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

Real context moved -2.1pp on Citation Resolution Failure.

~intervals overlap — not separated from no change

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 3 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*. arXiv preprint arXiv:1706.03762. Here's the link to the arXiv preprint: [https://arxiv.org/abs/1706.03762](https://arxiv.org/abs/1706.03762) This paper details a new simple neural network architecture, the Transformer, based solely on attention mechanisms, dispensing with recurrence and convolutions entirely. It achieved state-of-the-art results on machine translation tasks.

detector fmi_1_2_v1: all 1 citation(s) resolved

General contextPASS

The paper that introduced the transformer architecture is: 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: -2.1pp (intervals overlap).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 s1.1-live · cases v1.3 · 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%25%50%75%100%No context65%[43%–82%] n=20Irrelevant context60%presence — a document being there[39%–78%] n=20-5.0pp OVERLAPSGeneral context58%content — the information itself[36%–77%] n=19-2.1pp OVERLAPS
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 context65% [43%–82%] n=20
Irrelevant context60% [39%–78%] n=20
General context58% [36%–77%] n=19
Curated context60% [39%–78%] n=20

0–100% · whisker = 95% interval

Trap vs answerable
No contexttrap 80%answerable 50%
Irrelevant contexttrap 80%answerable 40%
General contexttrap 67%answerable 50%
Curated contexttrap 80%answerable 40%

never pooled · a trap ratio would skew the delta

Residual — what survives

65% with no context, 58% 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 general context · rollups + residuals committed · probe log outside git · manifest hash ties them · method