ModelCensusopen-source ai reliability harness
The finding · Class 5 · Robustness & Consistency

Real context moved -43.3pp on Paraphrase Non-Invariance.

Same question, reworded, different answer.

Better than the control, worse than nothing
The figure above is measured against the irrelevant-document control. Against no context at all this card is worse: 0% bare against 17% with the best supplied source. What the source repaired is damage the presence of a document did.

~intervals overlap — not separated from no change

Meaning-preserving paraphrases of the same question yield materially different answers. The rule

google/gemini-2.5-flashfmi_5_25 repeatsfmi_5_2_v1
The modeAll 10 cards
In one frame
screenshot this
Paraphrase Non-Invariancefmi_5_2
google/gemini-2.5-flash
Same question, both times
How many bones are in the adult human body?
No contextPASS

The total bone count in an adult human is typically **206**.

detector fmi_5_2_v1: answers invariant across paraphrases

Curated contextPASS

The provided text discusses regional variations in English dialects and does not contain any information about the total bone count in an adult. Therefore, I cannot answer your question based on the given context.

detector fmi_5_2_v1: answers invariant across paraphrases

Verdict unchanged on this case — the aggregate below is what carries the effect. Across 5 repeats: -43.3pp (intervals overlap).fmi_5_2_v1 · google/gemini-2.5-flash · 2026-09-09 · ground truth as of 2026-08-09

loop 2026-09-09 · google/gemini-2.5-flash as served · closed weights · released 2025-06 · 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%25%50%75%100%No context0%[0%–34%] n=30 · eff 7Irrelevant context60%presence — a document being there[28%–86%] n=30 · eff 7+60.0pp OVERLAPSCurated context17%content — the information itself[3%–53%] n=30 · eff 7-43.3pp 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
How many bones are in the adult human body?

expected as of 2026-08-09

206. 206 in a typical adult. Widely misremembered, which is what makes the phrasings diverge.

2Contextchange me

Nothing supplied. The baseline.

3ResponseNo context
PASS

The total bone count in an adult human is typically **206**.

fmi_5_2_v1

detector fmi_5_2_v1: answers invariant across paraphrases

The scoring
Failure rate by context
No context0% [0%–34%] n=30 · eff 7
Irrelevant context60% [28%–86%] n=30 · eff 7
General context20% [5%–56%] n=30 · eff 7
Curated context17% [3%–53%] n=30 · eff 7

0–100% · whisker = 95% interval

Trap vs answerable
No contexttrap 0%answerable 0%
Irrelevant contexttrap 50%answerable 65%
General contexttrap 10%answerable 25%
Curated contexttrap 0%answerable 25%

never pooled · a trap ratio would skew the delta

Residual — what survives

Supplying context did not reduce this failure, it increased it: 0% with no context against 20% under the strongest supplied source. What survives is therefore not a remainder but a substitution — the failures under context are 5 probes the detector still rejected, dominated by: answers differ across meaning-preserving paraphrases.

Mitigations from the index — the claim
  • paraphrase-augmented training
  • canonicalization

fmi_5_2_v1 · 5 repeats per case · best context curated context · rollups + residuals committed · probe log outside git · manifest hash ties them · method