Project · signal collector

Sydney Startup Radar

An internal 14-day, provenance-first Sydney startup/AI signal collector. It is not a public site and does not claim that public attention signals equal willingness to pay.

Boundary: No outreach, no public deployment, no LinkedIn automation. Exactly five source lanes: Luma Sydney, Startup Daily RSS, Fishburners, Stone & Chalk, and public X via Rodion's existing grokx session.

Data model

SQLite persists seen_at, signal_type, person_org, startup, place, ask_or_offer, source_url, source_date, first_seen, confidence, duplicate_cluster, plus source/first-source bookkeeping. Tracking query parameters are removed before URL dedupe. Cross-source duplicate clusters preserve the first source and first-seen time.

Output

Each run renders at most 10 genuinely new clusters for the requested date; event-only signals are capped at 40%. Output defaults to a dated library digest path.

Verification

Run ./run.sh (runs test suite before collection). --skip-x for deterministic/offline verification. Source failures are reported explicitly in the final JSON rather than replaced with fabricated data.