Governed boundary
Load-bearing foundations
The governed object is the set of assumptions and dependency relations whose validity materially affects whether a consequential result or exercise of authority remains eligible.
Patent-aligned definition · Core Enforcement Substrate
SafeAssumption
Critical Assumption Integrity · Consequence-Gated Assumption Control
When correct reasoning produces catastrophic results. An advanced AI can reason coherently, follow its assigned goal, and still cause serious harm—because one crucial assumption beneath its reasoning was wrong. SafeAssumption provides a dynamic fail-safe for advanced systems by governing the assumptions whose validity is necessary to support a consequential conclusion, recommendation, authorization, pathway, or action.
No verified critical assumption, no authority to proceed.
35th established component
26th Core Enforcement Substrate
Evidence proportionate to consequence
Authority follows foundation state
Control mechanism
Bind evidence to consequence
Each critical assumption receives an evidence contract, a foundation state, a dependency map, and revalidation conditions proportionate to possible harm or irreversibility.
Enforcement output
Dependent authority contracts
If a foundation becomes stale, unsupported, expired, contradicted, or invalid, dependent eligibility and authority are reduced, suspended, withdrawn, or rerouted.
I. Canonical definition
What boundary Critical Assumption Integrity governs
Critical Assumption Integrity governs the continuing sufficiency of assumptions that materially support consequential artificial-intelligence reasoning, conclusions, recommendations, decisions, authorizations, pathway selections, or actions.
It addresses the structural failure in which an AI system silently adopts or inherits an unverified foundation, builds a coherent chain of reasoning upon it, and converts that chain into a consequential output or exercise of authority. The defect may persist even when every later reasoning step is internally consistent.
The control identifies critical assumptions, determines which are load-bearing, assigns consequence-proportionate evidence requirements, records dependent conclusions and permissions, establishes a current foundation state, and continuously revalidates that state. Eligibility to proceed is therefore conditional on the continuing sufficiency of the foundations on which it depends.
Canonical distinction: SafeAssumption is SafeWave’s 35th established architectural component and 26th Core Enforcement Substrate. It provides Critical Assumption Integrity as a distinct enforceable control while working through verified boundaries with other SafeWave components.
II. Canonical mapping
Risk, governed object, trigger conditions, mechanism, and output
Risk or instability surface
A consequential AI result may rest on an assumption that is unverified, weakly supported, stale, contextually inapplicable, contradicted, or no longer true while remaining hidden beneath apparently coherent reasoning.
Governed object
Critical assumptions, their evidence contracts and foundation states, the dependency relations that make them load-bearing, and the eligibility or authority of every dependent output or action.
Trigger conditions
Formation or use of a critical assumption; a proposed consequential output or action; evidence expiry or staleness; material context change; contrary evidence; dependency change; or any event specified by the assumption's revalidation contract.
Control mechanism and output
Consequence-gated evidence contracts, counterfactual load-bearing tests, dependency tracking, scoped eligibility, event-driven revalidation, and automatic reduction or withdrawal of dependent authority until verified restoration.
III. Critical and load-bearing tests
Not every assumption carries the same control burden
An assumption is critical when its validity could materially affect a consequential result, recommendation, decision, authorization, pathway, or action. Criticality is determined by dependency and possible consequence—not by whether the assumption appears prominent in the system's explanation.
A critical assumption becomes load-bearing when a material counterfactual change in that assumption—such as treating it as false, unresolved, expired, or only conditionally sufficient—would change the eligibility, scope, safety, or authority of a dependent output or action.
A load-bearing foundation is identified through demonstrated dependency. Importance, confidence, repetition, or persuasive phrasing alone does not establish the test.
IV. Evidence contracts
Required support rises with possible consequence
- Evidence requirements are specified for the particular assumption, context, dependency, and consequence class.
- Higher-consequence, less reversible, or more expansive authority requires stronger, more independent, and more current support.
- The contract may specify source classes, corroboration, independence, freshness, scope, contradiction handling, and revalidation events.
- Absence of contradiction is not automatically evidence of sufficiency.
- A confidence score cannot substitute for fulfillment of the required evidence contract.
Critical Assumption Integrity invariant
No consequential eligibility or authority may remain stronger than the verified foundation on which it depends.
V. Foundation states and revalidation
Support is a continuing state, not a one-time check
Each load-bearing assumption receives a foundation state describing whether it is sufficiently supported for a defined use and consequence. Relevant states may include supported, conditionally sufficient, unresolved, expired, or contradicted. The state is scoped: evidence sufficient for one low-consequence use may be insufficient for a broader or more consequential use.
- Time or evidence expiry can require revalidation without proving that the assumption is false.
- Material changes in environment, scope, objective, model, tool, data, authority, or dependency can invalidate earlier sufficiency.
- Contradictory evidence triggers state reassessment and dependent control responses.
- Revalidation occurs before consequential use and whenever a defined event makes prior support potentially unreliable.
- Failure to complete required revalidation prevents silent continuation under the old state.
VI. Dependency invalidation and authority response
Failure propagates to what the assumption was supporting
A failed foundation is not handled as an isolated metadata warning. The system identifies every conclusion, recommendation, authorization, pathway, action, or delegated permission that materially depends on it and changes their eligibility accordingly.
Reduce
Narrow the permitted scope, duration, consequence, autonomy, or execution pathway to a level supported by the remaining evidence.
Suspend
Temporarily prevent dependent use or execution while required evidence, review, or revalidation remains incomplete.
Withdraw
Revoke dependent eligibility or authority when the assumption is contradicted, invalid, or insufficient for the proposed consequence.
Reroute
Select a lower-consequence pathway, request missing evidence, expose the unresolved foundation, or route to independently authorized review.
Recompute
Re-evaluate dependent reasoning, conclusions, plans, and permissions under the changed foundation state rather than preserving the prior output.
Record
Preserve the assumption, evidence state, dependency path, trigger, response, and resulting authority change for later verification and audit.
VII. Restoration and reauthorization
Authority does not return merely because new evidence appears
Restoration requires the evidence contract to be satisfied again for the relevant context and consequence. The system must update the foundation state, recompute materially dependent results, verify that the original dependency path remains applicable, and preserve evidence of the restoration process.
Technical restoration
Fresh or newly sufficient evidence supports a revised foundation state; dependent reasoning and eligibility are recomputed against that state.
Authority restoration
Any permission or authority previously reduced, suspended, or withdrawn is restored only through the required reauthorization path and only within verified scope.
Restoration of an assumption's support does not automatically validate every earlier conclusion or reactivate every earlier permission. Dependency-sensitive recomputation and reauthorization remain required.
VIII. Auditability and verification
The foundation and its consequences must remain inspectable
The control produces evidence sufficient to determine what assumption was used, why it was classified as critical or load-bearing, what evidence contract applied, what evidence was evaluated, which foundation state was assigned, and which outputs or permissions depended on it.
Audit records also preserve revalidation events, contradictions, expiry or staleness, eligibility changes, enforcement responses, recomputation, restoration, and reauthorization. The record must distinguish reported evidence, system inference, unresolved uncertainty, and externally supplied authorization.
Auditability does not replace enforcement. A system that records a failed assumption while allowing dependent authority to continue unchanged has not satisfied Critical Assumption Integrity.
IX. What Critical Assumption Integrity is not
A foundation-control method—not a general confidence label
- It is not ordinary hallucination detection or a general claim that an output may be wrong.
- It is not confidence scoring, calibration, explainability, provenance, monitoring, or human review by itself.
- It does not require every incidental premise to receive the same evidence burden.
- It does not treat staleness as automatic falsity; staleness changes the support state and triggers required revalidation.
- It does not permit polished reasoning, model consensus, or an absence of alerts to substitute for required evidence.
The distinguishing function is the complete control chain from assumption discovery and consequence-proportionate evidence through dependency, scoped eligibility, revalidation, and enforceable authority response.
X. Relationship to the SafeWave architecture
A distinct component working through verified architectural boundaries
SafeAssumption owns the distinct governed object of load-bearing assumptions, their evidence contracts and foundation states, their dependency relations, and the eligibility or authority that depends upon them. Its findings may be consumed or enforced by other specialized components, but those relationships must be assigned through the SafeWave Canonical Mapping Rule rather than inferred from component names.
SafeProvenance
May establish the origin and history of evidence. Provenance alone does not determine whether an assumption is sufficiently supported for a particular consequence.
SafeMemory
May govern persistent cognitive state. Critical Assumption Integrity governs whether a stored or newly formed assumption remains sufficient for dependent use.
SafeProcess
May govern formation of autonomous capability through intermediate reasoning. It does not replace validation of the load-bearing foundations supporting that reasoning.
SafeTelemetry
May observe state and signal changes. Critical Assumption Integrity additionally requires dependency-aware eligibility and enforcement responses.
SafePathway, SafeRuntime, and SafeControl
May select or enforce constrained pathways, continuation conditions, and control responses produced by the assumption-integrity process.
SafeAdmission boundary
The method's concept of positive admission or eligibility must not be confused with SafeAdmission, which canonically governs node participation and re-entry under instability.
Assess Critical Assumption Integrity directly
SafeAssumption can now be used as a specific architectural mapping in SafeWave assessment logic, assessment reports, and engineering documents. Its relationships with other named components must still be verified against each component’s canonical governed boundary.
Use the assessment without identifying yourself. Responses remain in your browser unless you deliberately submit them. No name, employer, organization, system name, or personal or work email is required. You may complete and save the questionnaire without sending anything to SafeWave; the optional free report may be delivered through a separate, non-identifying return address that will not be used for promotion.
SafeAssumption is SafeWave’s 35th established component and 26th Core Enforcement Substrate. It provides Critical Assumption Integrity while working across verified boundaries with other components in SafeWave’s evolving preventive AI systems architecture.