BizHallu Portfolio Demo: historical case readout

q_0064: product amounts and ranking relations

What were the top 3 products by merchandise net revenue in April 2011? Give the rank, stock code, product name, and net revenue.

Original Qwen answer; historical selected spans

The top 3 products by merchandise net revenue in April 2011 were:  
1. **REGENCY CAKESTAND 3 TIER** (rank 1) with a net revenue of GBP 14,280.90.  
2. **PARTY BUNTING** (rank 2) with a net revenue of GBP 10,323.87.  
3. **WOODEN UNION JACK BUNTING** (rank 3) with a net revenue of GBP 4,173.18.

Highlight colors preserve the earlier context-bound span judgments. A red amount is not necessarily a number copied from another product. These marks do not assign new relationship labels.

Current interpretation: row fidelity is separate from rank

Ranking within the eight rows shown to Qwen; historical corpus scope is separately documented. Both answers omit the requested stock codes. The six curated relationships are not a whole-answer correctness score.

q_0069: product amounts and ranking relations

What were the top 3 products by merchandise net revenue in September 2011? Give the rank, stock code, product name, and net revenue.

Original Qwen answer; historical selected spans

The top 3 products by merchandise net revenue in September 2011 were:  
1. **JUMBO BAG RED RETROSPOT** (rank 1) with a net revenue of GBP 8630.45.  
2. **PAPER CHAIN KIT 50'S CHRISTMAS** (rank 2) with a net revenue of GBP 5997.25.  
3. **REGENCY CAKESTAND 3 TIER** (rank 3) with a net revenue of GBP 9315.03.

Highlight colors preserve the earlier context-bound span judgments. A red amount is not necessarily a number copied from another product. These marks do not assign new relationship labels.

Current interpretation: row fidelity is separate from rank

Ranking within the eight rows shown to Qwen; historical corpus scope is separately documented. Both answers omit the requested stock codes. The six curated relationships are not a whole-answer correctness score.

What the case readouts do not prove

A list marker is generated before the following product and amount. Its low token uncertainty does not establish that the model was confident about the completed business relationship. Raw teacher-forced token scores and completed-answer evidence checks have different information budgets.

Positive and negative transaction values are sign-based accounting quantities. Negative value is not a verified measure of physical returns; fees and adjustments may be included. Merchandise scope uses a stock-code heuristic, and the historical product grain is stock code plus description.

There is no independent human benchmark, automatic claim extraction or completed-answer detector here. Seven selected examples of atomic outcomes cannot establish detector superiority, error prevalence or measured business loss.

Historical thresholds: one_minus_min_top2_margin >= 0.334333; mean_token_entropy >= 0.004817; mean_spilled_probability_mass_after_top2 >= 0.000225.

Exploratory max test AUPRC 0.835073; exploratory max test F1 0.779412. These are different test-selected signals, not preregistered winners.