transitions, justified deterministic reconciliation, complete ancestry, and exactly one authoritative history at the frozen termination tick. Selecting a survivor, timing out, or deleting an unresolved branch is not reconciliation.

Reserved fixtures and seeds must be disjoint from harness dry runs. Once the protocol package is time-locked, any change to fixtures, projection, code, thresholds, schedule, exclusions, or failure rules creates a new protocol and new reserved run. Only pre-execution corruption, digest mismatch, unavailable declared runtime, or failure before the first branch transition may be excluded. Post-start crashes, timeouts, budget overruns, unresolved conflicts, replay disagreement, missing provenance, control failures, and validator errors count as failures. The reserved run starts only after an independent gate verifies the receipt, digests, complete arm matrix, and all frozen decision rules.

Cost model

Every run reports wall-clock time separately from semantic outcomes and accounts for thread CPU, memory and retained context, view construction, serialization, dump storage, merge/barrier time, reconstruction, conflict resolution, and discarded information. Resource comparisons must define the budget and task fidelity in advance. A torus, compression, or state reuse does not remove compute, storage, or thermodynamic costs.

Invariants and acceptance checks

  1. Only the trunk can authoritatively advance world time or protected-ledger entries.
  2. Every committed event has deterministic provenance and is applied once.
  3. Split/merge preserves identity keys and ledger referential integrity.
  4. No protected entry disappears without a typed loss record and policy.
  5. Conflicting writes are rejected or resolved by the frozen rule, never by iteration order.
  6. Replay from an identical trunk, residual set, configuration, and seed is identical within the declared semantic state and observation horizon.
  7. Dump and merge are auditable: input/output digests, costs, losses, and conflicts are recorded.
  8. Narrative closure, if implemented, must satisfy both forward causal continuity and the explicit return constraint; it is not inferred from a cycle alone.
  9. A branch-reconvergence result is ineligible unless its prospective protocol package passed the time-lock gate before any reserved branch transition.
  10. Branch reduction, information loss, continuity through loss, deterministic replay, and reciprocal closure are never collapsed into one success label.

Failure modes

Test and report at least: stale-view writes; duplicate or lost events; broken identity joins; contradictory relations; unresolved merge conflicts; ledger overflow; summary hallucination or under-specification; reconstruction drift; unbounded thread or buffer growth; deadlock at the barrier; nondeterministic ordering; hidden full-world copies; resource-budget overruns; and a closure boundary that exposes a seam or fails to supply its declared initial conditions. Failure is informative and must not be relabeled as fidelity.

Evidence boundary

The public record describes this architecture as a proposed synthesis. Tests 01–04 are separate operational studies; they do not validate the combined simulator, its narrative closure, comparative fidelity, information creation, or a physical-universe claim.