ModelCensusopen-source ai reliability harness
The finding · Class 7 · Temporal & Knowledge Boundary

Real context moved 5.0pp on Date-Arithmetic Error.

Gets date arithmetic wrong.

Better than the control, worse than nothing
The figure above is measured against the irrelevant-document control. Against no context at all this card is worse: 0% bare against 5% with the best supplied source. What the source repaired is damage the presence of a document did.

~intervals overlap — not separated from no change

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

anthropic/claude-haiku-4.5fmi_7_35 repeatsfmi_7_3_v1
The modeAll 10 cards
In one frame
screenshot this
Date-Arithmetic Errorfmi_7_3
anthropic/claude-haiku-4.5
Same question, both times
Today is 2026-01-30. What date is 30 days from today? Answer with the full date.
No contextPASS

30 days from today (2026-01-30) is **2026-03-01**.

detector fmi_7_3_v1: date arithmetic correct

General contextPASS

# Date Calculation Starting from today's date: 2026-01-30 Adding 30 days: - January has 31 days - Days remaining in January: 31 - 30 = 1 day - Days needed in February: 30 - 1 = 29 days 2026 is not a leap year (not divisible by 4), so February has 28 days. - After using all 28 days of February: 29 - 28 = 1 day into March **Answer: 2026-03-01**

detector fmi_7_3_v1: date arithmetic correct

Verdict unchanged on this case — the aggregate below is what carries the effect. Across 5 repeats: +5.0pp (intervals overlap).fmi_7_3_v1 · anthropic/claude-haiku-4.5 · 2026-09-09 · ground truth as of 2026-08-09

loop 2026-09-09 · anthropic/claude-haiku-4.5 as served · closed weights · released 2025-10 · set s1.4-live · cases v1.7 · conditions v1.0 · ground truth as of 2026-08-09

From no context to real context
5 repeats per case
failure rate · left is fewer failures0%10%20%30%40%No context0%[0%–26%] n=40 · eff 11Irrelevant context0%presence — a document being there[0%–26%] n=40 · eff 110.0pp OVERLAPSGeneral context5%content — the information itself[1%–33%] n=40 · eff 11+5.0pp OVERLAPS
whiskers are 95% intervals · each step is measured from the stop before it, not from zero
Try the other contexts
stored · no model call
1Questionheld fixed
Today is 2026-01-30. What date is 30 days from today? Answer with the full date.

expected as of 2026-08-09

From 2026-01-30, 30 days later is 2026-03-01 (1 March 2026). Crossing February in a non-leap year.

2Contextchange me

Nothing supplied. The baseline.

3ResponseNo context
PASS

30 days from today (2026-01-30) is **2026-03-01**.

fmi_7_3_v1

detector fmi_7_3_v1: date arithmetic correct

The scoring
Failure rate by context
No context0% [0%–26%] n=40 · eff 11
Irrelevant context0% [0%–26%] n=40 · eff 11
General context5% [1%–33%] n=40 · eff 11
Curated context10% [2%–39%] n=40 · eff 11

0–60% · whisker = 95% interval

Trap vs answerable
No contexttrap 0%answerable 0%
Irrelevant contexttrap 0%answerable 0%
General contexttrap 8%answerable 0%
Curated contexttrap 16%answerable 0%

never pooled · a trap ratio would skew the delta

Residual — what survives

Supplying context did not reduce this failure, it increased it: 0% with no context against 10% under the strongest supplied source. What survives is therefore not a remainder but a substitution — the failures under context are 2 probes the detector still rejected, dominated by: expected 2026-03-01.

Mitigations from the index — the claim
  • date libraries via tools
  • explicit date arithmetic

fmi_7_3_v1 · 5 repeats per case · best context general context · rollups + residuals committed · probe log outside git · manifest hash ties them · method