▸case-01 We analyzed a credit application and rejected it due to a high debt-to-income ratio. We need to perform a boundary analysis along the debt-to-income dimension to see the smallest shift in this ratio that would reverse our rejection decision. Please run the evaluation and provide the minimal required change, how far that shift is on a 0.0 to 1.0 scale, your confidence level, and the sequence of steps taken during the search. | fail→fail | 12,575 | 32,636 | +160% | 1 | 1 | 0% | 2,256 | 3,356 | +49% | 0 | 0 | — |
▸case-02 In our safety assessment of an autonomous driving model, the current verdict is 'fail' under heavy rain conditions. Run a perturbation search along the rainfall intensity axis to identify the boundary where the verdict changes to 'pass'. I need a report showing the threshold change that flips the outcome, a normalized distance metric from actual conditions, your confidence score, and the step-by-step search trail. | fail→fail | 31,365 | 18,272 | -42% | 1 | 1 | 0% | 1,557 | 3,676 | +136% | 0 | 0 | — |
▸case-03 We have a legal document classifier that categorized a draft agreement as 'high risk'. I want to isolate the clause severity dimension and find the smallest edit to the document that alters the risk classification to 'low risk'. Return the exact minimal modification that alters the verdict, the relative distance metric from the original draft, your confidence in the result, and the trail of search steps. | fail→fail | 26,930 | 9,497 | -65% | 1 | 1 | 0% | 1,861 | 1,842 | -1% | 0 | 0 | — |
▸case-04 A patient triage model flagged an emergency intake record as 'urgent' based on recorded heart rate. Determine the exact heart rate threshold where the triage decision flips from 'urgent' to 'non-urgent'. Evaluate along the heart rate dimension and provide the output containing the threshold change, distance from baseline (0.0 to 1.0), confidence score, and search steps. | fail→fail | 11,357 | 16,939 | +49% | 1 | 1 | 0% | 2,192 | 3,598 | +64% | 0 | 0 | — |
▸case-05 An e-commerce review classifier rated a customer review as 'negative'. Perform an iterative perturbation along the word sentiment polarity dimension to find the boundary where the verdict turns 'positive'. Deliver the results including the minimal flipping modification, distance metric, confidence rating, and the iteration path. | fail→fail | 11,171 | 15,615 | +40% | 1 | 1 | 0% | 1,998 | 3,221 | +61% | 0 | 0 | — |
▸case-06 An enterprise mail gateway classified an outbound marketing email as 'spam'. Test the keyword density dimension using binary search to discover the exact point where it gets reclassified as 'ham'. Return the threshold modification, normalized distance, confidence score, and search step history. | fail→fail | 27,091 | 14,564 | -46% | 1 | 1 | 0% | 1,922 | 3,264 | +70% | 0 | 0 | — |
▸case-07 A wire transfer monitor flagged a $15,000 international transfer as 'suspect'. Determine the minimal transaction amount adjustment needed to flip the classification to 'cleared'. Provide the exact change needed, the 0.0 to 1.0 distance score, confidence level, and recorded search trail. | fail→fail | 15,235 | 21,953 | +44% | 1 | 1 | 0% | 743 | 3,513 | +373% | 0 | 0 | — |
▸case-08 A content moderation model flagged a user comment as 'toxic'. Run a perturbation search along the toxicity score dimension to locate the boundary where the post becomes 'compliant'. Submit the resulting flip modification, distance metric from origin, confidence, and search trail. | fail→fail | 12,216 | 15,121 | +24% | 1 | 1 | 0% | 2,118 | 2,720 | +28% | 0 | 0 | — |
▸case-09 An automated underwriting system denied a mortgage application due to applicant credit score. Conduct a binary search along the credit score parameter to locate the exact threshold where the decision flips to 'approved'. Provide the minimal score delta, distance score, confidence, and search history. | fail→fail | 25,254 | 16,001 | -37% | 1 | 1 | 0% | 1,809 | 3,413 | +89% | 0 | 0 | — |
▸case-10 A recruiting filter marked a candidate profile as 'rejected' for a Senior Engineer role. Perturb the years-of-experience attribute to find the tipping point where the decision becomes 'shortlisted'. Provide the minimal change, 0.0-1.0 distance, confidence, and search steps. | fail→fail | 9,871 | 15,284 | +55% | 1 | 1 | 0% | 1,872 | 3,377 | +80% | 0 | 0 | — |
▸case-11 An industrial defect vision model categorized a silicon wafer image as 'defective'. Find the exact illumination brightness adjustment needed to flip the output to 'pass'. Return the perturbation value, normalized distance, confidence level, and search trail. | fail→fail | 22,260 | 12,759 | -43% | 1 | 1 | 0% | 4,426 | 701 | -84% | 0 | 0 | — |
▸case-12 A network intrusion detection model flagged a traffic burst as 'exfiltration attack'. Conduct a binary search along the packet payload size dimension to find where the verdict flips to 'benign'. Report the minimal size shift, distance metric, confidence rating, and search steps. | fail→fail | 35,336 | 29,272 | -17% | 1 | 1 | 0% | 3,157 | 2,761 | -13% | 0 | 0 | — |
▸case-13 A customer success dashboard flagged a subscription account as 'high risk of churn'. Search along the weekly active login count dimension to find the tipping point where the prediction flips to 'retained'. Provide the required change, 0.0 to 1.0 distance, confidence score, and search steps. | fail→fail | 10,839 | 11,923 | +10% | 1 | 1 | 0% | 2,050 | 2,381 | +16% | 0 | 0 | — |
▸case-14 An automated valuation model evaluated a residential property as 'overpriced'. Run a binary perturbation search along the living space square footage dimension to find where the model flips to 'fairly priced'. Return the delta, distance score, confidence, and step log. | fail→fail | 29,514 | 14,306 | -52% | 1 | 1 | 0% | 3,496 | 3,404 | -3% | 0 | 0 | — |
▸case-15 An insurance claims analyzer categorized an auto collision claim as 'audit required'. Find the minimal claim amount perturbation along the estimated repair cost dimension that flips the verdict to 'auto approve'. Return the flip point, distance (0.0-1.0), confidence, and search path. | fail→fail | 16,206 | 13,136 | -19% | 1 | 1 | 0% | 3,385 | 3,018 | -11% | 0 | 0 | — |
▸case-16 An oncology imaging classifier flagged an MRI scan section as 'malignant'. Search along the lesion diameter dimension using binary search to find where the prediction flips to 'benign'. Provide the minimal diameter change, normalized distance, confidence, and search path. | fail→fail | 36,036 | 19,069 | -47% | 1 | 1 | 0% | 2,947 | 4,248 | +44% | 0 | 0 | — |
▸case-22 We have a multivariate random forest model predicting customer lifetime value. Generate a static correlation matrix showing the Pearson correlation coefficients across all 12 input features simultaneously. | pass→pass | 23,788 | 15,547 | -35% | 1 | 1 | 0% | 5,444 | 3,495 | -36% | 0 | 0 | — |
▸case-17 A static analysis tool failed a pull request build for 'excessive complexity'. Run a binary search on the method cyclomatic complexity dimension to determine the exact threshold where the build status flips to 'pass'. Output the required decrease, distance from current value, confidence, and search steps. | fail→fail | 11,153 | 30,671 | +175% | 1 | 1 | 0% | 2,189 | 2,580 | +18% | 0 | 0 | — |
▸case-18 An ad delivery engine categorized an ad creative as 'low engagement'. Find the boundary along the historical CTR dimension where the classification flips to 'high engagement'. Report the flip point, relative distance (0.0 to 1.0), confidence, and search path. | fail→fail | 14,758 | 15,066 | +2% | 1 | 1 | 0% | 2,822 | 3,405 | +21% | 0 | 0 | — |
▸case-19 A climate risk assessment model rated a coastal facility as 'extreme flood hazard'. Perform a perturbation search along the elevation above sea level dimension to find the threshold where the rating flips to 'moderate hazard'. Return the elevation delta, distance metric, confidence, and search path. | fail→fail | 14,030 | 37,724 | +169% | 1 | 1 | 0% | 2,604 | 2,316 | -11% | 0 | 0 | — |
▸case-20 An e-commerce pricing model flagged a candidate retail price of $89.99 for a product as 'margin negative'. Conduct a binary perturbation search along the unit price dimension to identify where the classification flips to 'margin positive'. Provide the flip point, normalized distance (0.0 to 1.0), confidence, and search path. | fail→fail | 18,402 | 24,253 | +32% | 1 | 1 | 0% | 4,047 | 5,375 | +33% | 0 | 0 | — |
▸case-21 We have a customer support ticket from user #8492 reading 'My order was delayed three days and I want a refund immediately'. Classify whether this ticket is 'urgent' or 'standard' priority based on standard ticketing guidelines. | pass→pass | 6,507 | 6,041 | -7% | 1 | 1 | 0% | 931 | 1,125 | +21% | 0 | 0 | — |
▸case-23 Please rewrite this error message 'Database query failed due to connection timeout' to be more user-friendly for non-technical end users in our web application. | pass→pass | 26,345 | 5,376 | -80% | 1 | 1 | 0% | 1,189 | 1,109 | -7% | 0 | 0 | — |