ModelCensusopen-source ai reliability harness
Failure Mode Index · fmi_7_3

Date-Arithmetic Error.

Temporal & Knowledge Boundaryinstrumented · tier 2 · status: stable

Given an injected "current date," the model computes relative dates/intervals incorrectly.

The question that exposes it

Today is 2026-03-01. What calendar date was exactly 90 days ago?

Probes: Whether relative-date math against an injected 'today' is correct.

Trap: Off-by-days and wrong-month errors are common in date arithmetic.

Signature

Inject a current date in-context; ask relative-date questions; compare against a date oracle. Failure = wrong interval/target date.

Example

Told "today is 2026-03-01," says "90 days ago was January 1."

Severity axes
prevalence
medium
harm
medium
stealth
medium
Engineering guidance · class 7
  • Inject the current date explicitly in context. Never assume the model knows what day it is.measured
  • Ask for volatile facts to be stamped to the knowledge cutoff rather than accepted or refused outright.
  • Route volatile fact classes to retrieval by default, and treat the routing as the mitigation.
  • Record each model's published training cutoff in your own config. Staleness behaviour is uninterpretable without one.
Mitigations
  • date libraries via tools
  • explicit date arithmetic
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_7_3_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%–26%n=40 · eff 110% (n=25)0% (n=15)
Irrelevant context
null · Length-matched irrelevant context. The control.
0%0%–26%n=40 · eff 110% (n=25)0% (n=15)
General context
recipe_a · Broad retrieval — the first passage the source returns, unreviewed.
0%0%–26%n=40 · eff 110% (n=25)0% (n=15)
Curated context
recipe_b · Hand-checked retrieval — the same passage plus the provenance chain behind it.
0%0%–26%n=40 · eff 110% (n=25)0% (n=15)
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
  • date libraries via tools
  • explicit date arithmetic

replay this card →