ModelCensusopen-source ai reliability harness
Failure Mode Index · fmi_4_1

Method/Execution Split.

Reasoning & Calculationinstrumented · tier 1 · status: stable

The model chooses a correct method but executes it wrongly (or vice versa) — a right-approach/wrong-answer or wrong-approach/right-answer split.

The question that exposes it

The car wash is a 5-minute walk away. Do you drive the car there, or walk?

Probes: Whether commonsense survives a prompt engineered to trigger the wrong pattern.

Trap: '5-minute walk' primes 'just walk' — models forget the whole point is to wash the car, which has to come.

Signature

On math-checkable cases, compare the stated method and the final numeric answer against an oracle; failure = correct method with wrong computed result, or an answer not entailed by the shown work.

Example

Sets up the right integral, then arithmetic-errors the evaluation.

Severity axes
prevalence
high
harm
medium
stealth
medium
Engineering guidance · class 4
  • Do arithmetic in code. Hand the model a calculator and ground the answer in what it returns.
  • Assert units and dimensions separately from the value.
  • Ask for method and answer as separate fields, then check they agree.
  • Do not ask a model to verify its own output in the same turn — it confirms far more readily than it catches.
Mitigations
  • tool-assisted computation
  • verify final against method
Measurement

One tab per model, newest first. Each card is that model against this mode under every condition on a single day. The mitigations above are the claim; the residual on each card is what actually survived the intervention.

fmi_4_1_v1 · 5 repeats · served 2026-09-10

loop 2026-09-10 · x-ai/grok-4.6 as served · closed weights · released 2026-01 · set s1.4-live · cases v1.7 · conditions v1.0 · ground truth as of 2026-08-09

ConditionFailure rate95% intervalnTrapAnswerable
No context
bare · The model alone, with no supplied context.
0%0%–10%n=50 · eff 350% (n=30)0% (n=20)
Irrelevant context
null · Length-matched irrelevant context. The control.
0%0%–10%n=50 · eff 350% (n=30)0% (n=20)
General context
recipe_a · Broad retrieval — the first passage the source returns, unreviewed.
0%0%–10%n=50 · eff 350% (n=30)0% (n=20)
Curated context
recipe_b · Hand-checked retrieval — the same passage plus the provenance chain behind it.
0%0%–10%n=50 · eff 350% (n=30)0% (n=20)
Presence effect · no context → irrelevant context
0.0pp

intervals overlap — not separated

What a document being there did, before any of its content. Should be small.

Content effect · irrelevant → best real context
0.0pp

intervals overlap — not separated

What the information itself did. This is the publishable claim.

Residual — what survives

Unchanged by context: 0% with none, 0% with the strongest supplied source. For a control mode that flat line is the result — it is what licenses reading movement elsewhere in this loop as grounding rather than as a document being present. Nothing survives at the best condition on the cases measured here.

Mitigations from the index — the claim
  • tool-assisted computation
  • verify final against method

replay this card →