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

Real context moved 15.0pp on Cross-Turn Schema Validity.

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: 5% bare against 15% with the best supplied source. What the source repaired is damage the presence of a document did.

~intervals overlap — not separated from no change

When asked for a structured output (JSON/schema), the model emits output that fails validation — especially after several turns. The rule

meta-llama/llama-3.1-8b-instructfmi_3_35 repeatsfmi_3_3_v1
The modeAll 10 cards
In one frame
screenshot this
Cross-Turn Schema Validityfmi_3_3
meta-llama/llama-3.1-8b-instruct
Same question, both times
Return a JSON object for a fictional person with fields name (string), age (integer), email (string).
No contextPASS

{ "name": "Max Wells", "age": 42, "email": "max.wells@rockstar.com" }

detector fmi_3_3_v1: all turns produced schema-valid JSON

General contextPASS

{ "name": "Max Wells", "age": 42, "email": "max.wells@rockstar.com" }

detector fmi_3_3_v1: all turns produced schema-valid JSON

Verdict unchanged on this case — the aggregate below is what carries the effect. Across 5 repeats: +15.0pp (intervals overlap).fmi_3_3_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 context5%[1%–24%] n=20Irrelevant context0%presence — a document being there[0%–16%] n=20-5.0pp OVERLAPSGeneral context15%content — the information itself[5%–36%] n=20+15.0pp 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
Return a JSON object for a fictional person with fields name (string), age (integer), email (string).

expected as of 2026-08-09

Schema-valid JSON on all 3 turns. The third turn adds a detail with nowhere to put it — an extra key is the failure.

2Contextchange me

Nothing supplied. The baseline.

3ResponseNo context
PASS

{ "name": "Max Wells", "age": 42, "email": "max.wells@rockstar.com" }

fmi_3_3_v1

detector fmi_3_3_v1: all turns produced schema-valid JSON

The scoring
Failure rate by context
No context5% [1%–24%] n=20
Irrelevant context0% [0%–16%] n=20
General context15% [5%–36%] n=20
Curated context15% [5%–36%] n=20

0–60% · whisker = 95% interval

Trap vs answerable
No contexttrap 10%answerable 0%
Irrelevant contexttrap 0%answerable 0%
General contexttrap 30%answerable 0%
Curated contexttrap 30%answerable 0%

never pooled · a trap ratio would skew the delta

Residual — what survives

Supplying context did not reduce this failure, it increased it: 5% with no context against 15% under the strongest supplied source. What survives is therefore not a remainder but a substitution — the failures under context are 3 probes the detector still rejected, dominated by: turn 3: data must NOT have additional properties.

Mitigations from the index — the claim
  • schema-constrained decoding
  • validate-and-repair loops

fmi_3_3_v1 · 5 repeats per case · best context general context · rollups + residuals committed · probe log outside git · manifest hash ties them · method