xpertsystems/mfg007-sample

MFG-007 — Warehouse Operations Dataset (Sample)

A schema-identical preview of MFG-007, the XpertSystems.ai synthetic warehouse operations + WMS activity dataset for picking productivity ML, inventory accuracy modeling, labor optimization, perfect order rate analysis, dock-to-stock cycle time forecasting, and fulfillment efficiency research. The full product covers 50,000-100,000 records. This sample is HF-sized at 3,000 records.

Built by XpertSystems.ai — Synthetic Data Platform Contact pradeep@xpertsystems.ai · xpertsystems.ai License CC-BY-NC-4.0 (sample); commercial license available for the full product.

---

What MFG-007 does — completing the 7-SKU Manufacturing vertical

MFG-007 is the seventh Manufacturing & Industrial Systems SKU in the XpertSystems catalog, completing a 7-SKU vertical covering FIVE major business functions:

Function	SKUs	Buyer Audience
Reliability Engineering	MGG-001 + MFG-002 + MFG-003	CMRP, CMMS, reliability software
Quality Engineering	MFG-004	CQE/CSSBB, QMS, SPC software
Operations Management	MFG-005	MES, OEE software, TPM, Lean
Supply Chain Risk	MFG-006	SCRM platforms, procurement
Warehouse Operations	MFG-007	WMS vendors, 3PL providers, fulfillment robotics, e-commerce

Where MFG-006 captures upstream supply chain risk, MFG-007 captures downstream fulfillment operations — the final piece of the manufacturing-to-customer value chain. This is the data shape that flows into WMS (Warehouse Management Systems) platforms:

Buyer Persona	Use Case
Manhattan Active WM (NASDAQ:MANH, $14B+ market cap)	Pick productivity + inventory accuracy ML
Blue Yonder Luminate Warehouse Edge (Panasonic-owned $7.1B)	Slotting + labor optimization
SAP EWM (publicly traded SAP $200B+)	Extended warehouse management analytics
Oracle WMS Cloud ($200B+ Oracle)	Cloud WMS feature ML
Infor WMS (private $10B+)	WMS productivity benchmarking
Fishbowl, Cin7, ShipBob	SMB + e-commerce WMS analytics
Amazon Fulfillment Services + Shopify Fulfillment Network	Marketplace fulfillment ML
3PL Providers (Penske Logistics, Ryder, XPO, DHL Supply Chain, GXO, NFI)	3PL operational benchmarking
MHI Warehouse Robotics (AutoStore, Symbotic, Locus Robotics, 6 River, GreyOrange)	Robotics ROI + AMR ML training
Pick-to-Light / Voice (Honeywell Voice, Lucas Systems, Vocollect)	Pick method effectiveness ML
Inventory Optimization (RELEX, ToolsGroup, OMP, o9)	Demand-driven replenishment
Labor Management (Manhattan LMS, JDA WLM, MercuryGate)	Labor productivity ML
WERC (Warehousing Education and Research Council)	DC Measures Benchmark Studies
APICS CSCP / CPIM Training	Inventory management case-study data

This is the substrate WMS vendors, 3PL providers, fulfillment robotics companies, e-commerce platforms, MHI material handling equipment vendors, and warehouse research consultancies have been waiting for: a coherent warehouse-event dataset where picking × inventory × labor × throughput all interact with WERC Benchmark Studies / GS1 Global Standards / OSHA 1910 / MHI / APICS CPIM / NRF retail shrink / Frazelle 2002 World-Class Warehousing-grade calibration.

---

What's inside

Single cross-sectional dataframe, one row per warehouse activity event with joined picking + inventory + labor + throughput data.

Output	Rows (sample)	Columns	Size
mfg007_warehouse_data.csv	3,000	116	~2.3 MB

Schema provided in MFG_007_schema.json.

Module structure (116 columns total, 7 modules)

Module	Cols	Coverage
Event identity	17	event_id, warehouse_id, zone, aisle, bay, shift_id, dates + timestamps, event_type, warehouse_type (5), industry sector (10), WMS system (7), automation level, peak season, edge case type
Picking	23	order_id, line_id, SKU, pick_method (8), picker_id + experience, qty ordered/actual, accuracy (yes/no), error type (5), start/end times, duration, travel + pick + confirmation time, picks/units/lines per hour, travel distance, path efficiency, location type (5), shortpick flag, units, substitution
Inventory	22	snapshot_id, location_id, on-hand/available/reserved/in-transit/on-order units, days of supply, reorder point, safety stock, fill rate, accuracy, shrinkage units + %, putaway cycle time, replenishment trigger (6) + lead time, expiry date, FIFO compliance, slotting score, turn rate, dead stock days, overstock flag
Labor	22	labor_record_id, operator_id + role (8), shift type (7), scheduled/actual headcount, absenteeism, utilization, productive/indirect/idle/overtime hours, labor cost per unit + shift, training hours, safety incidents, near miss, ergonomic risk, fatigue index, task completion, pick rate vs standard, quality error rate
Throughput	29	throughput_record_id, measurement period, inbound/outbound units, orders shipped, lines/units picked, dock-to-stock hours, order cycle time, on-time shipment rate, order fill rate, perfect order rate, dock utilization, conveyor + sorter throughput, equipment downtime + availability, bottleneck zone (8) + severity (5), WMS transactions, RF scan accuracy, carrier on-time pickup, returns processing + restocking, value-added services, cubic utilization, slotting optimization flag, warehouse management score
Equipment & systems	3	conveyor + sorter throughput, WMS transaction volume

---

Calibration sources

Every distribution is anchored to named warehousing standards and benchmark studies. The headline anchors are WERC DC Measures Benchmark Studies, APICS CPIM / CSCP Body of Knowledge, and Frazelle 2002 World-Class Warehousing. Other anchors:

• WERC (Warehousing Education and Research Council) DC Measures Annual Benchmark Studies — pick accuracy, inventory accuracy, perfect order rate, dock-to-stock, cubic utilization.
• WERC Picking Productivity Benchmarks — UPH by pick method: manual 60-120, batch 150-250, voice 180-300, put-to-light 250-500, AS/RS 500-2000.
• GS1 Global Standards — barcode, RFID, EPCIS event standards driving RF scan accuracy benchmarks.
• OSHA 1910 Subpart D (Walking-Working Surfaces) + OSHA 1910 Subpart N (Materials Handling) — warehouse safety incident benchmarks.
• BLS Warehouse Industry Statistics — OSHA recordable incident rate ~4.8 per 100 FTEs annually for warehousing/storage NAICS 4931.
• MHI (Material Handling Industry) Annual State of the Industry — conveyor + sorter + AS/RS + AGV + AMR availability and adoption.
• APICS CPIM (Certified in Production and Inventory Management) — inventory turn rate, days of supply, ABC analysis fundamentals.
• APICS CSCP (Certified Supply Chain Professional) — perfect order rate, SCOR Model KPI framework.
• NRF (National Retail Federation) Annual Retail Security Survey — shrinkage % benchmarks across retail subsegments.
• Frazelle 2002 World-Class Warehousing & Material Handling — comprehensive warehouse productivity framework.
• Supply Chain Council SCOR Model — Perfect Order Rate definition (on-time + complete + undamaged + correct documentation).
• Tompkins Associates Warehouse Design Studies — slotting optimization, cubic utilization, bottleneck identification.
• ISO 28000 Supply Chain Security Management — facility security, inventory integrity.
• ISA-95 / IEC 62264 Enterprise-Control Integration — WMS to ERP/MES data integration.

---

Validation scorecard

The wrapper ships a 10-metric WERC/GS1/OSHA/MHI/APICS-anchored scorecard (validation_scorecard.json) that re-scores the dataset on every generation. Default seed 42 result:

ID	Metric	Target	Observed	Source
M01	Pick Accuracy (FLOOR ≥96%)	≥96%	98.67%	WERC DC Measures
M02	Inventory Accuracy % (FLOOR ≥95%)	≥95%	98.17%	WERC + APICS CPIM
M03	Shrinkage % (CEILING ≤2.5%)	≤2.5%	0.27%	NRF Annual Retail Security Survey
M04	Perfect Order Rate % (FLOOR ≥85%)	≥85%	94.70%	WERC + SCOR Perfect Order
M05	Picks Per Hour	30–230	133.54	WERC Picking Benchmarks
M06	Safety Incidents/Shift (CEILING ≤0.3)	≤0.3	0.066	OSHA 1910 + BLS NAICS 4931
M07	Order Fill Rate % (FLOOR ≥90%)	≥90%	97.47%	WERC + SCOR
M08	Dock-to-Stock Hours (CEILING ≤10)	≤10	6.10	WERC DC Measures
M09	Cubic Utilization %	66–90%	80.17	WERC + MHI
M10	Equipment Availability % (FLOOR ≥92%)	≥92%	95.99	MHI Material Handling Reliability

Grade: A+ (100/100). Verified across seeds 42, 7, 123, 2024, 99, 1.

Standout calibration depth — directly matches WERC benchmark ranges:

• M01 Pick accuracy 98.67% — within WERC typical 98-99% range
• M02 Inventory accuracy 98.17% — within WERC typical 97-99% range
• M04 Perfect Order Rate 94.70% — at WERC world-class 90-95% threshold 🎯
• M07 Order Fill Rate 97.47% — within WERC typical 95-98% range
• M08 Dock-to-Stock 6.10 hrs — within WERC typical 4-6 hrs range
• M09 Cubic Utilization 80.17% — within WERC 70-85% range 🎯
• M10 Equipment Availability 95.99% — within MHI 92-98% range 🎯

This dataset is directly benchmarkable against WERC DC Measures published reports — meaningful for the WERC member community of 1,000+ warehouse operations professionals.

---

Suggested use cases

• Pick productivity ML — picker experience + method + location + fatigue × UPH/UPH prediction.
• Pick accuracy classification — pick features × accuracy yes/no for error-prediction ML.
• Inventory accuracy modeling — replenishment + slotting + WMS features × inventory_accuracy_pct regression.
• Perfect order rate prediction — composite OTIF + complete + undamaged + documentation accuracy.
• Dock-to-stock cycle time forecasting — inbound + putaway features × cycle time prediction.
• Labor optimization — fatigue + experience + shift type × pick rate vs standard.
• Safety incident prediction — ergonomic risk + fatigue + shift × safety_incidents classification.
• Bottleneck identification — throughput + WMS features × bottleneck_zone classification.
• Slotting optimization — pick path efficiency + cubic utilization × slotting_score regression.
• Returns processing efficiency — returns flow × restocking rate.
• Edge case detection — labor mass absenteeism / robotics failure / cold chain breach / WMS migration / inventory integrity failure classification.
• Warehouse benchmarking — industry sector × warehouse type × KPI comparison for WERC-style benchmarking.
• WMS system effectiveness — 7 WMS systems × performance metrics for vendor comparison.
• Pick method effectiveness — 8 pick methods (discrete/batch/zone/ wave/cluster/voice/put-to-light/RF) × UPH + accuracy.

---

Loading

from datasets import load_dataset

ds = load_dataset(
    "xpertsystems/mfg007-sample",
    data_files="mfg007_warehouse_data.csv",
    split="train",
)

Or with pandas directly:

import pandas as pd
from huggingface_hub import hf_hub_download

path = hf_hub_download(
    repo_id="xpertsystems/mfg007-sample",
    filename="mfg007_warehouse_data.csv",
    repo_type="dataset",
)
df = pd.read_csv(path)

# Pick productivity by method (WERC benchmarks)
by_method = df.groupby("pick_method").agg(
    uph=("picks_per_hour", "mean"),
    accuracy=("pick_accuracy", lambda s: (s == "yes").mean()),
).round(3)
print(by_method.sort_values("uph", ascending=False))

# Perfect order rate by warehouse type
print(df.groupby("warehouse_type")["perfect_order_rate_pct"].mean().sort_values())

# WMS system effectiveness comparison
wms_perf = df.groupby("wms_system").agg(
    inv_accuracy=("inventory_accuracy_pct", "mean"),
    perfect_order=("perfect_order_rate_pct", "mean"),
    dock_to_stock=("dock_to_stock_hours", "mean"),
).round(2)
print(wms_perf)

The dataset ships with MFG_007_schema.json providing per-column dtypes for pipeline integration:

import json
schema = json.load(open("MFG_007_schema.json"))

This dataset is cross-sectional with event-level granularity — one row per warehouse activity event. For warehouse-level aggregation, group by warehouse_id. For SKU-level, group by sku_id.

---

Schema highlights

Event identity — event_id, warehouse_id, zone_id, aisle_id, bay_id, shift_id, event_date, event_timestamp, shift_start_time, shift_end_time, event_type, warehouse_type ∈ {fulfillment_center, distribution_center, cross_dock, cold_storage, dark_store}, industry_sector (10 sectors), wms_system ∈ {Manhattan, Blue_Yonder, SAP_EWM, Oracle_WMS, Infor, Fishbowl, legacy}, automation_level ∈ {manual, semi_automated, highly_automated, lights_out, mixed}, is_peak_season, edge_case_type.

Picking — order_id, order_line_id, sku_id, sku_category, pick_method ∈ {discrete, batch, zone, wave, cluster, voice_directed, put_to_light, RF_scan}, picker_id, picker_experience_months, pick_quantity_ordered, pick_quantity_actual, pick_accuracy (yes/no), pick_error_type ∈ {wrong_sku, wrong_qty, wrong_location, damaged_pick, missing_item, NaN}, pick_start_time, pick_end_time, pick_duration_seconds, travel_time_seconds, pick_time_seconds, confirmation_time_seconds, picks_per_hour, units_per_hour, lines_per_hour, travel_distance_meters, pick_path_efficiency, pick_location_type ∈ {pallet_rack, carton_flow, shelving, floor_slot, mezzanine, pick_tower, carousel, conveyor_feed}, shortpick_flag, shortpick_units, substitution_applied.

Inventory — inventory_snapshot_id, location_id, inventory_on_hand_units, inventory_available_units, inventory_reserved_units, inventory_in_transit_units, inventory_on_order_units, days_of_supply_onhand, reorder_point_units, safety_stock_units, fill_rate_pct, inventory_accuracy_pct, shrinkage_units, shrinkage_pct, putaway_cycle_time_mins, replenishment_trigger ∈ {min_max, MRP, demand_driven, auto_reorder, manual, none}, replenishment_lead_time_hours, expiry_date, fifo_compliance, slotting_score, inventory_turn_rate, dead_stock_days, overstock_flag.

Labor — labor_record_id, operator_id, operator_role ∈ {picker, packer, receiver, putaway, replenisher, forklift_operator, returns_processor, supervisor}, shift_type ∈ {day, swing, night, weekend, overtime, casual, agency}, headcount_scheduled, headcount_actual, absenteeism_pct, labor_utilization_pct, productive_hours, indirect_hours, idle_hours, overtime_hours, labor_cost_per_unit_usd, labor_cost_total_shift_usd, training_hours_mtd, safety_incidents, near_miss_events, ergonomic_risk_score, operator_fatigue_index, task_completion_rate, pick_rate_vs_standard, quality_error_rate_pct.

Throughput — throughput_record_id, measurement_period, inbound_units_received, outbound_units_shipped, orders_shipped, lines_picked, units_picked, dock_to_stock_hours, order_cycle_time_hours, on_time_shipment_rate_pct, order_fill_rate_pct, perfect_order_rate_pct, dock_utilization_pct, conveyor_throughput_units_hr, sorter_throughput_units_hr, equipment_downtime_minutes, equipment_availability_pct, bottleneck_zone ∈ {none, picking, packing, putaway, replenishment, shipping, receiving, returns}, bottleneck_severity ∈ {none, low, medium, high, critical}, wms_transaction_volume, rf_scan_accuracy_pct, carrier_on_time_pickup_pct, returns_processing_time_mins, returns_restocking_rate_pct, value_added_services_units, cubic_utilization_pct, slotting_optimization_flag, warehouse_management_score.

---

Calibration notes & limitations

In the spirit of honest synthetic data, a few things buyers of the sample should know:

1. Shrinkage 0.27% is below NRF retail benchmark of 1.5-1.8%, but this reflects warehouse-only shrinkage (no retail floor exposure to customer theft). The major NRF shrinkage drivers — customer theft, dishonest employees, administrative errors — are reduced in pure-DC environments. The 0.27% is realistic for B2B fulfillment and DC operations.

2. Automation level is skewed semi_automated 99.7% at this sample size. The generator's automation_level parameter defaults to "mixed" but in practice produces predominantly semi-automated records. For automation-tier-specific analysis (manual vs lights_out vs highly_automated), the full product supports explicit automation tier filtering via --automation_level.

3. Days of supply 80.65 is high (typical 30-60 days). Reflects sector mix where pharma and automotive carry longer inventory.

4. Operator fatigue index 8.65 (of 10) is high — reflects the generator's emphasis on physically-demanding warehouse work; the pick_rate_vs_standard 0.69 (below 1.0) shows fatigue impact on productivity.

5. Overstock rate 61.5% is high — reflects safety-stock-heavy replenishment strategy. For lean-inventory modeling, the full product supports demand-driven MRP-only configurations.

6. Pick error type column is 98.67% NaN because errors only populate when pick_accuracy == "no". When picks are accurate, no error type applies. For error-classification ML, filter to inaccurate picks only (~1.3% of records).

7. 3% of records carry edge_case_type labels including labor mass absenteeism (0.67%), peak season surge (0.67%), robotics failure (0.33%), equipment cascade failure (0.30%), cold chain breach (0.30%), inventory record integrity failure (0.27%), WMS migration cutover (0.23%). These are valuable for edge-case classification ML and operational risk modeling.

8. Peak season events 19% of records — reflects realistic Q4 + back-to-school + Mother's Day patterns. For peak-season-specific modeling, filter is_peak_season == True.

9. 8 operator roles balanced ~10% each — generator chooses to distribute evenly across roles rather than reflecting actual warehouse role mix (typically picker-heavy 30-40%). For role-mix- realistic modeling, the full product supports weighted role distributions.

10. Deterministic seeding. Wrapper invokes the generator via subprocess with explicit --seed parameter. Seed sweep verifies Grade A+ across {42, 7, 123, 2024, 99, 1}.

---

Commercial / full product

The full MFG-007 product covers 50,000-100,000 warehouse activity records with configurable --labor_profile (high_performance / average / challenged / mixed), --automation_level (manual / semi_automated / highly_automated / lights_out / mixed) for automation-tier-specific modeling, --order_profile (B2B_bulk / B2C_singles / omnichannel / mixed), expanded pick method effectiveness scenarios, refined slotting

• wave planning logic, pre-built feature engineering for pick productivity ML (lag features, rolling-7-day pick rates, fatigue recovery curves), demand-driven replenishment scenarios (Demand-Driven MRP), peak-season stress-test cohorts, robotics integration scenarios (AutoStore + Symbotic + Locus AMR), cold chain compliance subsets (pharma + grocery), and value-added services (kitting + assembly + gift wrap + ticketing). Available under commercial license — contact pradeep@xpertsystems.ai.

XpertSystems.ai also publishes synthetic data products across Oil & Gas (17 SKUs), Healthcare/Neurology (10 SKUs), and Manufacturing (7 SKUs — complete coverage across reliability + quality + operations + supply chain + warehouse):

• MGG-001: Factory Sensor Dataset (IIoT sensor streams)
• MFG-002: Machine Failure Event Records (CMMS, ISO 14224)
• MFG-003: Predictive Maintenance Dataset (RUL ML training)
• MFG-004: Quality Control Dataset (SPC, MSA, 6 Sigma)
• MFG-005: Manufacturing Line Performance (OEE, TPM, Lean)
• MFG-006: Supply Chain Disruption Dataset (SCRM, bullwhip)
• MFG-007: Warehouse Operations Dataset (WMS, picking, perfect order) — this SKU

Catalog: huggingface.co/xpertsystems.