---
dataset_info:
  features:
  - name: shipment_id
    dtype: string
  - name: route
    dtype: string
  - name: carrier
    dtype: string
  - name: origin
    dtype: string
  - name: destination
    dtype: string
  - name: departure_time
    dtype: timestamp[ns]
  - name: arrival_time
    dtype: timestamp[ns]
  - name: delay_minutes
    dtype: int64
  - name: damage_flag
    dtype: int64
  - name: temperature_deviation
    dtype: float64
  - name: humidity_deviation
    dtype: float64
  - name: fuel_consumption
    dtype: float64
  - name: quality_score
    dtype: float64
  splits:
  - name: train
    num_bytes: 314282
    num_examples: 2000
  download_size: 98296
  dataset_size: 314282
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
---

# Brazil–Argentina Logistics Daily Transport Quality Dataset

> **Project:** `project_20260813_014231_9dbe9212`
> **Operator:** São Paulo SRE / Quality Improvement Team (China–Brazil joint-venture logistics company)
> **Operational date:** 2026-08-13
> **Owner account:** `test_user_9dbe9212@example.com`
> **Daily snapshot #:** day 1 of the quality-improvement programme

This dataset is the **daily transport-quality snapshot** for the South American logistics network. It records
individual truck shipments collected from the main logistics hubs of Brazil, Argentina, Uruguay and Paraguay
(São Paulo, Rio de Janeiro, Curitiba, Porto Alegre, Santos, Campinas, Buenos Aires, Rosario, Córdoba,
Montevideo, Asunción) and is used to drive the company's continuous **quality improvement program (QIP)** —
full traceability, reproducibility and sharing of every shipment's quality outcome.

## 1. Data source & collection

- **Source:** GPS telematic boxes on the fleet (truck trajectories), on-board temperature/humidity loggers,
  warehouse handover records, border-crossing toll/queue telemetry and fuel-card refuelling events.
- **Collection:** ingested from the fleet data-lake each night (00:30 BRT) and consolidated into a single daily
  snapshot. Each row = one finished shipment.
- **Recording method:** the fields are a *simulated but statistically calibrated* reconstruction of the real
  operational telemetry, designed to faithfully reproduce the failure modes observed on the South American
  network (border congestion, cold-chain drift on Argentine agri-produce, wet-season damage). The generator is
  seeded (`seed=42`) and fully reproducible.

## 2. Update frequency

- **Daily** snapshot. A new parquet file is written to the `train` split every operational day.
- File naming convention: `data/train-00000-of-00001.parquet`; the Hub dataset-viewer auto-registers the
  `train` split (see YAML metadata above).
- Previous days are kept for trend analysis; the model card describes the **daily retraining** policy.

## 3. Data scale

- **Records:** 2,000 shipments (within the target 1,000–10,000 per day).
- **Splits:** `train` = 2,000 rows (100%), Parquet (columnar), LFS-tracked on the Hub.
- **Geography:** BR (São Paulo, Rio de Janeiro, Curitiba, Porto Alegre, Santos, Campinas),
  AR (Buenos Aires, Rosario, Córdoba), UY (Montevideo), PY (Asunción).
- **Carriers:** TransBrasil, AndesCargo, MercosulLog, PampaFreight, RioGrandeTruck,
  PatagoniaExpress, LitoralLog, SerraTruck.

## 4. Field dictionary

| Field | Type | Description |
|-------|------|-------------|
| `shipment_id` | string | Unique shipment identifier, e.g. `SH-20260813-00001` |
| `route` | string | Origin–destination pair, e.g. `São Paulo → Buenos Aires` |
| `carrier` | string | Haulage company (TransBrasil, AndesCargo, MercosulLog, ...) |
| `origin` | string | Departure hub (city name) |
| `destination` | string | Arrival hub (city name) |
| `departure_time` | timestamp | Actual departure time (America/Sao_Paulo, UTC-3) |
| `arrival_time` | timestamp | Actual arrival at destination |
| `delay_minutes` | int | Total arrival delay in minutes (0 if on time) |
| `damage_flag` | int | 1 = cargo damaged, 0 = intact |
| `temperature_deviation` | float | Deviation from target temperature (°C, signed) |
| `humidity_deviation` | float | Deviation from target relative humidity (%RH, signed) |
| `fuel_consumption` | float | Total fuel consumed (litres) |
| `quality_score` | float | Composite quality score 0–100 (100 = perfect) |

## 5. Known South American logistics failure modes reflected in the data

- **Border congestion (Brazil–Argentina / Brazil–Uruguay / Brazil–Paraguay):** cross-border shipments
  (`São Paulo↔Buenos Aires`, `Porto Alegre↔Buenos Aires`, `Porto Alegre↔Montevideo`, `Curitiba↔Asunción`)
  show systematically higher delays due to queueing at Foz do Iguaçu / Paso de los Libres and customs
  paperwork.
- **Cold-chain drift on Argentine agri-produce:** shipments on the Pampas corridors (`Buenos Aires↔Rosario`,
  `Córdoba↔Rosario`, `Córdoba↔Buenos Aires`) carry higher `temperature_deviation` and `humidity_deviation`,
  reflecting chilled/fresh grain, beef and citrus cargo handling.
- **Wet-season road conditions:** high-rainfall corridors show an elevated `damage_flag` rate and extra fuel
  burn (rain factor modelled per corridor).

## 6. Quality improvement goals (QIP)

1. Reduce the **high-risk order ratio** below 5% (today: 5.0% flagged by the anomaly model).
2. Cut **average border-crossing delay** by 15% over 30 days via slot-booking at the frontier.
3. Reduce **cold-chain deviations** on AR produce by 20% through pre-trip reefer calibration.
4. Lower **wet-season damage rate** by 10% with route re-routing and load-securing standards.
5. Achieve **-10% fuel consumption per tonne-km** by avoiding congested corridors.

**Daily KPI report:** high-risk share, mean delay, delay improvement vs. trailing 7-day baseline, damage rate,
avg temp/humidity deviation, and fuel efficiency. See the model card for the anomaly-detection definition.

## 7. Usage

```python
from datasets import load_dataset
ds = load_dataset("toolathon123/project_20260813_014231_9dbe9212", split="train")
```

Companion artefact: `model/` directory contains the IsolationForest risk-scoring model trained on this
snapshot.

---

*Generated by the QIP daily pipeline — reproducible with the seeded generator in the repository provenance.*