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

Real context moved 12.5pp on Date-Arithmetic Error.

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: 10% bare against 13% 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

google/gemini-2.5-flash-litefmi_7_35 repeatsfmi_7_3_v1
The modeAll 10 cards
In one frame
screenshot this
Date-Arithmetic Errorfmi_7_3
google/gemini-2.5-flash-lite
Same question, both times
Today is 2026-03-10. What date is 30 days from today? Answer with the full date.
No contextPASS

To calculate the date 30 days from today (2026-03-10), we can proceed as follows: * **Days remaining in March:** March has 31 days. From March 10th, there are 31 - 10 = 21 days left in March. * **Remaining days to count:** We need to count 30 days in total. After using the 21 days in March, we have 30 - 21 = 9 days left to count. * **Counting into the next month:** The next month is April. We need to count 9 days into April. Therefore, 30 days from 2026-03-10 is April 9, 2026. The answer is **2026-04-09**.

detector fmi_7_3_v1: date arithmetic correct

General contextPASS

2026-04-09

detector fmi_7_3_v1: date arithmetic correct

Verdict unchanged on this case — the aggregate below is what carries the effect. Across 5 repeats: +12.5pp (intervals overlap).fmi_7_3_v1 · google/gemini-2.5-flash-lite · 2026-08-15 · ground truth as of 2026-08-09

loop 2026-08-15 · google/gemini-2.5-flash-lite as served · closed weights · released 2025-07 · set s1.3-live · cases v1.6 · 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 context10%[4%–23%] n=40Irrelevant context0%presence — a document being there[0%–9%] n=40-10.0pp OVERLAPSGeneral context13%content — the information itself[5%–26%] n=40+12.5pp 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

No exhibit stored for this context. The full probe log lives outside the repository.

The scoring
Failure rate by context
No context10% [4%–23%] n=40
Irrelevant context0% [0%–9%] n=40
General context13% [5%–26%] n=40
Curated context13% [5%–26%] n=40

0–40% · whisker = 95% interval

Trap vs answerable
No contexttrap 16%answerable 0%
Irrelevant contexttrap 0%answerable 0%
General contexttrap 20%answerable 0%
Curated contexttrap 20%answerable 0%

never pooled · a trap ratio would skew the delta

Residual — what survives

Unchanged by context: 10% with none, 13% 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