The finding · Class 7 · Temporal & Knowledge Boundary

Real context moved -17.5pp on Date-Arithmetic Error.

~intervals overlap — not separated from no change

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

google/gemma-3-27b-itfmi_7_35 repeatsfmi_7_3_v1

The mode All 10 cards

In one frame

screenshot this

Date-Arithmetic Errorfmi_7_3

google/gemma-3-27b-it

Same question, both times

Today is 2026-01-30. What date is 30 days from today? Answer with the full date.

No contextPASS

Today is 2026-01-30. We want to find the date 30 days from today. January has 31 days. So, from 2026-01-30, we add 30 days. Since January has 31 days, there is 31 - 30 = 1 day left in January. So, we add 1 day to reach the end of January, which is 2026-01-31. We have 30 - 1 = 29 days remaining. The next month is February. In 2026, February has 28 days since 2026 is not a leap year. So, we add 28 days to reach the end of February, which is 2026-02-28. We have 29 - 28 = 1 day remaining. The next month is March. We add 1 day to reach 2026-03-01. So, 30 days from 2026-01-30 is 2026-03-01. Final Answer: The final answer is $\boxed{2026-03-01}$

detector fmi_7_3_v1: date arithmetic correct

Curated contextPASS

Today is 2026-01-30. Adding 30 days to 2026-01-30: January has 31 days, so 31 - 30 = 1 day remaining in January. 30 - 1 = 29 days remaining to add. February 2026 has 28 days (2026 is not a leap year). 29 - 28 = 1 day remaining to add. Therefore, the date is March 1, 2026. Answer: 2026-03-01

detector fmi_7_3_v1: date arithmetic correct

loop 2026-08-15 · google/gemma-3-27b-it as served · open weights · released 2025-03 · 27B · 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

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

fmi_7_3_v1

detector fmi_7_3_v1: date arithmetic correct

The scoring

Failure rate by context

No context20% [10%–35%] n=40

Irrelevant context23% [12%–38%] n=40

General context8% [3%–20%] n=40

Curated context5% [1%–17%] n=40

0–60% · whisker = 95% interval

Trap vs answerable

No contexttrap 32%answerable 0%

Irrelevant contexttrap 36%answerable 0%

General contexttrap 12%answerable 0%

Curated contexttrap 8%answerable 0%

never pooled · a trap ratio would skew the delta

Residual — what survives

20% with no context, 5% under the best supplied source. What survives is 2 failing probes at that best condition, the detector's reason on most of them being: 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 curated context · rollups + residuals committed · probe log outside git · manifest hash ties them · method