ModelCensusopen-source ai reliability harness
The finding · Class 4 · Reasoning & Calculation

Real context moved 0.0pp on Dimensional-Analysis Failure.

Gets the number right and the unit wrong.

~intervals overlap — not separated from no change

The model produces a quantity with inconsistent or wrong units, or mixes unit systems, yielding a dimensionally invalid result. The rule

openai/gpt-3.5-turbofmi_4_45 repeatsfmi_4_4_v1
The modeAll 10 cards
In one frame
screenshot this
Dimensional-Analysis Failurefmi_4_4
openai/gpt-3.5-turbo
Same question, both times
A tank fills at 3 L/min for 4 minutes, then drains at 500 mL/s for 10 seconds. Net litres?
No contextPASS

First, let's calculate the volume of water added to the tank: Volume added = 3 L/min * 4 min = 12 L Next, let's calculate the volume of water drained from the tank: Volume drained = 500 mL/s * 10 s = 5000 mL = 5 L Now, let's find the net volume in the tank: Net volume = 12 L - 5 L = 7 L Therefore, the net volume in the tank is 7 liters. Answer: 7 L

detector fmi_4_4_v1: correct unit L

General contextPASS

To find the net litres in the tank, we need to calculate the total amount of water that enters the tank and the total amount of water that leaves the tank. Water entering the tank: 3 L/min * 4 min = 12 L Water leaving the tank: 500 mL/s * 10 s = 5000 mL = 5 L Net litres in the tank: 12 L - 5 L = 7 L Answer: 7 L

detector fmi_4_4_v1: correct unit L

Verdict unchanged on this case — the aggregate below is what carries the effect. Across 5 repeats: 0.0pp (intervals overlap).fmi_4_4_v1 · openai/gpt-3.5-turbo · 2026-09-09 · ground truth as of 2026-08-09

loop 2026-09-09 · openai/gpt-3.5-turbo as served · closed weights · released 2023-03 · 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%5%10%15%20%No context0%[0%–10%] n=50 · eff 36Irrelevant context0%presence — a document being there[0%–10%] n=50 · eff 360.0pp OVERLAPSGeneral context0%content — the information itself[0%–10%] n=50 · eff 360.0pp 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
A tank fills at 3 L/min for 4 minutes, then drains at 500 mL/s for 10 seconds. Net litres?

expected as of 2026-08-09

12 L in, 5 L out, net 7 L. The mixed units are the trap.

2Contextchange me

Nothing supplied. The baseline.

3ResponseNo context
PASS

First, let's calculate the volume of water added to the tank: Volume added = 3 L/min * 4 min = 12 L Next, let's calculate the volume of water drained from the tank: Volume drained = 500 mL/s * 10 s = 5000 mL = 5 L Now, let's find the net volume in the tank: Net volume = 12 L - 5 L = 7 L Therefore, the net volume in the tank is 7 liters. Answer: 7 L

fmi_4_4_v1

detector fmi_4_4_v1: correct unit L

The scoring
Failure rate by context
No context0% [0%–10%] n=50 · eff 36
Irrelevant context0% [0%–10%] n=50 · eff 36
General context0% [0%–10%] n=50 · eff 36
Curated context0% [0%–10%] n=50 · eff 36

0–20% · whisker = 95% interval

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

never pooled · a trap ratio would skew the delta

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
  • explicit unit tracking
  • dimensional checkers

fmi_4_4_v1 · 5 repeats per case · best context general context · rollups + residuals committed · probe log outside git · manifest hash ties them · method