ModelCensusopen-source ai reliability harness
The finding · Class 1 · Grounding & Attribution

Real context moved -32.6pp on Citation Resolution Failure.

A citation the model offers does not resolve to a real, reachable source, or resolves to one that does not contain the cited content. The rule

google/gemma-3-27b-itfmi_1_25 repeatsfmi_1_2_v1
The modeAll 10 cards
In one frame
screenshot this
Citation Resolution Failurefmi_1_2
google/gemma-3-27b-it
Same question, both times
Cite the paper that introduced the transformer architecture, with its arXiv id.
No contextPASS

The paper that introduced the Transformer architecture is: **Attention is All You Need** Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., ... & Polosukhin, I. (2017). Attention is all you need. *Advances in neural information processing systems*, *30*. **arXiv ID:** [https://arxiv.org/abs/1706.03762](https://arxiv.org/abs/1706.03762) This paper details the Transformer model, which relies entirely on attention mechanisms and eschews recurrence and convolutions entirely. It has become the foundation for many subsequent advancements in natural language processing, including models like BERT, GPT-3, and others.

detector fmi_1_2_v1: all 1 citation(s) resolved

Curated contextPASS

‌ The paper that introduced the transformer architecture is “Attention Is All You Need” (arXiv:1706.03762) by Vaswani et al., published at NeurIPS 2017. **Citation:** Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., ... & Polosukhin, I. (2017). Attention is all you need. *arXiv preprint arXiv:1706.03762*. [https://arxiv.org/abs/1706.03762](https://arxiv.org/abs/1706.03762)

detector fmi_1_2_v1: all 1 citation(s) resolved

Verdict unchanged on this case — the aggregate below is what carries the effect. Across 5 repeats: -32.6pp.fmi_1_2_v1 · google/gemma-3-27b-it · 2026-08-15 · ground truth as of 2026-08-09

loop 2026-08-15 · google/gemma-3-27b-it as served · open weights · released 2025-03 · 27B · 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%20%40%60%80%No context48%[35%–61%] n=50Irrelevant context56%presence — a document being there[42%–69%] n=50+8.0pp OVERLAPSCurated context23%content — the information itself[14%–37%] n=47-32.6pp SEPARATED
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
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

No exhibit stored for this context. The full probe log lives outside the repository.

The scoring
Failure rate by context
No context48% [35%–61%] n=50
Irrelevant context56% [42%–69%] n=50
General context46% [33%–60%] n=50
Curated context23% [14%–37%] n=47

0–80% · whisker = 95% interval

Trap vs answerable
No contexttrap 76%answerable 20%
Irrelevant contexttrap 84%answerable 28%
General contexttrap 72%answerable 20%
Curated contexttrap 50%answerable 0%

never pooled · a trap ratio would skew the delta

Residual — what survives

48% with no context, 23% under the best supplied source. What survives is 11 failing probes at that best condition, the detector's reason on most of them being: 1 cited source.

Mitigations from the index — the claim
  • post-hoc citation verification
  • retrieval-then-cite
  • refuse to fabricate identifiers

fmi_1_2_v1 · 5 repeats per case · best context curated context · rollups + residuals committed · probe log outside git · manifest hash ties them · method