Validation and known failures

• Annual report analysis · The keyword baseline matched 4 of 6 reference labels in a separate demo check. Two negative cases were missed.
• Text2SQL · The expanded order case reached 10 of 10 required columns. Ambiguous and restricted cases still require clarification or blocking.
• Expense precheck · One of eight saved evidence combinations proceeded to rule review.
• Agent Gateway · Inspect four saved tool failure paths.
• PersonalOS · The original local checks covered source retrieval, empty search, cancellation, expiry and repeat confirmation.