Workbench / Issues

Same Instructions Allow Different Outputs

The same instructions are broad or underspecified enough to allow materially different outputs that all appear compliant.

What This Looks Like

The same instructions produce materially different outputs, and each output can still be defended as following the prompt. The differences may involve structure, recommendation, level of detail, source choice, field selection, wording, or decision path. The problem is not obvious noncompliance; it is too much compliant variation.

Why It Matters

Instructions that allow many different compliant outputs are hard to validate. Users may not know whether the AI did something wrong or whether the task was too broad to control the result. This can make repeated runs, review comparisons, and workflow automation unreliable.

Structural Signal

The instructions define a wide enough solution space that different outputs can all satisfy the stated request. The issue is not random variation alone; it is that the governing instructions fail to narrow the output to the needed equivalence class.

Common Triggers

When to Use This Issue

Use this Issue when the same instructions allow materially different outputs that all appear compliant, making the result unstable or difficult to evaluate.

When Not to Use This Issue

Do not use this Issue for harmless wording variation. Do not use it when the outputs differ because the input, context, source set, model, or workflow changed.

Category

Inconsistent Behavior

Primary Pattern

PAT-0290 — Divergent Outputs

Declared Patterns

Derived Primary Lenses

Derived Secondary Lenses

Search Intents

Ontology Metadata

Code
ISS-0072
Version
ISS-0072@0.1.0
Ontology release
0.1.0
Updated
2026-05-10T00:00:00Z

History

No public history entries recorded.

View full ontology changelog