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

Real context moved -5.0pp on Cross-Turn Schema Validity.

Breaks the JSON shape partway through a conversation.

~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

mistralai/mistral-small-3.2-24b-instructfmi_3_35 repeatsfmi_3_3_v1

The mode All 10 cards

In one frame

screenshot this

Cross-Turn Schema Validityfmi_3_3

mistralai/mistral-small-3.2-24b-instruct

Same question, both times

Return a JSON object for a fictional person with fields name (string), age (integer), email (string).

No contextPASS

```json { "name": "Sarah Williams", "age": 28, "email": "sarah.williams@example.com" } ```

detector fmi_3_3_v1: all turns produced schema-valid JSON

Curated contextPASS

```json { "name": "Jane Smith", "age": 25, "email": "jane.smith@example.com" } ```

detector fmi_3_3_v1: all turns produced schema-valid JSON

loop 2026-09-10 · mistralai/mistral-small-3.2-24b-instruct as served · open weights · released 2025-06 · 24B · set s1.4-live · cases v1.7 · 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

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 2 turns. Two turns, one shape. The second turn is where drift starts.

2Contextchange me

Nothing supplied. The baseline.

3ResponseNo context

PASS

```json { "name": "Sarah Williams", "age": 28, "email": "sarah.williams@example.com" } ```

fmi_3_3_v1

detector fmi_3_3_v1: all turns produced schema-valid JSON

The scoring

Failure rate by context

No context45% [15%–79%] n=20 · eff 6

Irrelevant context40% [13%–76%] n=20 · eff 6

General context40% [13%–76%] n=20 · eff 6

Curated context35% [10%–72%] n=20 · eff 6

0–100% · whisker = 95% interval

Trap vs answerable

No contexttrap 90%answerable 0%

Irrelevant contexttrap 80%answerable 0%

General contexttrap 80%answerable 0%

Curated contexttrap 70%answerable 0%

never pooled · a trap ratio would skew the delta

Residual — what survives

45% with no context, 35% under the best supplied source. What survives is 7 failing probes at that best condition, the detector's reason on most of them being: 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 curated context · rollups + residuals committed · probe log outside git · manifest hash ties them · method