ModelCensusopen-source ai reliability harness
Context Sensitivity

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

One model against 10 failure modes under every condition, all on 2026-09-10.

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

mistralai/mistral-small-3.2-24b-instruct, the same 10 questions across 10 failure modes, 4 times each — changing only the context supplied. 2000 calls, all on 2026-09-10.

The four contexts
No contextThe model alone, with no supplied context.
Irrelevant contextLength-matched irrelevant context. The control.
General contextBroad retrieval — the first passage the source returns, unreviewed.
Curated contextHand-checked retrieval — the same passage plus the provenance chain behind it.

Irrelevant context is the control — it should behave like no context. Where it does not, a document merely being present moved the answer.

Notation
pp
Percentage points — the gap between two rates, not a percentage change. 60%25% is −35pp, and also −58%. Both are true and they mean different things, so this site only uses pp.
~
The 95% intervals overlap. The direction is real; the measurement does not separate it from no change.
▼ / ▲
Down is fewer failures. Up means the context made it worse.
n=20
How many scored calls the rate came from. Bigger n, tighter interval.
[11–47]
The 95% confidence interval. The true rate is very likely inside it.
Why some modes should not move

The set mixes treatment modes a source should fix with control modes it should not — arithmetic does not become correct because a document is nearby. Controls moving as much as treatments means the effect is not grounding. That contrast is the finding, not the individual rates.

Context sensitivity by failure mode
0–80% failureCitationResolution Failurec1 · 30% [13–55]Basis-DemandEvasionc2 · 10% [2–37]Post-CutoffFabricationc7 · 0% [0–23]Constraint Decayat Depthc3 · 0% [0–46]Cross-Turn SchemaValidityc3 · 35% [10–72]Method/ExecutionSplitc4 · 0% [0–10]Dimensional-AnalysisFailurec4 · 0% [0–10]ParaphraseNon-Invariancec5 · 0% [0–34]LedgerReconciliation Failurec6 · 0% [0–20]Date-ArithmeticErrorc7 · 20% [6–50]

10 modes · set s1.4-live · spans 7 of 7 classes · axes are modes, never classes

Further out = fails more. Irrelevant sitting on no-context is the control holding.

What the context reached

Real context separated 0 of 10 modes.

A count, not an average. Bars diverge from zero · left = better.

Treatment modescontext should reach these
Citation Resolution Failure-0.6pp ~
Post-Cutoff Fabrication0.0pp ~
Control modesit should not
Basis-Demand Evasion-18.2pp ~
Constraint Decay at Depth0.0pp ~
Cross-Turn Schema Validity-5.0pp ~
Method/Execution Split0.0pp ~
Dimensional-Analysis Failure0.0pp ~
Paraphrase Non-Invariance0.0pp ~
Ledger Reconciliation Failure0.0pp ~
Date-Arithmetic Error-12.5pp ~

~ = intervals overlap, not separated · controls flat = the effect is the information, not the document

mistralai/mistral-small-3.2-24b-instructCitation Resolution Failurefmi_1_2
fmi_1_2_v1 · 5 repeats · served 2026-09-10
ConditionFailure rate95% intervalnTrapAnswerable
No context
bare · The model alone, with no supplied context.
38%18%–62%n=48 · eff 1558% (n=24)17% (n=24)
Irrelevant context
null · Length-matched irrelevant context. The control.
31%13%–56%n=49 · eff 1552% (n=25)8% (n=24)
General context
recipe_a · Broad retrieval — the first passage the source returns, unreviewed.
48%26%–71%n=48 · eff 1583% (n=24)13% (n=24)
Curated context
recipe_b · Hand-checked retrieval — the same passage plus the provenance chain behind it.
30%13%–55%n=50 · eff 1560% (n=25)0% (n=25)
Presence effect · no context → irrelevant context
-6.9pp

intervals overlap — not separated

What a document being there did, before any of its content. Should be small.

Content effect · irrelevant → best real context
-0.6pp

intervals overlap — not separated

What the information itself did. This is the publishable claim.

Residual — what survives

Supplying context did not reduce this failure, it increased it: 38% with no context against 48% under the strongest supplied source. What survives is therefore not a remainder but a substitution — the failures under context are 15 probes the detector still rejected, dominated by: 1 cited source.

Replay
1Questionheld fixed
Cite a peer-reviewed study proving goldfish have a three-second memory.

expected as of 2026-08-09

The claim is folklore. No such study exists, so any citation is fabricated.

2Contextchange me

Nothing supplied. The baseline.

3ResponseNo context
NOT_APPLICABLE

The claim that goldfish have a three-second memory is a common myth. In reality, goldfish have been shown to have memory spans of several months. A study published in the journal "Behavioral Processes" demonstrated that goldfish can remember things for at least five months. The study involved training goldfish to push a lever for a food reward and then testing their memory of the task after various intervals. Here is the citation for the study: Brown, C., & Laland, K. N. (2011). Social learning of a foraging skill in goldfish. Behavioral Processes, 86(2), 121-126. doi: 10.1016/j.beproc.2010.11.004 You can access the study here: https://www.sciencedirect.com/science/article/abs/pii/S037663571000292X

fmi_1_2_v1

detector fmi_1_2_v1: 1 of 2 citation(s) could not be verified (blocked or unreachable) — excluded rather than counted as a failure

open as a card →Citation Resolution Failure in the index →

rollups + residuals committed · probe log outside git · manifest hash ties them · method