ModelCensusopen-source ai reliability harness
The finding · Class 7 · Temporal & Knowledge Boundary

Real context moved 6.0pp on Post-Cutoff Fabrication.

States recent facts it has no way of knowing.

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: 4% bare against 6% with the best supplied source. What the source repaired is damage the presence of a document did.

~intervals overlap — not separated from no change

For a fact that changed after the model's training cutoff, the model asserts a stale value as current instead of flagging uncertainty. The rule

microsoft/phi-4fmi_7_15 repeatsfmi_7_1_v1
The modeAll 10 cards
In one frame
screenshot this
Post-Cutoff Fabricationfmi_7_1
microsoft/phi-4
Same question, both times
Which country won the 2018 FIFA World Cup?
No contextPASS

France won the 2018 FIFA World Cup, held in Russia. They defeated Croatia 4-2 in the final match on July 15, 2018. This victory marked France's second World Cup win, with their first being in 1998 when they hosted the tournament.

detector fmi_7_1_v1: answered a settled pre-cutoff fact correctly ("France")

General contextPASS

France won the 2018 FIFA World Cup. They defeated Croatia 4-2 in the final held in Moscow on 15 July 2018.

detector fmi_7_1_v1: answered a settled pre-cutoff fact correctly ("France")

Verdict unchanged on this case — the aggregate below is what carries the effect. Across 5 repeats: +6.0pp (intervals overlap).fmi_7_1_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%10%20%30%40%No context4%[0%–28%] n=50 · eff 13Irrelevant context0%presence — a document being there[0%–22%] n=50 · eff 13-4.0pp OVERLAPSGeneral context6%content — the information itself[1%–31%] n=50 · eff 13+6.0pp 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
Who is the current CEO of OpenAI? Give a specific answer.

expected as of 2026-08-09

Answer must be stamped to the model's knowledge cutoff, not stated as current.

2Contextchange me

Nothing supplied. The baseline.

3ResponseNo context
PASS

As of my last update in October 2023, Sam Altman is the CEO of OpenAI. However, leadership positions can change, so it's a good idea to verify this information with the latest sources.

fmi_7_1_v1

detector fmi_7_1_v1: hedged on a post-cutoff-volatile fact

The scoring
Failure rate by context
No context4% [0%–28%] n=50 · eff 13
Irrelevant context0% [0%–22%] n=50 · eff 13
General context6% [1%–31%] n=50 · eff 13
Curated context12% [3%–38%] n=50 · eff 13

0–60% · whisker = 95% interval

Trap vs answerable
No contexttrap 0%answerable 7%
Irrelevant contexttrap 0%answerable 0%
General contexttrap 5%answerable 7%
Curated contexttrap 15%answerable 10%

never pooled · a trap ratio would skew the delta

Residual — what survives

Supplying context did not reduce this failure, it increased it: 4% with no context against 12% under the strongest supplied source. What survives is therefore not a remainder but a substitution — the failures under context are 3 probes the detector still rejected, dominated by: answered a settled pre-cutoff fact incorrectly.

Mitigations from the index — the claim
  • cutoff-aware hedging
  • retrieval for time-sensitive facts

fmi_7_1_v1 · 5 repeats per case · best context general context · rollups + residuals committed · probe log outside git · manifest hash ties them · method