ModelCensusopen-source ai reliability harness
Failure Mode Index · fmi_3_4

Needle-Position Sensitivity.

Instruction Adherence & Long-Contextinstrumented · tier 2 · status: stable

Retrieval of a specific fact from long context depends on where in the context the fact sits ("lost in the middle").

The question that exposes it

Somewhere in this 60-page document is a vault code. What is it?

Probes: Whether recall depends on where the needle sits (lost-in-the-middle).

Trap: Facts at the start and end are recalled; the middle vanishes.

Signature

The same needle swept across positions (start/25%/50%/75%/end) of a long filler context; failure = recall that drops at particular positions.

Example

Recalls a planted code at the top and bottom but not the middle of a 60k-token context.

Severity axes
prevalence
high
harm
medium
stealth
medium
Engineering guidance · class 3
  • Restate hard constraints in the last message, not only in the system prompt.
  • Validate structured output programmatically on every turn. Sampling turns is how schema drift ships.
  • Carry a canary constraint through long sessions and alert when it decays.
  • Compact before you extend. A shorter conversation holds a rule better than a longer window does.
Mitigations
  • position-robust attention
  • retrieval over raw stuffing
Measurement

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