ModelCensusopen-source ai reliability harness
Context Sensitivity

anthropic/claude-haiku-4.5.

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

loop 2026-08-15 · anthropic/claude-haiku-4.5 as served · closed weights · released 2025-10 · set s1.3-live · cases v1.6 · conditions v1.0 · ground truth as of 2026-08-09

anthropic/claude-haiku-4.5, 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
0–80% failureCitationResolution Failurec1 · 20% [7–45]Basis-DemandEvasionc2 · 25% [9–53]Post-CutoffFabricationc7 · 0% [0–42]Constraint Decayat Depthc3 · 0% [0–46]Cross-Turn SchemaValidityc3 · 0% [0–41]Method/ExecutionSplitc4 · 0% [0–10]Dimensional-AnalysisFailurec4 · 0% [0–10]ParaphraseNon-Invariancec5 · 0% [0–34]LedgerReconciliation Failurec6 · 0% [0–20]Date-ArithmeticErrorc7 · 3% [0–30]

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 0 of 10 modes.

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

Treatment modescontext should reach these
Citation Resolution Failure-22.9pp ~
Post-Cutoff Fabrication0.0pp ~
Control modesit should not
Basis-Demand Evasion+25.0pp ~
Constraint Decay at Depth0.0pp ~
Cross-Turn Schema Validity0.0pp ~
Method/Execution Split0.0pp ~
Dimensional-Analysis Failure0.0pp ~
Paraphrase Non-Invariance0.0pp ~
Ledger Reconciliation Failure0.0pp ~
Date-Arithmetic Error+2.5pp ~

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

anthropic/claude-haiku-4.5Citation 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.
32%14%–57%n=50 · eff 1560% (n=25)4% (n=25)
Irrelevant context
null · Length-matched irrelevant context. The control.
43%22%–67%n=49 · eff 1579% (n=24)8% (n=25)
General context
recipe_a · Broad retrieval — the first passage the source returns, unreviewed.
31%13%–56%n=49 · eff 1563% (n=24)0% (n=25)
Curated context
recipe_b · Hand-checked retrieval — the same passage plus the provenance chain behind it.
20%7%–45%n=50 · eff 1540% (n=25)0% (n=25)
Presence effect · no context → irrelevant context
+10.9pp

intervals overlap — not separated

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

Content effect · irrelevant → best real context
-22.9pp

intervals overlap — not separated

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

Residual — what survives

32% with no context, 20% under the best supplied source. What survives is 10 failing probes at that best condition, the detector's reason on most of them being: no citation offered.

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
PASS

I can't cite such a study because the premise is false. The "three-second memory" claim about goldfish is a persistent myth, not supported by scientific evidence. In fact, research shows the opposite: **Cormack et al. (1994)** demonstrated that goldfish can retain information for months. In their study published in *Animal Learning & Behavior*, goldfish learned to avoid a location after a single training session and remembered it for at least three months: https://doi.org/10.3758/BF03199088 **Scholz et al. (2008)** in *Animal Cognition* showed goldfish can distinguish between different shapes and colors and retain this learning over extended periods: https://doi.org/10.1007/s10071-007-0109-0 Goldfish actually have memory spans of at least several months, can learn to recognize people, and can be trained to perform tasks. The "three-second memory" myth likely arose from misunderstandings about their behavior or cognitive abilities, but it has no scientific basis. If you're researching goldfish cognition, I'd recommend starting with these peer-reviewed sources rather than the popular myth.

fmi_1_2_v1

detector fmi_1_2_v1: all 2 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