ModelCensusopen-source ai reliability harness
Failure Mode Index · fmi_6_1

Tool-Routing Error.

Tool Use & Agentic Controlinstrumented · tier 2 · status: stable

Given multiple tools, the model routes a request to the wrong tool despite adequate labels/descriptions.

The question that exposes it

Given weather, currency, and translate tools — convert 100 USD to EUR.

Probes: Whether it routes to the right tool among distractors.

Trap: The model reaches for a plausible-but-wrong tool despite clear labels.

Signature

Tasks with a known correct tool among distractors; failure = a call to a tool whose description does not fit the task.

Example

Uses a `weather` tool to answer a currency-conversion request.

Severity axes
prevalence
medium
harm
medium
stealth
low
Engineering guidance · class 6
  • Validate every tool call against its schema before executing it.
  • Cap iterations and detect repeated identical calls. Non-termination is a failure mode, not a hang.
  • Diff the final answer against the tool return rather than trusting the summary of it.
  • Replay the call log to compute true state. Never ask the model for the running total.
Mitigations
  • better tool descriptions
  • routing verification
Measurement

No published loop has measured this mode yet. When one does, its cards appear here, newest first.