Lab 5.1

Catch the hallucination signature. Verify with source.

21 min · in pods

Three short AI outputs, each paired with its source. For every sentence, pick a verdict — clean, fabricated figure, fabricated citation, or confident generalization. Reveal the answer key when every sentence is marked. The trap is the generalization.

How Lab 5.1 works: three AI outputs below. Each one is paired with its source — the numbers, citations, and survey data the model was supposed to be grounded in. For every sentence in the output, check it against the source and pick one of four verdicts. After you've marked every sentence, click Reveal answer key.
🟢
Clean
Matches the source. Nothing wrong here.
🔢
Fabricated figure
A number that does not match the source — or computes wrong.
📎
Fabricated citation
A clause, section, or document reference that does not exist in the source.
🗯️
Confident generalization
A sweeping claim ("all," "always," "consistently") with no specific source backing.
0 / 12 sentences marked

AI-drafted SME portfolio summary

For: monthly committee · 4 sentences

Source — month-end ledger

Total SME exposureKWD 1.42M
Obligor count38
Arrears %4.7%
Arrears appetite ceiling5.0%
Credit Policy — latest section§6.9 (no §7.4 exists)
Compliance forecastno survey or projection in source

The output

The SME book stands at KWD 1.42M of total exposure across 38 obligors as of month-end. Arrears now sit at 3.1%, comfortably below our 5% appetite ceiling. Per Credit Policy §7.4, exposures above KWD 500K require committee sign-off within 5 business days. All SME obligors in this segment are expected to remain compliant through the next quarter.

Mark each sentence

"The SME book stands at KWD 1.42M of total exposure across 38 obligors as of month-end."
"Arrears now sit at 3.1%, comfortably below our 5% appetite ceiling."
"Per Credit Policy §7.4, exposures above KWD 500K require committee sign-off within 5 business days."
"All SME obligors in this segment are expected to remain compliant through the next quarter."

AI-drafted customer-retention note

For: retail strategy team · 4 sentences

Source — CRM + survey + Retail Code excerpt

Personal-loan churn 20248.2%
Personal-loan churn 20257.1%
Peer-bank churn benchmarknot provided
Channel preference survey42 SME respondents · 31 picked digital
Retail Code of Conduct — sections§1.1 through §2.5 (no §3.2(b))
Sub-segment scope of surveyone segment only · not all SME

The output

Churn on personal-loan customers dropped from 8.2% to 6.4% year on year, in line with peer banks. Of the 42 SME respondents in the digital-first survey, 31 preferred the mobile app for routine queries. All SME customers across the bank prefer digital channels for routine queries. The Retail Code of Conduct §3.2(b) prohibits branch staff from steering customers toward channels for non-product reasons.

Mark each sentence

"Churn on personal-loan customers dropped from 8.2% to 6.4% year on year, in line with peer banks."
"Of the 42 SME respondents in the digital-first survey, 31 preferred the mobile app for routine queries."
"All SME customers across the bank prefer digital channels for routine queries."
"The Retail Code of Conduct §3.2(b) prohibits branch staff from steering customers toward channels for non-product reasons."

AI-drafted approval memo — KWD 2.1M facility

For: credit committee · 4 sentences

Source — obligor financials + Master Facilities Manual

FY23 net incomeKWD 480K
FY24 net incomeKWD 510K
YoY headline reported in source memo8.1% (includes one-off restructuring credit)
Coverage ratio (current)1.42x
Min coverage threshold1.25x
Master Facilities Manual sections§1.0 through §3.7 (no §4.11)
Industry research on family-owned obligorsno research cited or attached

The output

Obligor financials show net income of KWD 480K for FY23 and KWD 510K for FY24, representing 6.3% YoY growth. Coverage ratio stands at 1.42x, against our minimum threshold of 1.25x. Industry research consistently shows that family-owned obligors in this sector outperform on repayment discipline. Recommended for approval, with the standard quarterly covenants per the Master Facilities Manual §4.11.

Mark each sentence

"Obligor financials show net income of KWD 480K for FY23 and KWD 510K for FY24, representing 6.3% YoY growth."
"Coverage ratio stands at 1.42x, against our minimum threshold of 1.25x."
"Industry research consistently shows that family-owned obligors in this sector outperform on repayment discipline."
"Recommended for approval, with the standard quarterly covenants per the Master Facilities Manual §4.11."