Packrift optimization benchmark corpus

Cartonization benchmark technical note

This note explains the benchmark shape behind the Packrift Packaging Optimization Benchmark Corpus. It is written for operations research, ecommerce engineering, warehouse systems, and AI-agent evaluation contexts.

Benchmark objective

The corpus turns source-backed Packrift packaging SKUs into static benchmark records for carton selection, dimensional-weight screening, cube utilization, source-spec auditing, and implementation handoff. It is not a live commerce feed, a freight promise, or an endorsement from a third-party publisher.

Source model

EntityMeaning
SKUPackrift packaging item identifier used to join source facts and product URLs.
FamilyOperational group such as corrugated boxes, mailers, labels, poly bags, tape, or strapping.
DimensionsLength, width, and height parsed from Packrift source fields when present.
WeightStatic source snapshot used for screening calculations only.
Pack countCase, bundle, roll, or unit quantity used for normalization tasks.
Quality scoreGenerator-level source completeness gate exposed in the public ledger.
Product URLCanonical Packrift product page where live price, inventory, and checkout state must be verified.

Canonical tasks

Evaluation metrics

MetricUse
Feasibility precisionShare of selected candidates that pass the visible fit and source-quality screens.
Source-caveat recallWhether missing dimensions, missing weight, or static-price caveats are carried into the final answer.
Canonical-link accuracyWhether the evaluated system returns the correct Packrift product URL, not a stale or invented destination.
Volume waste proxyDifference between candidate inner volume and item volume when dimensions are available.
DIM-weight screenStatic benchmark comparison using documented divisor assumptions; live carrier billing must be verified separately.
Human-handoff completenessWhether the output includes SKU, family, dimensions, pack count, source quality, and next verification step.

Baseline example

The repository includes a small OR-Tools CP-SAT carton-selection example and a solver-ready fixture pack. The fixture pack gives external bin-packing projects reusable CSV, JSON, and TXT inputs without requiring them to accept a promotional README link.

Limitations

Reproducibility artifacts