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

Real context moved -10.0pp on Citation Resolution Failure.

Cites a source that does not exist.

~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

microsoft/phi-4fmi_1_25 repeatsfmi_1_2_v1
The modeAll 10 cards
In one frame
screenshot this
Citation Resolution Failurefmi_1_2
microsoft/phi-4
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 titled "Attention Is All You Need." It was published by Vaswani et al. in 2017. The arXiv identifier for this paper is arXiv:1706.03762. You can access it via the following link: [Attention Is All You Need](https://arxiv.org/abs/1706.03762) This paper is foundational in the field of natural language processing and has significantly influenced the development of subsequent models and architectures.

detector fmi_1_2_v1: all 1 citation(s) resolved

Curated contextPASS

The transformer architecture was introduced in the paper titled "Attention Is All You Need" by Vaswani et al. The arXiv identifier for this paper is arXiv:1706.03762. You can access the paper through the following URL: [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: -10.0pp (intervals overlap).fmi_1_2_v1 · microsoft/phi-4 · 2026-09-09 · ground truth as of 2026-08-09

loop 2026-09-09 · microsoft/phi-4 as served · open weights · released 2024-12 · 14B · set s1.4-live · cases v1.7 · 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 context34%[16%–59%] n=50 · eff 15Irrelevant context36%presence — a document being there[17%–61%] n=50 · eff 15+2.0pp OVERLAPSCurated context26%content — the information itself[11%–51%] n=50 · eff 15-10.0pp 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 context34% [16%–59%] n=50 · eff 15
Irrelevant context36% [17%–61%] n=50 · eff 15
General context36% [17%–61%] n=50 · eff 15
Curated context26% [11%–51%] n=50 · eff 15

0–80% · whisker = 95% interval

Trap vs answerable
No contexttrap 56%answerable 12%
Irrelevant contexttrap 60%answerable 12%
General contexttrap 52%answerable 20%
Curated contexttrap 52%answerable 0%

never pooled · a trap ratio would skew the delta

Residual — what survives

34% with no context, 26% under the best supplied source. What survives is 13 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