POLARIS / INVOICE EXCEPTION INVESTIGATION Run: uk-wholesale-1015-153431 SCOPE Retrospective analysis of a public dataset; historical replay, not a customer deployment. Period: H1 2011 (2011-01-01 to 2011-06-30). Invoice date in H1 2011, unit price > 0 and quantity > 0. Initial flag: Unit price deviates from the SKU’s most common unit price in the same month by more than max(£0.01, 2%). Classification: Heuristic attribution for this exercise, not a confirmed price list or independently validated error labels. AGGREGATE RESULT Valid invoice lines: 174,619 Initially flagged outside tolerance: 43,109 Matched to one of three heuristic patterns: 41,958 Unexplained: 1,151 CUSTOMER-TYPE PRICING / 33,849 lines These lines match a higher-price pattern associated with missing customer IDs. Comparing those records against one monthly common price creates apparent exceptions. Next step: Confirm whether a separate walk-in or customer-type price list applies, then use that price list as the comparison reference. Source: report:Management Conclusions, item 1; report:Hypothesis Tree and Findings, H3 VOLUME PRICING / 7,721 lines These lines follow a lower-price pattern at larger order quantities, consistent with quantity tiers. Next step: Confirm the approved quantity tiers and compare each invoice against the relevant tier. Source: report:Management Conclusions, item 1; report:Hypothesis Tree and Findings, H2 PRICE CHANGES / 388 lines These lines match price changes between adjacent months rather than a single stable monthly reference. Next step: Check the effective date of the price change and the price that should have applied to the invoice. Source: report:Management Conclusions, item 1; report:Hypothesis Tree and Findings, H4 NEEDS INVESTIGATION / 1,151 lines These lines do not match the three identified patterns. The analysis does not establish whether they are mistakes or negotiated prices. Next step: Review the underlying invoice, agreement and any override approval before deciding whether a correction is needed. Source: report:Management Conclusions, items 1 and 5; report:Evidence Grading Detail METHOD 1. Recompute the 174,619 valid invoice lines for January–June 2011. 2. Use the most common unit price for each SKU in the same month as the initial reference. 3. 43,989 lines differ exactly from that reference; the displayed review set is the narrower 43,109 lines outside max(£0.01, 2%) tolerance. 4. Assign the 43,109 flagged lines to customer-type, volume, price-change or unexplained categories using the exercise’s heuristic rules. 5. The first three categories sum to 41,958 pattern-matched lines. The remaining 1,151 lines need further investigation. 6. Treat this as a historical reclassification of the review set. It does not measure deployed workload reduction, hours saved, recovered revenue or corrected errors. EXPLANATION CHECKS Pilot: PASS. Expanded: PASS. Both checks require a valid category output: tier, customer_type, price_change or unexplained. Passing these checks does not validate the accuracy of the assigned labels. Agreement with the host rule was report-only pending a 50-row human spot check reaching 90%; no completed human validation is recorded in the provided run. SEPARATE QUOTE PROTOTYPE Pilot: PASS — 500 held-out lines. Expanded: FAIL — 3,000 held-out lines. Failed overall_gain: Required hit-rate improvement of at least 5 percentage points over the strongest trivial baseline. Failed unseen_family_ape: Required median absolute percentage error on eligible held-out product-family descriptions no greater than 0.8 times the noun median-price baseline. The separate suggested-price prototype did not meet expanded acceptance criteria and was not recommended for use. The website example shows invoice exception investigation. It does not present the failed quote prototype as a successful production pricing system. FULL RUN STATUS Display state: delivered. Overall delivery gate passed: false. Research closed: false. A run snapshot was delivered. The investigation remained open (requires_input) and the overall delivery gate did not pass. report_exists: PASS required_sections: PASS evidence_tags_present: PASS citations_resolve: PASS questions_file: PASS research_closed: FAIL privacy_masked: PASS Open questions: - Whether formal customer-type price lists and quantity tiers exist. - Whether unexplained deviations are errors or negotiated prices. - Who owns price lists and override approvals. DATASET ATTRIBUTION Chen, D. (2012). Online Retail II. UCI Machine Learning Repository. https://doi.org/10.24432/C5CG6D License: CC BY 4.0 — https://creativecommons.org/licenses/by/4.0/ Analysis: Analysis of a valid-line subset covering January–June 2011; aggregate results and heuristic categories prepared by Polaris. SOURCE PROVENANCE run: polaris-demo/static/run-review/uk-wholesale-1015-153431.json SHA-256: 001c82999473423201a459fc34896a266912e52ad08eef267907d719b96f5d29 Source location: /charts/0/points/0; /sections; /gate; /frozen_checks; /frozen_criteria report: polaris-demo/runs/uk-wholesale-1015-153431/translations/artifacts__report.md.en.md SHA-256: 5ea35a4c9e3a1d9e15c64752ff9e72c1396a35a2f609cfcb445fca9d4c1fd9f4 Source location: Management Conclusions; Reproducible Baseline; Evidence Grading Detail; Restructuring and Improvement Plan READING THIS RESULT - Public-dataset retrospective analysis, not a measured customer deployment. - Pattern matches are heuristic explanations, not validated labels or confirmed pricing errors. - No production changes were deployed and no automatic invoice correction is shown. - No time-saving, recovered-revenue or realized workload-reduction claim is made. - The separate quote prototype failed expanded validation; the full run remained open pending human input.