CoolFace
Datasetpublic

Parthsoni10/supplychain-agent-input-samples

Supply Chain Agent Input Samples Synthetic input samples for a five-agent supply-chain platform (aizenio/supplychain-ops). Each row is one internally consistent snapshot that satisfies the input contract of every agent skill — 67 columns covering demand forecasting, inventory monitoring, logistics/shipment evaluation, anomaly detection and strategic analysis. A single row can be fed to any agent without post-processing. Why it exists The agents needed realistic… See the full description on the dataset page: https://huggingface.co/datasets/Parthsoni10/supplychain-agent-input-samples.

sourceHugging Facemitupdated 10d agoView on Hugging Face
0likes54downloads
Dataset Card

Supply Chain Agent Input Samples

Synthetic input samples for a five-agent supply-chain platform (aizenio/supplychain-ops).

Each row is one internally consistent snapshot that satisfies the input contract of every agent skill — 67 columns covering demand forecasting, inventory monitoring, logistics/shipment evaluation, anomaly detection and strategic analysis. A single row can be fed to any agent without post-processing.

Why it exists

The agents needed realistic, reproducible inputs to be exercised against without standing up the live world simulation. This dataset is the recorded input side of that system, generated from the platform's own reference data and formulas so the values stay consistent with agent behaviour.

Contents

Column groupFields
Identitysample_id, tick, sim_date, scenario, applicable_skills
Productproduct_id, product_name, category, base_demand, unit_cost
Locationlocation, location_name, location_type, capacity
Demandexternal_factors, current_demand, predicted_demand, trend, spike_detected, spike_magnitude, historical_avg, std_deviation
Demand anomalydemand_z_score, demand_anomaly, demand_anomaly_severity
Inventorycurrent_stock, daily_demand, reorder_point, fill_rate, status, days_of_supply, needs_reorder
Reallocationfrom_location, to_location, quantity, reason
Logisticsshipment_id, origin, destination, distance_km, weather_condition, traffic_level, expected_hours, revised_hours, delay_hours, delay_reason, shipment_status, progress, departed_tick, current_tick, expected_duration_ticks
Anomalyinventory_anomaly_type, inventory_anomaly_severity, logistics_anomaly_type, logistics_anomaly_severity
Correlationanomalies_json
GOD agentinventory_data_json, demand_data_json, events_summary_json, alerts_json, event_history_json, risks_json, causes_json
Insightpredictions_json, recommendations_json, summary, confidence

Rows: 250,000 × 67 columns.

How it was generated

Values are derived from the platform's own reference data and formulas, not sampled arbitrarily:

  • —PRODUCTS, LOCATIONS, ROUTES and _CAPACITY from app/core/simulation.py — 8 SKUs × 8 locations, real coordinates used for haversine distances.
  • —Seasonal curve 1 + 0.25·sin(2π(doy−80)/365) and the external-factor weight map from demand_agent.py.
  • —Weather/traffic delay multipliers from logistics_agent.py (clear 1.0 … storm 1.8, low 1.0 … severe 2.0).
  • —reorder_point = daily_demand × 7 and fill-rate status thresholds (<0.20 critical, <0.40 low, >0.90 excess) from inventory_agent.py.
  • —z-score anomaly bands (>3 critical, >2 high, >1.5 medium) from anomaly_agent.py.

The sampling universe is 8 SKUs × 8 locations × 365 ticks = 23,360 unique combinations, so 250k rows covers every combination roughly ten times over.

Regenerate with:

bash
python scripts/generate_agent_input_samples.py 250000 --out docs/samples.csv
python scripts/build_local_sqlite.py --rows 250000

Usage

python
from datasets import load_dataset

ds = load_dataset("<hf-username>/supplychain-agent-input-samples", split="train")
row = ds[0]

Intended use and limits

  • —Intended: exercising, testing and benchmarking the agent network; query-heavy demos; as a template for supply-chain tabular schemas.
  • —Not intended: real operational planning. The data is synthetic and the network topology is a fixed 8×8 reference graph.
  • —Values reflect one platform's formulas, so statistical properties are internally consistent rather than drawn from real-world distributions.

License

MIT, matching the source repository.