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

Real context moved 0.0pp on Method/Execution Split.

~intervals overlap — not separated from no change

The model chooses a correct method but executes it wrongly (or vice versa) — a right-approach/wrong-answer or wrong-approach/right-answer split. The rule

qwen/qwen3-235b-a22b-2507fmi_4_15 repeatsfmi_4_1_v1
In one frame
screenshot this
Method/Execution Splitfmi_4_1
qwen/qwen3-235b-a22b-2507
Same question, both times
A shop sells 17 boxes holding 24 items each, then 8 items are returned. How many items were sold?
No contextPASS

First, calculate the total number of items in the 17 boxes: 17 boxes × 24 items per box = 408 items Then, subtract the 8 items that were returned: 408 − 8 = 400 items Answer: 400

detector fmi_4_1_v1: answer matches oracle

Curated contextPASS

First, calculate the total number of items sold initially: 17 boxes × 24 items per box = 408 items Then, subtract the 8 items that were returned: 408 − 8 = 400 items Answer: 400

detector fmi_4_1_v1: answer matches oracle

Verdict unchanged on this case — the aggregate below is what carries the effect. Across 5 repeats: 0.0pp (intervals overlap).fmi_4_1_v1 · qwen/qwen3-235b-a22b-2507 · 2026-08-15 · ground truth as of 2026-08-09

loop 2026-08-15 · qwen/qwen3-235b-a22b-2507 as served · open weights · released 2025-07 · 235B-A22B · 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%5%10%15%20%No context0%[0%7%] n=50Irrelevant context0%presence — a document being there[0%7%] n=500.0pp OVERLAPSCurated context0%content — the information itself[0%7%] n=500.0pp 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 shop sells 17 boxes holding 24 items each, then 8 items are returned. How many items were sold?

expected as of 2026-08-09

400. Order matters: multiply before subtracting.

2Contextchange me

Nothing supplied. The baseline.

3ResponseNo context
PASS

First, calculate the total number of items in the 17 boxes: 17 boxes × 24 items per box = 408 items Then, subtract the 8 items that were returned: 408 − 8 = 400 items Answer: 400

fmi_4_1_v1

detector fmi_4_1_v1: answer matches oracle

The scoring
Failure rate by context
No context0% [0%7%] n=50
Irrelevant context0% [0%7%] n=50
General context4% [1%13%] n=50
Curated context0% [0%7%] n=50

0–20% · whisker = 95% interval

Trap vs answerable
No contexttrap 0%answerable 0%
Irrelevant contexttrap 0%answerable 0%
General contexttrap 7%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
  • tool-assisted computation
  • verify final against method

fmi_4_1_v1 · 5 repeats per case · best context curated context · rollups + residuals committed · probe log outside git · manifest hash ties them · method