Issue
Similar Cases Route to Different Outcomes
Similar inputs, cases, prompts, or workflow states are routed to different outcomes without a declared difference that explains the split.
What This Looks Like
Two cases appear equivalent to the user, but the AI system routes them to different answers, decisions, tools, workflows, reviewers, or next steps. The user may see one request approved and another rejected, one case escalated and another handled normally, or one prompt sent down a different path without any visible reason for the difference.
Why It Matters
Routing differences become hard to trust when the deciding distinction is not visible. If similar cases produce different outcomes, users cannot tell whether the system found a meaningful difference, followed hidden state, used inconsistent context, or simply varied across runs. That weakens review, debugging, policy enforcement, and user confidence.
Structural Signal
The issue is not merely that two outputs are different. The signal is that the system treats comparable cases as structurally different without exposing the condition, rule, state, or authority that caused the split.
Common Triggers
- Similar prompts carry small hidden differences in context, memory, metadata, or prior state
- Routing depends on thresholds, classifiers, or confidence signals the user cannot see
- A workflow has multiple valid paths but no declared tie-breaker
- The system treats equivalent labels, fields, or case types as different
- Different tools, policies, or model paths are selected without a visible routing explanation
- A prior answer or intermediate step changes the route for a later case
When to Use This Issue
Use this Issue when the central problem is unexplained routing or outcome divergence between cases that should be handled the same way, or whose differences should be explicitly declared before they affect the result.
When Not to Use This Issue
Do not use this Issue when the cases are clearly different, when the user is mainly objecting to the quality of one answer, or when the failure is only that repeated runs vary. If the same exact prompt changes after retry, use a retry or inconsistent-output Issue instead.
Category
Primary Pattern
Declared Patterns
Non-Deterministic Execution
A structural condition where equivalent inputs and declared constraints produce divergent outputs across executions.
Divergent Outputs
A structural condition where parallel evaluations under comparable scope and shared authority produce non-equivalent outputs.
Authority-State Mismatch
A structural condition where observed system state no longer aligns with the declared authority state that is supposed to govern it.
Derived Primary Lenses
Convergence Lens
Compares parallel structural systems to determine whether they align under shared authority.
Determinism Lens
Evaluates whether identical structural inputs produce equivalent structural outputs across repeated executions.
Reconciliation Lens
Evaluates whether structural changes align with declared authority updates, version changes, or reconciliation rules.
Variance / Entropy Lens
Measures structural variability across repeated or comparable evaluations and identifies divergence beyond expected bounds.
Derived Secondary Lenses
Related AI-Adjacent Issues
Search Intents
- similar cases route to different outcomes
- same case gets different result
- AI routes similar requests differently
- similar prompts produce different decisions
- same workflow state gets different outcome
- AI decision path is inconsistent