← high signal

Gazetteer candidates

Tokens that appeared in events with no entity match in the last 720h — the strongest candidates to add to ai_infra_entities.csv. Scanned 500 unmapped events.

window24h72h7d30d

$TICKER mentions

0 candidates

Cashtag-prefixed tickers (e.g., $NVDA) — high confidence, the writer explicitly marked them as instruments.

no unmapped $tickers in the last 720h

Bare tickers

1 candidates

UPPERCASE tokens (3–5 chars) that match a real symbol in our equities universe but appear without a $ prefix. Higher noise floor than $TICKER — stoplisted common words filtered, but still review samples before promoting.

Bare entities (open-world)

30 candidates

Capitalized proper-noun phrases (1–3 words) that aren't yet seeded — most valuable for private companies that have no ticker at all (Anthropic, Mistral, Perplexity). Filtered by stoplist + ≥2 mentions or ≥2 sources to suppress noise. Manually verify before adding to the gazetteer.