BHBizHallu Detector interpretation

Presentation-level label confirmation

Review the examples behind the detector claims.

This packet selects the spans most likely to appear in the final story: detector FP/FN examples, top-3 confident misses, currency patterns, and the rank-marker offset regression checks. Every item remains pending human confirmation.

Selected spans15
Questions covered9
Error example rows20
StatusPending

How to use this packet

Confirm each highlighted span, gold reference, label, and reason. If any item is wrong, update the source annotation first, then rerun alignment, scoring, metrics, error review, and interpretation.

Selected confirmation items

confirm_001 Pending human review

q_0039 - currency_amount - correct_key_fact

In October 2011, did Germany or France generate more net revenue, and by how much?

...25,017.64 GBP) than Germany (30,604.27 GBP), with France earning a net revenue of -8,453.41 GBP. The difference in net revenue is [[5,586.83 GBP]].
Gold answer
Germany generated more net revenue than France by GBP 5,586.63 in October 2011.
Reason
Generated comparison difference is within tolerance of GBP 5,586.63.
Why selected
detector_error_example, currency_amount_pattern
Detector role
energy false_positive via mean_spilled_probability_mass_after_top2
Confirmation checklist
  • Highlighted span points to the intended generated claim.
  • Gold reference matches the deterministic gold answer/evidence.
  • Label is correct_key_fact only when the generated claim is supported; otherwise hallucinated_key_fact or unsupported_claim is justified.
  • Annotation reason is specific enough to explain the label in a presentation.
  • Detector error type is presentation-safe: false positive means correct span flagged; false negative means wrong/unsupported span missed.
Gold reference 
{
  "field": "revenue_delta",
  "fact_type": "currency_amount",
  "gold_value": 5586.63,
  "gold_display_value": "GBP 5,586.63",
  "tolerance": {
    "absolute": 1.0,
    "relative_percent": 0.5
  }
}
confirm_002 Pending human review

q_0054 - comparison_direction - correct_key_fact

How did net revenue change from May 2011 to June 2011? Provide the absolute change and percentage change.

From May 2011 to June 2011, net revenue [[decreased]] by **GBP 3,121.37**, representing a **-4.14%** change. The absolute change is **-GBP 3,121.37**, and the percentage change is **-...
Gold answer
Net revenue decreased by GBP 32,116.87 (-4.45%) from May 2011 to June 2011.
Reason
Generated direction matches the gold direction.
Why selected
detector_error_example
Detector role
energy false_positive via mean_spilled_probability_mass_after_top2
Confirmation checklist
  • Highlighted span points to the intended generated claim.
  • Gold reference matches the deterministic gold answer/evidence.
  • Label is correct_key_fact only when the generated claim is supported; otherwise hallucinated_key_fact or unsupported_claim is justified.
  • Annotation reason is specific enough to explain the label in a presentation.
  • Detector error type is presentation-safe: false positive means correct span flagged; false negative means wrong/unsupported span missed.
Gold reference 
{
  "field": "direction",
  "fact_type": "comparison_direction",
  "gold_value": "decrease",
  "gold_display_value": "decrease",
  "tolerance": "exact_normalized_match"
}
confirm_003 Pending human review

q_0054 - percentage - correct_key_fact

How did net revenue change from May 2011 to June 2011? Provide the absolute change and percentage change.

From May 2011 to June 2011, net revenue decreased by **GBP 3,121.37**, representing a **[[-4.14%]]** change. The absolute change is **-GBP 3,121.37**, and the percentage change is **-4.14%**.
Gold answer
Net revenue decreased by GBP 32,116.87 (-4.45%) from May 2011 to June 2011.
Reason
Generated percentage change is within tolerance of -4.45%.
Why selected
detector_error_example
Detector role
energy false_positive via mean_spilled_probability_mass_after_top2; simple false_positive via one_minus_min_top2_margin
Confirmation checklist
  • Highlighted span points to the intended generated claim.
  • Gold reference matches the deterministic gold answer/evidence.
  • Label is correct_key_fact only when the generated claim is supported; otherwise hallucinated_key_fact or unsupported_claim is justified.
  • Annotation reason is specific enough to explain the label in a presentation.
  • Detector error type is presentation-safe: false positive means correct span flagged; false negative means wrong/unsupported span missed.
Gold reference 
{
  "field": "percent_change",
  "fact_type": "percentage",
  "gold_value": -4.45,
  "gold_display_value": "-4.45%",
  "tolerance": {
    "percentage_points": 0.5
  }
}
confirm_004 Pending human review

q_0059 - comparison_direction - correct_key_fact

How did net revenue change from October 2011 to November 2011? Provide the absolute change and percentage change.

The net revenue for November 2011 was 145,614.50 GBP, which represents a **29.75%** [[increase]] from October 2011's 106,936.83 GBP. The absolute change is **38,677.67 GBP**, and the percentage change is calculated by dividing...
Gold answer
Net revenue increased by GBP 386,777.57 (36.17%) from October 2011 to November 2011.
Reason
Generated direction matches the gold direction.
Why selected
detector_error_example
Detector role
energy false_positive via mean_spilled_probability_mass_after_top2
Confirmation checklist
  • Highlighted span points to the intended generated claim.
  • Gold reference matches the deterministic gold answer/evidence.
  • Label is correct_key_fact only when the generated claim is supported; otherwise hallucinated_key_fact or unsupported_claim is justified.
  • Annotation reason is specific enough to explain the label in a presentation.
  • Detector error type is presentation-safe: false positive means correct span flagged; false negative means wrong/unsupported span missed.
Gold reference 
{
  "field": "direction",
  "fact_type": "comparison_direction",
  "gold_value": "increase",
  "gold_display_value": "increase",
  "tolerance": "exact_normalized_match"
}
confirm_005 Pending human review

q_0063 - ranking - hallucinated_key_fact

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

...BERRY** (85099F) with a net revenue of GBP 5853.80. 2. **JUMBO BAG PINK POLKADOT** (22386) with a net revenue of GBP 5079.29. [[3.]] **VINTAGE UNION JACK BUNTING** (21621) with a net revenue of GBP 5214.33. The rankings are based on the provided net_revenue_g...
Gold answer
Top 3 products in March 2011: 1. 22423 (REGENCY CAKESTAND 3 TIER), GBP 16,817.85; 2. 85099B (JUMBO BAG RED RETROSPOT), GBP 10,116.43; 3. 47566 (PARTY BUNTING), GBP 9,098.27.
Reason
The rank-3 marker is tied to the wrong generated product.
Why selected
offset_regression_check
Detector role
No detector-error example; offset regression check only.
Confirmation checklist
  • Highlighted span points to the intended generated claim.
  • Gold reference matches the deterministic gold answer/evidence.
  • Label is correct_key_fact only when the generated claim is supported; otherwise hallucinated_key_fact or unsupported_claim is justified.
  • Annotation reason is specific enough to explain the label in a presentation.
  • Rank marker span is the list marker, not a decimal inside a currency amount.
Gold reference 
{
  "field": "top_products[2].rank_item",
  "fact_type": "ranking",
  "gold_value": "3. 47566",
  "gold_display_value": "3. 47566 / PARTY BUNTING"
}
confirm_006 Pending human review

q_0064 - currency_amount - correct_key_fact

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

...he 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 reve...
Gold answer
Top 3 products in April 2011: 1. 22423 (REGENCY CAKESTAND 3 TIER), GBP 14,280.90; 2. 47566 (PARTY BUNTING), GBP 10,323.87; 3. 22084 (PAPER CHAIN KIT EMPIRE), GBP 6,619.51.
Reason
Generated rank-1 revenue matches the gold rank-1 revenue.
Why selected
detector_error_example, top3_confident_miss_pattern, currency_amount_pattern
Detector role
simple false_positive via one_minus_min_top2_margin
Confirmation checklist
  • Highlighted span points to the intended generated claim.
  • Gold reference matches the deterministic gold answer/evidence.
  • Label is correct_key_fact only when the generated claim is supported; otherwise hallucinated_key_fact or unsupported_claim is justified.
  • Annotation reason is specific enough to explain the label in a presentation.
  • Detector error type is presentation-safe: false positive means correct span flagged; false negative means wrong/unsupported span missed.
Gold reference 
{
  "field": "top_products[0].merchandise_net_revenue",
  "fact_type": "currency_amount",
  "gold_value": 14280.9,
  "gold_display_value": "GBP 14,280.90"
}
confirm_007 Pending human review

q_0064 - product_name - correct_key_fact

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

...erchandise 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.
Gold answer
Top 3 products in April 2011: 1. 22423 (REGENCY CAKESTAND 3 TIER), GBP 14,280.90; 2. 47566 (PARTY BUNTING), GBP 10,323.87; 3. 22084 (PAPER CHAIN KIT EMPIRE), GBP 6,619.51.
Reason
Generated rank-2 product matches the gold rank-2 product.
Why selected
detector_error_example, top3_confident_miss_pattern
Detector role
simple false_positive via one_minus_min_top2_margin
Confirmation checklist
  • Highlighted span points to the intended generated claim.
  • Gold reference matches the deterministic gold answer/evidence.
  • Label is correct_key_fact only when the generated claim is supported; otherwise hallucinated_key_fact or unsupported_claim is justified.
  • Annotation reason is specific enough to explain the label in a presentation.
  • Detector error type is presentation-safe: false positive means correct span flagged; false negative means wrong/unsupported span missed.
Gold reference 
{
  "field": "top_products[1].description",
  "fact_type": "product_name",
  "gold_value": "PARTY BUNTING",
  "gold_display_value": "PARTY BUNTING"
}
confirm_008 Pending human review

q_0064 - ranking - hallucinated_key_fact

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

...AND 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.
Gold answer
Top 3 products in April 2011: 1. 22423 (REGENCY CAKESTAND 3 TIER), GBP 14,280.90; 2. 47566 (PARTY BUNTING), GBP 10,323.87; 3. 22084 (PAPER CHAIN KIT EMPIRE), GBP 6,619.51.
Reason
The rank-3 marker is tied to the wrong generated product.
Why selected
detector_error_example, offset_regression_check, top3_confident_miss_pattern
Detector role
energy false_negative via mean_spilled_probability_mass_after_top2; simple false_negative via one_minus_min_top2_margin
Confirmation checklist
  • Highlighted span points to the intended generated claim.
  • Gold reference matches the deterministic gold answer/evidence.
  • Label is correct_key_fact only when the generated claim is supported; otherwise hallucinated_key_fact or unsupported_claim is justified.
  • Annotation reason is specific enough to explain the label in a presentation.
  • Detector error type is presentation-safe: false positive means correct span flagged; false negative means wrong/unsupported span missed.
  • Rank marker span is the list marker, not a decimal inside a currency amount.
Gold reference 
{
  "field": "top_products[2].rank_item",
  "fact_type": "ranking",
  "gold_value": "3. 22084",
  "gold_display_value": "3. 22084 / PAPER CHAIN KIT EMPIRE"
}
confirm_009 Pending human review

q_0064 - currency_amount - hallucinated_key_fact

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

...**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]].
Gold answer
Top 3 products in April 2011: 1. 22423 (REGENCY CAKESTAND 3 TIER), GBP 14,280.90; 2. 47566 (PARTY BUNTING), GBP 10,323.87; 3. 22084 (PAPER CHAIN KIT EMPIRE), GBP 6,619.51.
Reason
Gold rank-3 revenue is GBP 6,619.51, not GBP 4,173.18.
Why selected
detector_error_example, top3_confident_miss_pattern, currency_amount_pattern
Detector role
energy false_negative via mean_spilled_probability_mass_after_top2; simple false_negative via one_minus_min_top2_margin
Confirmation checklist
  • Highlighted span points to the intended generated claim.
  • Gold reference matches the deterministic gold answer/evidence.
  • Label is correct_key_fact only when the generated claim is supported; otherwise hallucinated_key_fact or unsupported_claim is justified.
  • Annotation reason is specific enough to explain the label in a presentation.
  • Detector error type is presentation-safe: false positive means correct span flagged; false negative means wrong/unsupported span missed.
Gold reference 
{
  "field": "top_products[2].merchandise_net_revenue",
  "fact_type": "currency_amount",
  "gold_value": 6619.51,
  "gold_display_value": "GBP 6,619.51"
}
confirm_010 Pending human review

q_0068 - ranking - hallucinated_key_fact

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

...* (rank 1), with a net revenue of GBP 4,728.73. 2. **JUMBO BAG RED RETROSPOT** (rank 2), with a net revenue of GBP 10,441.77. [[3.]] **SPOTTY BUNTING** (rank 3), with a net revenue of GBP 7,864.43.
Gold answer
Top 3 products in August 2011: 1. 85099B (JUMBO BAG RED RETROSPOT), GBP 10,441.77; 2. 84879 (ASSORTED COLOUR BIRD ORNAMENT), GBP 9,867.04; 3. 47566 (PARTY BUNTING), GBP 9,655.25.
Reason
The rank-3 marker is tied to the wrong generated product.
Why selected
offset_regression_check
Detector role
No detector-error example; offset regression check only.
Confirmation checklist
  • Highlighted span points to the intended generated claim.
  • Gold reference matches the deterministic gold answer/evidence.
  • Label is correct_key_fact only when the generated claim is supported; otherwise hallucinated_key_fact or unsupported_claim is justified.
  • Annotation reason is specific enough to explain the label in a presentation.
  • Rank marker span is the list marker, not a decimal inside a currency amount.
Gold reference 
{
  "field": "top_products[2].rank_item",
  "fact_type": "ranking",
  "gold_value": "3. 47566",
  "gold_display_value": "3. 47566 / PARTY BUNTING"
}
confirm_011 Pending human review

q_0069 - ranking - hallucinated_key_fact

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

...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 ...
Gold answer
Top 3 products in September 2011: 1. 23243 (SET OF TEA COFFEE SUGAR TINS PANTRY), GBP 9,971.51; 2. 22423 (REGENCY CAKESTAND 3 TIER), GBP 9,315.03; 3. 85099B (JUMBO BAG RED RETROSPOT), GBP 8,630.45.
Reason
The rank-2 marker is tied to the wrong generated product.
Why selected
detector_error_example, top3_confident_miss_pattern
Detector role
energy false_negative via mean_spilled_probability_mass_after_top2; simple false_negative via one_minus_min_top2_margin
Confirmation checklist
  • Highlighted span points to the intended generated claim.
  • Gold reference matches the deterministic gold answer/evidence.
  • Label is correct_key_fact only when the generated claim is supported; otherwise hallucinated_key_fact or unsupported_claim is justified.
  • Annotation reason is specific enough to explain the label in a presentation.
  • Detector error type is presentation-safe: false positive means correct span flagged; false negative means wrong/unsupported span missed.
Gold reference 
{
  "field": "top_products[1].rank_item",
  "fact_type": "ranking",
  "gold_value": "2. 22423",
  "gold_display_value": "2. 22423 / REGENCY CAKESTAND 3 TIER"
}
confirm_012 Pending human review

q_0069 - ranking - hallucinated_key_fact

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

...(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.
Gold answer
Top 3 products in September 2011: 1. 23243 (SET OF TEA COFFEE SUGAR TINS PANTRY), GBP 9,971.51; 2. 22423 (REGENCY CAKESTAND 3 TIER), GBP 9,315.03; 3. 85099B (JUMBO BAG RED RETROSPOT), GBP 8,630.45.
Reason
The rank-3 marker is tied to the wrong generated product.
Why selected
detector_error_example, top3_confident_miss_pattern
Detector role
energy false_negative via mean_spilled_probability_mass_after_top2; simple false_negative via one_minus_min_top2_margin
Confirmation checklist
  • Highlighted span points to the intended generated claim.
  • Gold reference matches the deterministic gold answer/evidence.
  • Label is correct_key_fact only when the generated claim is supported; otherwise hallucinated_key_fact or unsupported_claim is justified.
  • Annotation reason is specific enough to explain the label in a presentation.
  • Detector error type is presentation-safe: false positive means correct span flagged; false negative means wrong/unsupported span missed.
Gold reference 
{
  "field": "top_products[2].rank_item",
  "fact_type": "ranking",
  "gold_value": "3. 85099B",
  "gold_display_value": "3. 85099B / JUMBO BAG RED RETROSPOT"
}
confirm_013 Pending human review

q_0069 - currency_amount - hallucinated_key_fact

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

...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]].
Gold answer
Top 3 products in September 2011: 1. 23243 (SET OF TEA COFFEE SUGAR TINS PANTRY), GBP 9,971.51; 2. 22423 (REGENCY CAKESTAND 3 TIER), GBP 9,315.03; 3. 85099B (JUMBO BAG RED RETROSPOT), GBP 8,630.45.
Reason
GBP 9,315.03 is the gold rank-2 revenue, but the generated answer uses it for rank 3.
Why selected
detector_error_example, top3_confident_miss_pattern, currency_amount_pattern
Detector role
energy false_negative via mean_spilled_probability_mass_after_top2; simple false_negative via one_minus_min_top2_margin
Confirmation checklist
  • Highlighted span points to the intended generated claim.
  • Gold reference matches the deterministic gold answer/evidence.
  • Label is correct_key_fact only when the generated claim is supported; otherwise hallucinated_key_fact or unsupported_claim is justified.
  • Annotation reason is specific enough to explain the label in a presentation.
  • Detector error type is presentation-safe: false positive means correct span flagged; false negative means wrong/unsupported span missed.
Gold reference 
{
  "field": "top_products[2].merchandise_net_revenue",
  "fact_type": "currency_amount",
  "gold_value": 8630.45,
  "gold_display_value": "GBP 8,630.45"
}
confirm_014 Pending human review

q_0093 - currency_amount - correct_key_fact

In April 2011, how much did cancellations and returns reduce gross positive revenue, and what was the final net revenue?

In April 2011, cancellations and returns reduced gross positive revenue by [[£44,600.65]], resulting in a final net revenue of £44,600.65.
Gold answer
Cancellations and returns reduced gross positive revenue by GBP 44,600.65 (8.31% of gross), leaving net revenue of GBP 492,367.84.
Reason
Generated reduction amount matches GBP 44,600.65; the pound symbol is treated as GBP-compatible in this context.
Why selected
detector_error_example, currency_amount_pattern
Detector role
energy false_positive via mean_spilled_probability_mass_after_top2; simple false_positive via one_minus_min_top2_margin
Confirmation checklist
  • Highlighted span points to the intended generated claim.
  • Gold reference matches the deterministic gold answer/evidence.
  • Label is correct_key_fact only when the generated claim is supported; otherwise hallucinated_key_fact or unsupported_claim is justified.
  • Annotation reason is specific enough to explain the label in a presentation.
  • Detector error type is presentation-safe: false positive means correct span flagged; false negative means wrong/unsupported span missed.
Gold reference 
{
  "field": "reduction_amount",
  "fact_type": "currency_amount",
  "gold_value": 44600.65,
  "gold_display_value": "GBP 44,600.65",
  "tolerance": {
    "absolute": 1.0,
    "relative_percent": 0.5
  }
}
confirm_015 Pending human review

q_0098 - currency_amount - correct_key_fact

In September 2011, how much did cancellations and returns reduce gross positive revenue, and what was the final net revenue?

In September 2011, cancellations and returns reduced gross positive revenue by [[GBP 38,838.51]], resulting in a net revenue of GBP 101,759.68.
Gold answer
Cancellations and returns reduced gross positive revenue by GBP 38,838.51 (3.68% of gross), leaving net revenue of GBP 1,017,596.68.
Reason
Generated reduction amount matches GBP 38,838.51.
Why selected
detector_error_example, currency_amount_pattern
Detector role
simple false_positive via one_minus_min_top2_margin
Confirmation checklist
  • Highlighted span points to the intended generated claim.
  • Gold reference matches the deterministic gold answer/evidence.
  • Label is correct_key_fact only when the generated claim is supported; otherwise hallucinated_key_fact or unsupported_claim is justified.
  • Annotation reason is specific enough to explain the label in a presentation.
  • Detector error type is presentation-safe: false positive means correct span flagged; false negative means wrong/unsupported span missed.
Gold reference 
{
  "field": "reduction_amount",
  "fact_type": "currency_amount",
  "gold_value": 38838.51,
  "gold_display_value": "GBP 38,838.51",
  "tolerance": {
    "absolute": 1.0,
    "relative_percent": 0.5
  }
}