ModelCensusopen-source ai reliability harness
Failure Mode Index · fmi_1_4

Planted-History Recall Failure.

Grounding & Attributioninstrumented · tier 2 · status: stable

The model misremembers, contradicts, or fabricates details of earlier turns that were explicitly established in the conversation.

The question that exposes it

[turn 1] My budget is exactly $4,000. … [turn 6] So what can I afford?

Probes: Whether a fact planted early survives to a later turn intact.

Trap: Under depth the model misremembers $4,000 as $14,000 and reasons from the wrong figure.

Signature

In a multi-turn protocol that plants specific facts early, a later answer conflicts with the planted history.

Example

User states "my budget is $4,000" in turn 1; by turn 6 the model reasons from "$14,000."

Severity axes
prevalence
medium
harm
high
stealth
medium
Engineering guidance · class 1
  • Supply the source rather than asking the model to recall one — a real passage substantially reduces post-cutoff fabrication across the panel.measured
  • Resolve every citation yourself. An HTTP or DOI check in your pipeline, never the model's word that a reference exists.
  • Diff what the model says it did against your tool log. Narration is not evidence of an action.
  • Make "no basis found" a passing answer in your own eval, or you are training over-assertion.
Mitigations
  • conversation-state grounding
  • explicit recall checks
Measurement

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