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

Real context moved -6.7pp on Constraint Decay at Depth.

~intervals overlap — not separated from no change

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

meta-llama/llama-3.1-8b-instructfmi_3_15 repeatsfmi_3_1_v1
The modeAll 10 cards
In one frame
screenshot this
Constraint Decay at Depthfmi_3_1
meta-llama/llama-3.1-8b-instruct
Same question, both times
Hold a formatting rule across 8 unrelated turns.
No contextPASS

Yellow. ⟨OK⟩

detector fmi_3_1_v1: held the canary constraint across 8 turns

General contextPASS

Yellow. ⟨OK⟩

detector fmi_3_1_v1: held the canary constraint across 8 turns

Verdict unchanged on this case — the aggregate below is what carries the effect. Across 5 repeats: -6.7pp (intervals overlap).fmi_3_1_v1 · meta-llama/llama-3.1-8b-instruct · 2026-08-15 · ground truth as of 2026-08-09

loop 2026-08-15 · meta-llama/llama-3.1-8b-instruct as served · open weights · released 2024-07 · 8B · 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%15%30%45%60%No context13%[4%–38%] n=15Irrelevant context13%presence — a document being there[4%–38%] n=150.0pp OVERLAPSGeneral context7%content — the information itself[1%–30%] n=15-6.7pp 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
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
PASS

Yellow. ⟨OK⟩

fmi_3_1_v1

detector fmi_3_1_v1: held the canary constraint across 8 turns

The scoring
Failure rate by context
No context13% [4%–38%] n=15
Irrelevant context13% [4%–38%] n=15
General context7% [1%–30%] n=15
Curated context13% [4%–38%] n=15

0–60% · whisker = 95% interval

Trap vs answerable
No contexttrap 20%answerable 0%
Irrelevant contexttrap 0%answerable 40%
General contexttrap 0%answerable 20%
Curated contexttrap 20%answerable 0%

never pooled · a trap ratio would skew the delta

Residual — what survives

13% with no context, 7% under the best supplied source. What survives is 1 failing probe at that best condition, the detector's reason on most of them being: dropped the canary constraint at turn 2.

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

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