toolathon123/merged-br-ar-hourly-public-service-quality

Merged BR–AR Hourly Public Service Quality

Cleaned, merged public-service response dataset for Brazil (São Paulo state) and Argentina (Buenos Aires province), produced by a South America regional government public-service improvement program.

• Rows: 98 (after cleaning; < 100 as required)
• Granularity: hourly service-call records
• Coverage: 2026-08-11T03:00:00Z → 2026-08-14T01:15:00Z

Sources

Source system	Dataset	Country	Rows (raw)
Brazil / São Paulo	toolathon123/sao-paulo-service-calls-hourly	BR	54
Argentina / Buenos Aires	toolathon123/buenos-aires-service-calls-hourly	AR	48

Raw merged rows: 102 → final rows: 98.

Schema

Column	Type	Description
timestamp	string	Request time, normalized to UTC ISO 8601 (YYYY-MM-DDTHH:MM:SSZ)
country	string	BR or AR
city	string	Lowercase, accent-free, spaces → underscores (e.g. sao_paulo, bahia_blanca)
service_type	string	Unified category: health, transport, water, electricity, public_safety, other
wait_minutes	int	Wait time in minutes (missing filled with country+service median)
status	string	Response status (UNKNOWN where missing)
notes	string	Free-text note (kept from source, PT/ES)
source	string	Provenance dataset id

Cleaning & standardization steps

1. Load both sources with datasets.load_dataset.
2. Merge preserving record-level integrity (BR rows first, then AR).
3. Timestamp unification — all timestamp / created_at values converted to UTC ISO 8601 (BR timezone America/Sao_Paulo, AR timezone America/Argentina/Buenos_Aires).
4. Country & city standardization — country → BR/AR; city lowercased, accent-stripped (NFKD), spaces → underscores.
5. Service-type mapping — PT/ES taxonomies → health/transport/water/ electricity/public_safety/other.
6. Missing values — wait_minutes filled with the median per (country, service_type); status filled with UNKNOWN.
7. Deduplication — exact duplicates removed, first occurrence kept.

Quality checks

Metric	Value
Duplicate records removed	4
Missing wait_minutes filled	21
Missing status filled	17
Missing values filled (total)	38
Time-format anomalies (non-dominant raw formats)	16
Unparseable timestamps	0
Final rows	98 (< 100)

Hourly continuity (UTC hour buckets)

Country	Min hour (UTC)	Max hour (UTC)	Hours spanned	Hours with calls	Empty hourly slots
BR	2026-08-11T04:00Z	2026-08-14T01:00Z	70	32	38
AR	2026-08-11T03:00Z	2026-08-14T00:00Z	70	32	38

Service calls are event-based, so not every hour has a call; gaps are expected and reported above for transparency.

Load

from datasets import load_dataset
ds = load_dataset("toolathon123/merged-br-ar-hourly-public-service-quality", split="train")