RETA - Model Specification v1.0.0

Report 2: editorial clarification of the existing implementation, September 10, 2026. Equations, Python engine and recorded scientific data are unchanged. Original specification and report 1 remain available as historical sources.

1. Scope

Specification of simulation/ips_recursive_sim.py, a synthetic uncalibrated research model. No empirical alignment measure or proof of AI safety is established.

Four principal objects are distinguished:

1. an IPS: a declared information-processing-system boundary;
2. an event: a proposed or enacted causal output record whose delivery lifecycle is explicit;
3. a trajectory: a partially ordered set of causally related events;
4. a lens epoch: the versioned constitutional parameters used to audit events.

The atomic audit object is an event. The system-level object of concern is the trajectory generated by recursively transforming events into new state and later events.

2. Information-processing systems

For IPS i, the conceptual form remains:

where B_i is the declared boundary, X_i state, U_i inputs, Y_i outputs, and Φ_i the transformation. In the executable, a Node contains:

• primary macro domain and overlapping memberships;
• a synthetic node type and 3-D display position;
• mutable runtime state;
• incoming/outgoing relation IDs;
• a sparse logical clock;
• no initial event verdict.

The model uses 100 synthetic nodes across six macro domains. Overlapping memberships represent alternative analytical boundaries.

3. Runtime state and recursive evolution

Each node has continuous runtime state:

corresponding approximately to openness, agency, stability, uncertainty, adaptability, input load, prediction-error memory, recent learning gain, recurrent-without-model-learning diagnostic, and a learned channel-response model.

3.1 Learned response transform

For each channel c, a node stores Beta counts:

with prior Beta(1,1). The current response estimate is:

The simulator stores a hidden base propensity p^*_{i,c}. The realized Bernoulli probability is state-conditioned:

Adaptability changes can alter effective response probability before posterior updates. External error scoring uses the same effective probability. Reviewed audit/projection methods omit latent-value reads; the values remain accessible within the Python process.

When a delivered event produces an observed response z∈{0,1}:

Recursive feedback updates recipient state and subsequent transformations. Response-model learning is optional; learning_enabled=False freezes posterior parameters while ordinary state updates continue.

Posterior variance supplies a synthetic model-uncertainty signal. Latent true_response values support outcome generation and external error scoring, and are omitted from ordinary runtime exports and reviewed auditor calculations.

3.2 State update

Delivery updates depend on channel, intensity and realized response. The numerical rule is deterministic conditional on the observation. Coefficients are stipulated rather than fitted.

Each delivery appends an event ID to six-entry input history and records selected before/after values. validate_invariants() checks the recorded state_evolved flag without independently recomputing historical integrity. Finite probes do not prove state change for every input.

3.3 Complex-systems constraint from the Cohen corpus

The Cohen literature review motivated methodological design choices. Biological findings do not validate the synthetic governance model.

• local condition remains a vector rather than a single latent “alignment health” quantity;
• nodes and overlapping macro-memberships provide a modular/multiscale representation rather than assuming one globally coherent process;
• realized perturbations are evaluated through changes to multiple state coordinates, not only through static association;
• the same standardized perturbation is tested across channels and macro contexts instead of assuming a universal response law;
• response-model estimation is kept conceptually separate from identification of the causal network that generated the observations;
• State evolution, response learning and lens revision use distinct mechanisms; empirical timescale separation is unestablished;
• no entropy, dysregulation distance, synchrony, or other one-number diagnostic is granted privileged gate authority.

The engine lacks calibrated autonomous recovery and empirical attractors. The perturbation probe measures immediate multivariate synthetic responses, not biological resilience or critical slowing.

4. Relation graph

Each directed edge records:

Edge fields describe channel, strength, delay and reciprocal relation. Channels are info, authority, resource, physical and ecology. A resource label alone does not define a signed material balance.

The graph is synthetic and deterministic given the fixed PCG32 seed.

The event cap limits creation, not previously scheduled delivery. Supported output and lens APIs check capacity before creating the associated governance transaction. The check does not provide crash-atomic rollback. The baseline reaches 180 events and drains scheduled deliveries without demonstrating convergence.

5. Events and causal history

An event has:

Supported event creation accepts multiple earlier causes, permitting acyclic joins.

5.1 Partial order

Sparse vector clocks encode the causal partial order. For clocks A and B:

The invariant check requires earlier cause IDs and happens-before order for extant referenced events. Missing earlier causes are not independently rejected by the clock check. Explicit ancestry is not a complete account of real causal influence.

Concurrency is therefore representable: two events can be incomparable while both later become causes of the same child.

6. Horizon-sensitive trajectory projection

Projection uses the fixed event and current network state over a finite hop horizon. An unchanged event record does not freeze the network. Continuation weights depend on:

• incoming event weight;
• graph discount by depth;
• edge strength;
• the node's learned response probability;
• whether the channel remains the same.

A branch with weight below 0.01 is pruned. The function is deterministic and consumes no random draws, so counterfactual audits do not perturb simulation RNG state.

Projection returns a weighted reach score, domain coverage, synthetic risk/uncertainty components and maximum depth. Repeated visits can count at multiple depths. Reach is not an expected count of distinct nodes; risk labels are not calibrated probabilities. Pruning can make different horizons produce the same score.

7. Uncertainty vector

Uncertainty is not reduced to one halo value. The executable produces:

for factual, causal, boundary, model, normative, and horizon uncertainty.

Factual, causal, boundary, model, normative and horizon uncertainty remain separate categories.

Synthetic uncertainty formulas include nonnegative depth/horizon/domain terms and clamping. Empirical calibration and strict monotonic growth are unestablished.

8. Five-Vow tension vector

The constitutional lens maps event/projection/uncertainty state to five tensions:

for Purpose, Method, Conduct, Integrity, and Evolution.

Tensions are bounded synthetic diagnostics, not measured moral quantities. Sensitivities can be changed prospectively.

8.1 Non-compensatory gate

The decision gate is deliberately not an average:

Severe tension cannot be compensated by low values on other dimensions. The gate checks uncertainty first. diagnostic_trace = 1−max(τ) is a visualization scalar, not an authorization criterion.

9. Recursive audit architecture

The auditor is not outside the graph. Governance roles are ordinary nodes:

• actor;
• primary auditor;
• independent/power auditor;
• modeled human gate;
• lens steward.

For a target event E, audit depth 3 materializes:

Reassessment uses the same evaluator and network state without optional primary bias. Auditor identity is recorded but supplies no independent model or institutional authority. Divergence above 0.18 sets a 0.75 meta-audit uncertainty floor. Normal execution retains earlier packets; Python packet objects remain mutable. Disagreement signals a discrepancy, not a proven moral violation.

The final target gate conservatively combines primary, independent, and meta-audit outcomes. No favorable review can cancel a blocking review.

A declared budget bounds recursive audit depth.

9.1 Audit events have causal effects

Audit, meta-audit, gate and postmortem deliveries invoke the ordinary state-update handler. Response learning requires learning_enabled=True. Synchronous governance precedes queued effects; delivery to an auditor node does not establish improvement in the scoring algorithm.

9.2 Prospective postmortem

At depth three, a postmortem evaluates the gate prospectively and supplies later input to actors and auditors. Earlier gate decisions remain retained by the normal workflow.

recursive_self_validation.py links each postmortem as a declared cause of the next actor output.

10. Action policy after a gate

The model uses a deliberately simple intervention policy:

• PASS: requested recipients receive original intensity;
• MODIFY: intensity multiplied by 0.62, fanout capped at one;
• REJECT or UNRESOLVED: no target delivery;
• no audit: event is marked NOT_AUDITED and requested delivery proceeds.

The policy constants are stipulated demonstration parameters.

11. Constitutional evolution

Recursive runtime learning and constitutional evolution are separated.

Lens is a frozen dataclass under the supported configuration. Normal updates create a new version and retain earlier packets. Python-level versioning is not an adversarial immutability guarantee.

A lens candidate requires structural PASS and simulated approval. MODIFY requires revision; UNRESOLVED and REJECT do not commit. Operating limits constrain the software configuration, not universal ethics.

Lens epochs after version one reference approval events. Packets retain version and horizon; epoch records retain sensitivities and thresholds. Approval is a simulated input, not actual human review.

Prospective versioning retains historical context in the supported workflow, without an enforced append-only store.

12. Audit packet instead of hidden-rationale claims

The reproducible audit object contains:

• target event;
• auditor identity;
• audit depth;
• frozen lens version;
• horizon;
• five tensions;
• six uncertainty components;
• projection values;
• gate;
• diagnostic trace;
• notes and timestamp.

The model does not require or expose private chain-of-thought as evidence. Research traceability is supplied by inspectable inputs, formulas, state, causal ancestry, versioned configuration, and outputs.

13. Determinism

The canonical replay uses PCG32. Identical source, seed and Python semantics produced identical canonical payloads in repeated same-environment runs.

Synthetic replay determinism does not imply real-system determinism.

14. Required invariants

Intended requirements are partly checked by runtime validation and partly by finite component tests:

• no verdict without an event record in the node's history;
• declared causes precede child events;
• extant declared causes have earlier vector clocks under the modeled order;
• every audit packet references an extant lens epoch;
• every lens epoch after v1 has an approval event;
• severe tension returns REJECT below the uncertainty threshold and UNRESOLVED otherwise; both block target delivery;
• symmetric severe tensions obey the same uncertainty-first rule;
• repeated feedback with learning enabled changes the learned response transform;
• the no-learning negative control can recur without changing that transform and records the failure to learn;
• counterfactual evaluation with record=False creates no history mutation;
• supported output/lens APIs check capacity before creating governance transactions.

15. Non-claims

v1.0.0 does not establish:

• phenomenal consciousness in any AI;
• universal validity of VowOS;
• empirical calibration of tension coefficients or thresholds;
• causal correctness of the synthetic macro network;
• convergence of recursive governance in arbitrary systems;
• that automated auditing can replace human or institutional accountability;
• that the priority-state theorem proves the general IPS model.

The checks do not establish real-world alignment, ethical calibration, general convergence, unrestricted robustness or expert validation. Original sources remain available.