Reality Admission
for Intelligent Systems.

Computation can propose reality
without owning it.

Autonomous systems can observe, reason, generate, call tools, compose workflows, and propose actions across enterprise boundaries. Existing security controls govern important prerequisites such as identity, authentication, access, network behavior, endpoint integrity, application permissions, and detection.

Otarx addresses a different control problem: whether a consequence proposed by computation should be allowed to become canonical enterprise reality.

Otarx separates proposal from realization. An action may be computed, requested, and evaluated without acquiring the authority to mutate protected canonical state.

The runtime evaluates a proposed consequence under a consequence-specific Reality Contract. The contract determines which enterprise perspectives are relevant, which conditions must be established, which contradictions matter, and what evidence is sufficient for admission.

Otarx then resolves relevant observers and authorities, evaluates their claims, converges evidence into a bounded representation of the reality surrounding the proposed consequence, and produces a signed Reality Decision.

Only a valid, consequence-bound ADMIT decision accepted at an Otarx-controlled enforcement boundary permits the protected consequence to become canonical.

Access is not the same
as legitimate consequence.

Traditional security architectures commonly establish whether a principal can authenticate, access a resource, invoke an operation, or enter a protected environment.

Those controls remain necessary. They do not necessarily establish whether the specific consequence being proposed is consistent with the surrounding enterprise reality.

An authenticated principal can still act outside valid authority. A valid credential can be stolen. An autonomous agent can execute an incorrect plan. A compromised application can issue syntactically valid requests. A model can reason from incomplete or misleading context.

Otarx therefore treats access to the claim interface and authority to create canonical consequence as separate properties.

Identity establishes who is presenting a claim. Reality admission establishes whether the enterprise conditions surrounding the proposed consequence are sufficiently supported for that consequence to become canonical.

S0, proposed consequence,
decision, S1.

Let S0 represent the current protected canonical state relevant to a proposed operation.

A caller proposes a consequence that, if realized, would produce a potential successor state S1.

The proposal itself does not authorize that transition.

If admission returns DENY, required proof cannot be established, or the resulting decision cannot be validated at enforcement, the protected state remains S0.

Describe reality.
Propose consequence.

Otarx distinguishes claims about existing enterprise reality from claims proposing a change to that reality.

Define what must be true
before asking whether it is.

A Reality Contract defines the proof obligations for a class of consequence.

Different consequences can require different evidence families, authority relationships, temporal conditions, contradiction handling, and levels of assurance.

A Reality Contract can define:

• target object or protected state domain,
• requested operation,
• consequence level,
• required observer families,
• authority requirements,
• relationship requirements,
• mandatory conditions,
• supporting evidence,
• contradiction rules,
• anomaly handling,
• freshness and temporal requirements,
• purpose or intent requirements,
• sequence requirements,
• bounded realization scope,
• decision validity rules.

The caller may provide evidence references. The caller does not determine what evidence is sufficient. The Reality Contract does.

Determine whose perspective
matters for this consequence.

A Reality Contract establishes the categories of enterprise perspective required to evaluate an Action Claim.

Otarx resolves the observers and authority sources capable of establishing those parts of reality.

The relevant observers are consequence-specific. A payroll change may require different perspectives from a production deployment. A financial transfer may require different authority and relationship evidence from a patient record mutation.

Observer resolution therefore occurs before evidence convergence. Otarx first establishes which perspectives are relevant, then asks whether those perspectives support a coherent bounded reality.

A consequential claim should survive
independent enterprise perspectives.

Otarx can evaluate Observer Claims and evidence from multiple independent enterprise sources rather than relying on one credential, one database, one agent, or one model output.

Relevant evidence families can include:

• source reliability and attestation,
• identity evidence,
• authority lineage,
• organizational and object relationships,
• timestamps and freshness,
• device and session context,
• contract conditions,
• declared purpose,
• expected sequence,
• supporting contextual evidence,
• verified contradictions,
• anomaly signals.

Evidence convergence does not attempt to establish universal or metaphysical truth.

Its purpose is narrower: determine whether the independent enterprise perspectives required by the Reality Contract support the reality necessary to judge this particular consequence.

Prove the slice of reality
the consequence actually depends on.

Otarx does not need to reconstruct all enterprise state before evaluating every consequential action.

The Reality Contract bounds the reality relevant to the proposed consequence.

That boundary can include:

• actor,
• delegate,
• target object,
• operation,
• relationship,
• authority,
• purpose,
• time,
• context,
• consequence level.

Bounded Reality is the contract-scoped representation of enterprise reality established from authorized Observer Claims, evidence, contradictions, temporal conditions, and contextual constraints for evaluating a specific Action Claim.

Bounded Reality is evidence for admission. It is not itself a mutation of protected canonical state.

Judge the action claim
against established reality.

Once the required bounded reality has been established, Otarx evaluates whether the proposed consequence satisfies the applicable Reality Contract.

Admission considers supporting evidence, mandatory conditions, contradictions, anomalies, temporal validity, authority, context, and other contract-defined requirements.

The result is bounded to the specific consequence under evaluation.

Sign the consequence,
not a generic yes.

A Reality Decision is a bounded result of the admission process.

It should identify enough context for the enforcement boundary to determine that the decision applies to the exact consequence being presented.

Conceptually, a decision can bind:

• decision identifier,
• Action Claim reference,
• Reality Contract reference,
• target object,
• requested operation,
• actor or delegated authority context,
• relevant state reference,
• bounded realization scope,
• issued timestamp,
• validity window,
• integrity signature.

The purpose of this binding is to prevent a valid decision for one consequence from becoming reusable authority for a different consequence.

A Reality Decision authorizes only the bounded consequence to which that decision is cryptographically and contextually bound.

Computation does not decide
what becomes canonical.

Otarx distinguishes potential state from canonical state.

An agent may construct a proposed S1 in memory. An application may calculate a new value. A workflow may generate an approval. An attacker may compose a technically valid request.

None of those events independently make the proposed consequence canonical.

• An Action Claim is not canonical.
• A proposed S1 is not canonical.
• An Observer Claim is not canonical.
• Converged evidence is not canonical.
• Bounded Reality is not canonical.

Canonical state changes only when a valid Reality Decision authorizing the exact consequence is accepted through the protected enforcement path and the mutation succeeds.

Keep admission logic separate
from system mutation.

The Enforcement Adapter is not where Otarx determines whether the Action Claim is legitimate.

Admission occurs before the protected mutation boundary.

The adapter validates the Reality Decision and converts that bounded result into the narrow system-specific capability required to perform the approved operation.

This separation allows admission logic to remain centralized while protected systems enforce admitted consequences through interfaces appropriate to their own architecture.

The first useful integration point is usually the point where a consequential change becomes durable:

• ledger mutation,
• production deployment,
• privileged infrastructure change,
• canonical record mutation,
• identity or entitlement change,
• financial approval,
• other high-consequence state transitions.

Missing proof should not silently
become permission.

Reality admission requires explicit failure semantics.

For a protected consequence, conditions such as the following can prevent admission or enforcement:

• a mandatory observer cannot be resolved,
• required evidence is missing,
• required evidence has expired,
• a mandatory condition is not satisfied,
• a critical contradiction remains unresolved,
• a Reality Decision has expired,
• decision integrity verification fails,
• the decision does not match the requested object or operation,
• the protected enforcement path cannot validate the decision.

Exact failure behavior may vary by consequence class and deployment policy. That behavior belongs to the Reality Contract and protected consequence policy rather than arbitrary caller preference.

Assume the proposer
may be wrong or compromised.

Otarx is designed for environments where successful authentication of the caller is not sufficient evidence that a proposed consequence should become canonical.

Relevant threat conditions can include:

• a compromised autonomous agent,
• a compromised calling application,
• stolen or misused credentials,
• incorrect delegated authority,
• a hallucinated or mistaken plan,
• malicious but syntactically valid tool invocation,
• stale contextual evidence,
• one incorrect or compromised observer,
• conflicting enterprise systems of record,
• an authenticated principal acting outside legitimate enterprise conditions.

The architecture is intended to prevent control of the proposer alone from automatically becoming control of protected consequence.

Compromising the proposer is not sufficient if the proposed consequence still cannot satisfy the independent enterprise reality required by its Reality Contract.

The intended invariants.

These properties depend on correct deployment of the protected mutation boundary, integrity of Reality Contracts and evidence sources, correct cryptographic key management, correct adapter implementation, trustworthy enforcement, and appropriate failure semantics.

A $500,000 transfer.

Otarx complements
prerequisite controls.

Reality admission does not replace encryption, IAM, endpoint security, application authorization, network security, observability, transaction monitoring, or security operations.

Those systems can become prerequisites, observers, evidence providers, or enforcement dependencies.

Reason freely.
Control consequence.

Otarx does not require all upstream computation to be prevented.

An intelligent system may:

• observe permitted information,
• reason,
• generate,
• construct plans,
• call exposed claim interfaces,
• propose consequential actions.

The protected architecture governs the point where those proposals would create durable canonical consequence.

AI proposes consequences. Otarx admits consequences.

The boundary must actually
protect the mutation.

Otarx cannot provide consequence admission for a state transition that can bypass the Otarx-controlled enforcement path.

The security properties described in this paper therefore depend on:

• protected mutation paths requiring valid Reality Decisions,
• integrity of Reality Contracts,
• appropriate independence and integrity of evidence providers,
• correct identity and authority mappings,
• secure cryptographic key management,
• correct decision verification,
• correct Enforcement Adapter implementation,
• appropriate fail-safe behavior,
• prevention or control of alternate mutation paths.

Otarx should therefore be evaluated as part of a complete protected consequence architecture rather than as an isolated decision API.

Reality admission does not remove
the need for trustworthy inputs.

Independent perspectives reduce reliance on a single source but cannot create truth from uniformly corrupted inputs.

Relevant engineering questions include:

• how observer independence should be measured,
• how conflicts between authoritative systems should be represented,
• how evidence freshness should vary by consequence,
• how degraded operation should behave under partial system failure,
• how Reality Contracts should evolve and be versioned,
• how decisions should be revoked or invalidated,
• how replay resistance should be enforced,
• how high-volume admission should be distributed without weakening consistency,
• how bounded reality should be represented across heterogeneous enterprise systems.

These are implementation and systems-design problems rather than reasons to collapse proposal and consequence back into the same authority boundary.

Challenge the architecture.

Otarx is being developed as a runtime and architectural control model for consequential intelligent systems.

We welcome technical review from practitioners working in infrastructure, distributed systems, enterprise architecture, security engineering, identity, AI systems, formal methods, transaction systems, and AI safety.