Historical transactions / simulated inventory As of 2011-11-30sl-61f664e09e9c9098

Public data source

The source citation, limits, and artifact status stay visible.

UCI Online Retail II · Dataset 502

Chen, D. (2012). Online Retail II. UCI Machine Learning Repository. https://doi.org/10.24432/C5CG6D. The repository reports 1,067,371 transaction rows from 2009-12-01 through 2011-12-09. This workbench shows an illustrative 14-SKU slice of it.

Open the official dataset page

License: Creative Commons Attribution 4.0 International. The SHA-256 recorded alongside this demo is a pinned reference value for the official UCI archive. Nothing in this repository downloads or hashes the source, so it has not been verified here — treat it as a value to check against, not as proof of provenance. The UI fixture is separate from the pending full transformation and backtest run.

Interface states

What the workbench shows while it is working, when a filter matches nothing, and when a view fails. All three are wired into the live surfaces; they are reproduced here so the handling is visible without breaking anything.

Recalculating the planShown while an edited scenario is applied to recommendations, risk, and budget allocation.

No products match these filtersShown when the product filters exclude every stock code in the subset.

This view could not be renderedShown by the error boundary that wraps the active view, with the scenario left untouched.

Cleaning contract

Each invalid row receives a reason code before aggregation.

Cancellation markerSeparate invoices that begin with the documented cancellation marker.
Returns and negative unitsKeep return events apart from gross positive demand.
Price validityQuarantine zero or negative unit prices with a reason code.
Product scopeExclude postage, adjustments, manual entries, and other non-merchandise codes.
IdentifiersRemove customer identifiers from every deployment artifact.
Temporal leakageChoose eligible products and model features from training windows only.

Core policy formulas

Inputs remain visible; no safety factor is hidden in a second term.

inventory_position = on_hand + on_order − backorders − allocatedUnits available or expected after commitments.
recommended_raw = max(0, order_up_to_level − inventory_position)Unrounded policy quantity before supplier rules.
recommended_quantity = round_to_case_pack(max(MOQ, recommended_raw))MOQ applies only when the raw quantity is positive.
unit_fill_rate = fulfilled_units / requested_unitsReturns and cancellations do not reduce the demand denominator.

Provenance key

Every displayed field uses one of three source types.

ObservedPublic transaction fieldStock code, description, timestamp, quantity, unit price, country.
DerivedCalculated from observed dataDemand, forecast, variation, segment, risk, and policy thresholds.
AssumptionSimulated or configured inputInventory, supplier, lead time, costs, MOQ, budget, and storage.

KPI definitions

Formula, unit, grain, window, and edge case are documented.

Unit fill rate: fulfilled units / requested units% · Demand event, then portfolio · Returns do not reduce requested demand; zero requested units return no value.
Cycle service level: eligible replenishment cycles without a stockout / eligible cycles% · SKU replenishment cycle · No eligible cycles return no value.
Stockout events: Count of transitions from available stock to unfilled demandevents · SKU-day · Consecutive stockout days count as one event until availability returns.
Stockout days: Count of active days with unfilled demand or backordersdays · SKU-day · A day with zero demand and an old backorder still counts in backorder mode.
Backordered or lost units: sum(requested units - fulfilled units)units · Demand event · The scenario states whether unfilled units carry forward or become lost sales.
Average on hand: sum(end-of-day on hand) / active daysunits · SKU-day · On hand cannot fall below zero.
Inventory position: on hand + on order - backorders - allocatedunits · SKU snapshot · The value can be negative when commitments exceed supply.
Unit turnover: fulfilled units / average on-hand unitsturns · SKU, then portfolio · Zero average on hand returns no value.

Case-study limits

These limits must travel with every recommendation.

The public source ends in 2011, so the default plan is historical.
Inventory, supplier, cost, budget, and storage values are simulated.
The current UI uses a representative fixture until the official full-source pipeline is run.
No recommendation submits a purchase order or claims measured business impact.