What This Looks Like
The documentation says a model, tool, connector, API, mode, field, permission, or product surface should behave one way, but the actual runtime behaves differently. A feature may be missing, a field may differ, a tool may reject documented inputs, or the runtime may produce behavior the docs do not describe.
Why Users Blame AI
The mismatch often appears inside an AI task, so the user may experience it as the AI ignoring instructions, making unsupported calls, or inventing behavior. The underlying problem may be documentation drift, rollout differences, feature flags, product-version mismatch, hidden limitations, or runtime behavior that changed before the docs did.
What to Check First
- Whether the docs match the exact version, product, API, mode, and account being used
- Whether the behavior depends on plan, workspace, region, feature flag, or rollout state
- Whether the runtime exposes a different schema than the docs show
- Whether examples are illustrative instead of contract-level
- Whether the docs describe expected behavior but not current product behavior
- Whether a hidden policy, permission, or tool constraint changes runtime behavior
When This Is AI-Adjacent
Use this AI-Adjacent Issue when documented behavior and observed runtime behavior disagree. If the mismatch creates a structural Workbench failure, use the related Issues for policy drift, output change, prompt behavior change, contract-name drift, or tool call contract mismatch.