State precedes answer
The failure occurs before behavior. A valid decision state either exists or it does not. Current systems have no mechanism for detecting when it does not. The question is when a system is even in a state where a correct answer can exist.
It is not that the answer is incorrect. The system is producing an answer in a state where correctness is impossible.
A distinct category
The RSS is distinct from alignment methods, training approaches, and evaluation layers. It does not attempt to improve model behavior. It defines whether a response is structurally admissible before behavior occurs.
Current frameworks shape system behavior toward correct outputs. The Realis Structural Standard establishes whether a load-bearing basis for any answer exists.
This distinction operates at the dependency-edge layer specified in DX-DE-001. AI systems are a cross-domain source of the correctness-edge failure mode: confident, well-formed output pointed at the wrong object, passing inspection by construction. Unlike validity edges, which remove the object of evaluation entirely and announce themselves as absence, correctness edges corrupt the object while leaving it in place. A system producing a correctness-edge failure looks coherent from inside. The precondition check specified in RSS-001-N1 converts this concealing failure mode into an honest validity edge by introducing a non-bypassable refusal condition when the decision basis is absent or indeterminate. The refusal announces the problem rather than concealing it. The structural distinction between these edge types is developed in DX-DE-001 (Dependency Edge Type).
The structural principle behind this is developed in WSA-002, Why Orientation, Not Prediction. Prediction extrapolates forward from historical patterns; orientation establishes position against conditions that apply regardless. The precondition check the standard requires is the orientation operation expressed at the AI decision boundary.
The five-second test
If a model is asked a question where the evidence is incomplete, ambiguous, or contradictory, what mechanism currently forces it to refuse rather than complete the answer?
The four operational categories that produce an unresolved decision basis are contradiction, vacuity, incompleteness, and ambiguity, specified in RSS-001-N1.
Most systems have no such mechanism. The architecture is oriented toward completion. Refusal is treated as error. The system proceeds fluently and confidently on a foundation that has not been established.
The question is not how to improve the system's output. It is whether output should be permitted at all.
A structural failure the architecture cannot correct from within
The failure occurs at the precondition layer. Resolving it through further output extends the completion logic. The instrumentation must be decoupled from the generation process itself.
A system optimized for completion cannot reliably refuse from inside its own architecture. The same processes produce confident outputs in invalid states as well. The signal that would distinguish valid from invalid states is the signal the architecture is not built to generate.
Alignment is the wrong category for this failure. Alignment shapes what the system produces. It does not introduce a non-bypassable refusal condition for cases where no valid output is structurally possible. That condition has to come from outside the generation process.
External does not refer to physical separation. It refers to constraint independence.
A precondition check meets this condition only if it is not subject to the optimization that rewards completion, and cannot be overridden or reinterpreted by the process producing outputs.
Subsystems trained on the same objective, or whose refusals can be bypassed by the generator, do not qualify.
Completion is not decision.
What the Realis Structural Standard changes
The RSS introduces a structural floor that operates before generation, not after.
- A precondition check that determines whether a load-bearing decision basis for a valid decision state exists
- Mandatory refusal when that basis is absent or indeterminate, treated as correct operation, not error
- A durable trace connecting every output to a verifiable decision chain
- Separation of fluency from validity. Fluency and validity are entangled, and that entanglement is the failure
Operational specification of constraint independence, deficit categories, and falsifiability appears in RSS-001-N1.
Authority does not override structure
The standard also defines what happens when a user, operator, or upstream system supplies input intended to resolve a detected deficit. Such input is treated as a candidate addition to the decision basis, not as authorization to proceed. The precondition check verifies the supplying party's standing, verifies the source independently, admits the input only if both are satisfied, and re-evaluates the basis for all four deficit categories. Refusal continues if any deficit remains.
The structural floor is not subject to authority override.
Operational specification appears in RSS-001-N2. A worked demonstration of the verification sequence under operator pressure appears in RSS-001-N2-WE1.
The downstream consequence
Without this, invalid outputs enter decision systems that assume validity, converting a local reasoning failure into a system-level decision error.
As model capability increases, the cost of undetected invalid outputs rises faster than the system's ability to internally detect them.
If this describes a failure mode you have observed
The standard defines a concrete mechanism that addresses it. It is publicly available, self-certified, and invokable without authorization from Realis Institute.