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

Real context moved -2.5pp on Date-Arithmetic Error.

Gets date arithmetic wrong.

~intervals overlap — not separated from no change

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

openai/gpt-4ofmi_7_35 repeatsfmi_7_3_v1
The modeAll 10 cards
In one frame
screenshot this
Date-Arithmetic Errorfmi_7_3
openai/gpt-4o
Same question, both times
Today is 2026-01-30. What date is 30 days from today? Answer with the full date.
No contextPASS

Thirty days from January 30, 2026, is March 1, 2026.

detector fmi_7_3_v1: date arithmetic correct

General contextPASS

30 days from January 30, 2026, is March 1, 2026.

detector fmi_7_3_v1: date arithmetic correct

Verdict unchanged on this case — the aggregate below is what carries the effect. Across 5 repeats: -2.5pp (intervals overlap).fmi_7_3_v1 · openai/gpt-4o · 2026-09-09 · ground truth as of 2026-08-09

loop 2026-09-09 · openai/gpt-4o as served · closed weights · released 2024-05 · 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%15%30%45%60%No context8%[1%–36%] n=40 · eff 11Irrelevant context5%presence — a document being there[1%–33%] n=40 · eff 11-2.5pp OVERLAPSGeneral context3%content — the information itself[0%–30%] n=40 · eff 11-2.5pp 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

Thirty days from January 30, 2026, is March 1, 2026.

fmi_7_3_v1

detector fmi_7_3_v1: date arithmetic correct

The scoring
Failure rate by context
No context8% [1%–36%] n=40 · eff 11
Irrelevant context5% [1%–33%] n=40 · eff 11
General context3% [0%–30%] n=40 · eff 11
Curated context8% [1%–36%] n=40 · eff 11

0–60% · whisker = 95% interval

Trap vs answerable
No contexttrap 12%answerable 0%
Irrelevant contexttrap 8%answerable 0%
General contexttrap 4%answerable 0%
Curated contexttrap 12%answerable 0%

never pooled · a trap ratio would skew the delta

Residual — what survives

Unchanged by context: 8% with none, 3% 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.

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