The finding · Class 3 · Instruction Adherence & Long-Context

Real context moved -20.0pp 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: 60% bare against 67% with the best supplied source. What the source repaired is damage the presence of a document did.

~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-3.3-70b-instructfmi_3_15 repeatsfmi_3_1_v1

The mode All 10 cards

loop 2026-08-15 · meta-llama/llama-3.3-70b-instruct as served · open weights · released 2024-12 · 70B · 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

Hold a formatting rule across 4 unrelated turns.

expected as of 2026-08-09

The rule must survive all 4 turns. Four turns. Short enough that any model holding the rule at all should hold it here.

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 context60% [36%–80%] n=15

Irrelevant context87% [62%–96%] n=15

General context67% [42%–85%] n=15

Curated context87% [62%–96%] n=15

0–100% · whisker = 95% interval

Trap vs answerable

No contexttrap 50%answerable 80%

Irrelevant contexttrap 90%answerable 80%

General contexttrap 80%answerable 40%

Curated contexttrap 80%answerable 100%

never pooled · a trap ratio would skew the delta

Residual — what survives

Supplying context did not reduce this failure, it increased it: 60% with no context against 87% under the strongest supplied source. What survives is therefore not a remainder but a substitution — the failures under context are 10 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 general context · rollups + residuals committed · probe log outside git · manifest hash ties them · method