COM-E2 · Synthetic source

REFERENCE / METHODOLOGY

A transparent decision model.

What FoodLoop knows, predicts, assumes and does not do.

01 — Original data

Seven immutable Refruiting source CSVs are synthetic challenge inputs. 45 offices, 900 delivery and 900 occupancy records, 30 directed zone edges, five fruit types, three vehicles, four reference strategies.

Not real-time; not personal inventory; no retail GTIN in source.

02 — Forecast cutoff

Predict from only occupancy observations strictly before the selected source date, with available-before-cutoff expected staff where flagged. Weight available expected staff at 65%, historical weekday/rolling baseline at 35%.

No target-day observed staff or consumption enters the prediction.

03 — Order minima

Predicted staff × office expected fruit kg/employee × operator safety factor. When below office minimum, deliver minimum or explicitly defer. No below-minimum delivery is silently scheduled.

Quantity precision: 3 decimals (kg).

04 — Zone routing

Optimize order among zones using directed distances. Vehicle load is a hard upper bound. Depot return is unavailable in the supplied matrix and reported infeasible when requested.

No office-level GPS route, actual punctuality or confirmed access slots.

05 — Inventory integrity

Every batch stores its own operator-confirmed expiry type and date. FEFO selects earliest date. Consumption/waste/donation are journaled, and negative stock is prevented by immediate SQLite transactions.

Use-by is a safety distinction; best-before is a quality date.

06 — Donation boundary

Offers and reservations live in SQLite. Transfers are explicit simulations. Expired use-by batches cannot be reserved or transferred. No charity integrations or actual dispatch are claimed.

Supplier stock defaults to UNKNOWN absent a verified report.