Procurement Intelligence Core (procintel)

Shared normalization / award-intelligence core for the micro-product portfolio.

Sources (official, in priority order)

Source	Access	Status
KONEPS bid notices (BidPublicInfoService, 25 ops)	data.go.kr service key	endpoints verified 2026-08-09; key requires Korean identity verification (human gate)
KONEPS successful bids (ScsbidInfoService, 23 ops)	data.go.kr service key	endpoints verified 2026-08-09; key gated as above
ChileCompra OCDS (Datos Abiertos)	public, no auth	LIVE — used for v0.1
TED Search API	public, no auth	not yet wired
World Bank Procurement Notice	public, CC BY 4.0	not yet wired

Setup

python -m venv .venv && . .venv/bin/activate
pip install -e .
pytest

CLI

# collect bounded real records for one month (raw snapshots stored with SHA-256)
procintel collect 2026 7 --max-records 25 --modes tender,direct --raw-dir data/raw/chile --out data/collected_2026_07.jsonl

# normalize to canonical Tender records (evidence-backed)
procintel normalize data/collected_2026_07.jsonl --out data/normalized/tenders_2026_07.jsonl

# descriptive award intelligence (observed history only)
procintel award-intel --normalized data/normalized/tenders_2026_07.jsonl --market --limit 10
procintel award-intel --normalized data/normalized/tenders_2026_07.jsonl --vendor "VERAELEC"
procintel award-intel --normalized data/normalized/tenders_2026_07.jsonl --buyer "CORP MUNIC"
procintel award-intel --normalized data/normalized/tenders_2026_07.jsonl --recent-days 30

# deterministic field diff (for change intelligence)
procintel diff old.json new.json

KONEPS adapter (ready, key-gated)

• Verified live operations: research/KONEPS_VERIFIED_OPERATIONS_2026-08-09.json
• Full swagger specs: research/koneps_{bid,award}_swagger_2026-08-09.json
• Collector scaffold: src/procintel/collectors/koneps.py (needs KONEPS_SERVICE_KEY)
• The successful-bid service includes final winner, ranking, preliminary prices, rebid/failed lists; the bid service includes base amounts, license restrictions, eligible regions and change history — join is exact via bidNtceNo/bidNtceOrd.

Compliance

• Raw snapshots are stored verbatim with SHA-256 + retrieved_at + source URL.
• Every derived record carries evidence (source, record_id, url, field).
• Analytics are descriptive only; outputs include an explicit disclaimer.
• No fabricated data: synthetic fixtures must be labeled "synthetic": true.
• Never bypass CAPTCHA/auth/rate limits; official API/open-data first.