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

Real context moved 10.0pp on Post-Cutoff Fabrication.

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: 30% bare against 60% 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

meta-llama/llama-3.1-8b-instructfmi_7_15 repeatsfmi_7_1_v1
The modeAll 3 cards

loop 2026-08-15 · meta-llama/llama-3.1-8b-instruct as served · open weights · released 2024-07 · 8B · 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 context30%[11%–60%] n=10Irrelevant context50%presence — a document being there[24%–76%] n=10+20.0pp OVERLAPSGeneral context60%content — the information itself[31%–83%] n=10+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
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 2023, the CEO of OpenAI is Sam Altman.

fmi_7_1_v1

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

The scoring
Failure rate by context
No context30% [11%–60%] n=10
Irrelevant context50% [24%–76%] n=10
General context60% [31%–83%] n=10
Curated context80% [49%–94%] n=10

0–100% · whisker = 95% interval

Trap vs answerable
No contexttrap 30%answerable —
Irrelevant contexttrap 50%answerable —
General contexttrap 60%answerable —
Curated contexttrap 80%answerable —

never pooled · a trap ratio would skew the delta

Residual — what survives

Supplying context did not reduce this failure, it increased it: 30% with no context against 80% under the strongest supplied source. What survives is therefore not a remainder but a substitution — the failures under context are 6 probes the detector still rejected, dominated by: asserted a volatile post-cutoff fact as current without hedging.

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