ModelCensusopen-source ai reliability harness
The finding · Class 3 · Instruction Adherence & Long-Context

Real context moved 53.3pp on Constraint Decay at Depth.

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: 87% bare against 93% with the best supplied source. What the source repaired is damage the presence of a document did.

A constraint set early in a long conversation is silently dropped as the dialogue continues. The rule

openai/gpt-3.5-turbofmi_3_15 repeatsfmi_3_1_v1
The modeAll 10 cards

loop 2026-08-15 · openai/gpt-3.5-turbo as served · closed weights · released 2023-03 · 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%25%50%75%100%No context87%[62%–96%] n=15Irrelevant context40%presence — a document being there[20%–64%] n=15-46.7pp OVERLAPSCurated context93%content — the information itself[70%–99%] n=15+53.3pp SEPARATED
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
Hold a formatting rule across 8 unrelated turns.

expected as of 2026-08-09

The rule must survive all 8 turns. Eight turns. The rule has left the recent window for most models.

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 context87% [62%–96%] n=15
Irrelevant context40% [20%–64%] n=15
General context100% [80%–100%] n=15
Curated context93% [70%–99%] n=15

0–100% · whisker = 95% interval

Trap vs answerable
No contexttrap 90%answerable 80%
Irrelevant contexttrap 10%answerable 100%
General contexttrap 100%answerable 100%
Curated contexttrap 90%answerable 100%

never pooled · a trap ratio would skew the delta

Residual — what survives

Supplying context did not reduce this failure, it increased it: 87% with no context against 100% under the strongest supplied source. What survives is therefore not a remainder but a substitution — the failures under context are 14 probes the detector still rejected, dominated by: dropped the canary constraint at turn 1.

Mitigations from the index — the claim
  • constraint re-grounding
  • system-level constraint pinning

fmi_3_1_v1 · 5 repeats per case · best context curated context · rollups + residuals committed · probe log outside git · manifest hash ties them · method