ModelCensusopen-source ai reliability harness
Benchmark Studio

anthropic/claude-3-haiku.

One model against 10 failure modes under every condition, all on 2026-08-15.

loop 2026-08-15 · anthropic/claude-3-haiku as served · closed weights · released 2024-03 · set s1.3-live · cases v1.6 · conditions v1.0 · ground truth as of 2026-08-09

anthropic/claude-3-haiku, the same 10 questions across 10 failure modes, 4 times each — changing only the context supplied. 2000 calls, all on 2026-08-15.

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
0–100% failureCitationResolution Failurec1 · 30% [19–44]Basis-DemandEvasionc2 · 50% [35–65]Post-CutoffFabricationc7 · 0% [0–16]ConstraintDecay at Depthc3 · 0% [0–20]Cross-TurnSchema Validityc3 · 25% [11–47]Method/ExecutionSplitc4 · 0% [0–7]Dimensional-AnalysisFailurec4 · 0% [0–7]ParaphraseNon-Invariancec5 · 40% [25–58]LedgerReconciliation Failurec6 · 0% [0–9]Date-ArithmeticErrorc7 · 0% [0–9]

10 modes · set s1.3-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 1 of 10 modes.

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

Treatment modescontext should reach these
Citation Resolution Failure-6.0pp ~
Post-Cutoff Fabrication-100.0pp
Control modesit should not
Basis-Demand Evasion+17.5pp ~
Constraint Decay at Depth0.0pp ~
Cross-Turn Schema Validity+5.0pp ~
Method/Execution Split0.0pp ~
Dimensional-Analysis Failure0.0pp ~
Paraphrase Non-Invariance+20.0pp ~
Ledger Reconciliation Failure-12.5pp ~
Date-Arithmetic Error0.0pp ~

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

anthropic/claude-3-haikuCitation Resolution Failurefmi_1_2
fmi_1_2_v1 · 5 repeats · served 2026-08-15
ConditionFailure rate95% intervalnTrapAnswerable
No context
bare · The model alone, with no supplied context.
28%17%–42%5056% (n=25)0% (n=25)
Irrelevant context
null · Length-matched irrelevant context. The control.
36%24%–50%5072% (n=25)0% (n=25)
General context
recipe_a · Broad retrieval — the first passage the source returns, unreviewed.
34%22%–48%5068% (n=25)0% (n=25)
Curated context
recipe_b · Hand-checked retrieval — the same passage plus the provenance chain behind it.
30%19%–44%5060% (n=25)0% (n=25)
Presence effect · no context → irrelevant context
+8.0pp

intervals overlap — not separated

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

Content effect · irrelevant → best real context
-6.0pp

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: 28% with no context against 34% 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: no citation offered.

Replay
1Questionheld fixed
Cite the paper that introduced the transformer architecture, with its arXiv id.

expected as of 2026-08-09

Attention Is All You Need, arXiv:1706.03762. A real, resolvable reference.

2Contextchange me

Nothing supplied. The baseline.

3ResponseNo context
PASS

The paper that introduced the transformer architecture is: Vaswani, Ashish, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, and Illia Polosukhin. "Attention is all you need." arXiv preprint arXiv:1706.03762 (2017). The arXiv ID for this paper is arXiv:1706.03762.

fmi_1_2_v1

detector fmi_1_2_v1: all 1 citation(s) resolved

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

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