Die Telemetry funktioniert. Der digitale Schlüssel funktioniert ebenfalls. Aber die Eigentumsquelle ist veraltet. SRA lässt den alten Zustand nicht einfach weiterlaufen.
STALE → CLOSURE NOT REUSABLE
STATE-REALITY ASSURANCE · SRA
THE BOUNDARY BEFORE AUTHORITY
Just because the system sees something does not mean it may act.
Modern systems see an enormous amount: sensors report states, APIs return status, vehicles stream telemetry, and AI agents interpret events. The problem begins when an observation silently turns into authority. That is exactly where State-Reality Assurance starts.
SRA does not decide whether an effect is authorized. It decides whether what runtime claims about the world is sufficiently assured to be presented to the Authority layer at all.
Reality may influence the question. It may never write the answer.
20SRA INVARIANTS
46/46MODELED EXPECTED OUTCOMES
0BOUNDARY BYPASS OBSERVED
falseexecution_permitted
THE PROBLEM
Observation is not yet truth. Truth is not yet Authority.
A sensor may measure correctly. An API may be reachable. An agent may describe a situation perfectly. None of that should automatically create real-world effect. Observation answers only: “What appears to be the case?” Authority answers a different question: “What is allowed to follow from it?”
01 · SENSOR
“The door is open.”That is a state. It is not authority to lock the door, trigger an alarm, or exclude someone.
02 · API
“The account is active.”That says nothing about whether the account may create this specific effect now, on this resource, for this purpose.
03 · AI AGENT
“I have enough information.”Enough information is not execution authority. Knowledge and Authority remain separate.
THE VISUAL
Runtime ends here.
SRA’s strongest property is not a score. It is the boundary. On the left, the world may be observed. On the right, Authority begins. Between them sits a controlled assurance gate.
RUNTIME · THE WORLD AS INPUT
Everything may be observed.
Signals may arrive, conflict, become stale, or disappear. They remain observations.
Sensors
APIs
Telemetry
AI Agents
Cameras
GPS
Cloud Systems
External Sources
OBSERVATION ≠ AUTHORITY
RUNTIME INPUT ∉ EXECUTION AUTHORITY
RUNTIME INPUT ∉ EXECUTION AUTHORITY
CONTROLLED BOUNDARY
SRACHECKASSURE
BIND CONTEXT
FORCE REASSESSMENT
SRA is not another Authority family. It is the universal pre-boundary that determines which observed state may enter Authority evaluation at all.
WHY THIS MATTERS
The most dangerous errors often look plausible.
SRA is built for situations where a system appears to have “enough data,” while that data is not sufficiently assured for the specific effect.
VEHICLEA vehicle reports: “Owner unchanged.”
AI AGENTAn agent reports a state with high confidence.
High model confidence is not provenance. The model may describe, plan, and request. It may not promote its own output into Authority.
MODEL OUTPUT → ASSURANCE, NEVER AUTHORITY
SOURCES20 APIs confirm the same value.
If all 20 depend on the same compromised upstream, they are not 20 independent sources epistemically. SRA detects shared provenance chains.
SOURCE COUNT ≠ INDEPENDENT EVIDENCE COUNT
REPLAYA previously valid artifact is presented again.
The artifact may be cryptographically genuine and still be insufficient for the new context. State, policy, dependency, or time changes force reassessment.
VALID ARTIFACT ≠ CURRENT SUFFICIENCY
FIVE PLANES
From observation to effect without mixing the layers.
01 · REALITYWhat is happening?
Sensors, APIs, Telemetry und externe Ereignisse liefern Rohbeobachtungen.
02 · ASSURANCEWhat do we know reliably?
SRA checks provenance, freshness, observability, consistency, dependencies, and uncertainty.
03 · AUTHORITYWhat may follow from it?
Only now are rights, mandates, rules, scope, purpose, and jurisdiction evaluated.
04 · EXECUTIONWhich effect may occur?
Only a closed, current Execution Right can carry an executable capability.
05 · EVIDENCEWhy was it allowed or blocked?
The receipt binds decision, state, context, and result for later verification.
WHAT SRA CHECKS
Not “How much data do we have?” but “How reliable is this state?”
1
ProvenanceWhere did the observation come from, and can its provenance be traced?
2
FreshnessIs the state still current enough for this exact domain?
3
ObservabilityDo we actually observe the relevant state or only a fragment?
4
ConsistencyDo sources conflict in a way that matters for the effect?
5
IndependenceAre multiple sources truly independent or merely copies of the same upstream?
6
UncertaintyIst die verbleibende Uncertainty klein genug, um Authority-Evaluation überhaupt zuzulassen?
OVERALL ARCHITECTURE
SRA protects the entrance. Continuing Authority protects the moment of effect.
This creates a symmetric architecture: before Authority, the system checks what may enter as sufficiently assured state. At the Point of Effect, it checks again whether the right still exists.
SRAWhat may enter Authority?
→
AUTHORITY CLOSUREWhat right exists?
→
CONTINUING AUTHORITYDoes that right still exist?
→
POINT OF EFFECTMay the effect occur now?
→
EVIDENCEWhy was it allowed or blocked?
AEG says: “The model may think. Authority stays outside the model.”
SRA adds: “The runtime may observe. Authority stays outside the runtime.”
SRA adds: “The runtime may observe. Authority stays outside the runtime.”
CONTEXT & REPLAY
A genuine artifact can still be the wrong artifact.
SRA binds assurance to the exact decision context. Resource, effect class, domain, tenant, jurisdiction, policy version, snapshot version, validity window, and provenance roots belong together. If the context does not match, the artifact is insufficient.
SRA-INV-019
Assurance Context BindingAn artifact for vehicle A cannot be used for vehicle B, an OTA artifact cannot be reused for ownership, and a foreign tenant context cannot be used for the current decision.
SRA-INV-020
No Assurance ReplayState change, policy change, material dependency change, or expiry make a previous artifact insufficient for a new Authority Closure.
ROOT LAUNDERING
Many sources are not automatically many pieces of evidence.Shared upstream elements are considered so apparently independent sources cannot hide the same compromised origin.
INTERNAL VERIFICATION
The boundary is not only described. It is attacked.
The internal adversarial SRA suite covers Threshold Downgrade, Policy/Artifact Mismatch, Artifact Replay, Cross-Domain Reuse, Root Laundering, Time Manipulation, Partial Observability, Artifact Substitution, Assurance-after-Authority, and Reassessment Suppression.
46/46MODELED EXPECTED OUTCOMES
20FORMALISIERTE SRA INVARIANTS
0OBSERVED BOUNDARY BYPASSES
1RUNTIME-DERIVED OBJECT BEYOND THE BOUNDARY
Claim Boundary: These figures describe the current internal, modeled, executed SRA test state. They are not external certification, regulatory approval, or a claim of universal security.
FOR TECHNICAL REVIEWERS
The technical depth remains available. It simply no longer gets in the way.
Reviewers who want to go deeper find the formal logic here — after the explanation, not before it.
E_base = P^wP · F^wF · O^wO · C^wC · D^wD · A^wA · N^wN
E = E_base · (1-U)^wU
ESTABLISHED → SUFFICIENT → DEGRADED → CONFLICTING → STALE → INSUFFICIENT → UNKNOWN
AUTHORITY_EVALUATION_PERMITTED ⇏ EXECUTION_ALLOWED
execution_permitted = false
E = E_base · (1-U)^wU
ESTABLISHED → SUFFICIENT → DEGRADED → CONFLICTING → STALE → INSUFFICIENT → UNKNOWN
AUTHORITY_EVALUATION_PERMITTED ⇏ EXECUTION_ALLOWED
execution_permitted = false
FAQ
The three questions SRA should answer immediately.
Is SRA another Authority family?
No. SRA sits before Authority. It is the universal pre-boundary that determines whether observed state is sufficiently assured to enter Authority evaluation at all.
Can a high SRA score permit Execution?
No. SRA never authorizes Execution. execution_permitted remains constant false. ESTABLISHED or SUFFICIENT only permit entry into Authority evaluation.
Why are multiple sources not enough?
Because multiple sources can share the same origin. SRA considers source independence and shared provenance chains so 20 dependent APIs do not look like 20 independent pieces of evidence.