Companion to the substrate-completeness report. tl;dr: corpus is 91% Benzinga wire — safe, deep, US-heavy. High-signal sources we're under-indexing: TechCrunch (18.2% deal-density per article, highest in the corpus), DealStreetAsia, Bloomberg, and the entire non-English world (1-2% of the corpus).
Article volume isn't the right way to rank sources. The right ranking is deals per article ingested — how much unique deal signal we extract per unit of ingest cost.
| Source domain | Articles | Funding deals | Merger deals | Density |
|---|---|---|---|---|
| www.benzinga.com | 2,259,486 | 42,894 | 31,226 | 1.9% |
| www.prnewswire.com | 48,475 | 2,050 | 1,735 | 4.2% |
| www.yahoo.com | 36,362 | 558 | 171 | 1.5% |
| www.finanznachrichten.de | 20,024 | 1,650 | 1,256 | 8.2% |
| news.yahoo.com | 16,248 | 740 | 108 | 4.6% |
| www.globenewswire.com | 15,358 | 585 | 476 | 3.8% |
| techcrunch.com | 9,297 | 1,692 | 221 | 18.2% |
| finance.yahoo.com | 8,094 | 143 | 171 | 1.8% |
| www.moneycontrol.com | 7,813 | 387 | 214 | 5.0% |
| www.thehindubusinessline.com | 7,567 | 733 | 306 | 9.7% |
| www.thehindu.com | 7,115 | 171 | 158 | 2.4% |
| www.cnbc.com | 7,105 | 470 | 166 | 6.6% |
| www.thestar.com.my | 3,972 | 126 | 69 | 3.2% |
| seekingalpha.com | 3,677 | 102 | 167 | 2.8% |
| www.livemint.com | 2,764 | 35 | 162 | 1.3% |
| www.dealstreetasia.com | 1,651 | 298 | 100 | 18.0% |
| tradearabia.com:443 | 1,637 | 18 | 114 | 1.1% |
| hindi.moneycontrol.com | 1,632 | 87 | 61 | 5.3% |
| www.prnewswire.com:443 | 1,444 | 42 | 34 | 2.9% |
| www.afr.com | 1,443 | 198 | 124 | 13.7% |
| www.bloomberg.com | 1,360 | 222 | 70 | 16.3% |
A serious research desk expects Bloomberg / WSJ / FT / Reuters coverage. Right now the substrate has almost none of it — the paywall + robots.txt combination means GDELT under-indexes premium press systematically.
| Source | Articles in substrate | Δ vs prior | Est. annual output | Fix path |
|---|---|---|---|---|
| Bloomberg | 1,401 | ~30,000 | rss team parquet — follow up | |
| WSJ | 42 | ~15,000 | Dow Jones API / rss | |
| Financial Times | 105 | ~12,000 | FT API (paid) / rss | |
| Reuters | 183 | ~20,000 | Refinitiv (paid) / rss | |
| The Information | 0 | ~2,000 | Subscription + direct ingest | |
| Axios | 0 | ~5,000 | RSS free; GDELT should catch | |
| Businesswire | 0 | ~150,000 | One-line source-list add | |
| TechCrunch | 9,297 | ~5,000 | Direct-ingest (task #64) | |
| CNBC | 7,105 | ~50,000 | GDELT-fed, healthy density | |
| PRNewswire | 49,919 | ~500,000 | Wire — depth is fine |
Of 2.47M articles, a small fraction are from non-English publishers. Germany is best-served (via Finanznachrichten). Everything else is essentially token coverage.
| Region / language | Articles | Δ vs prior | Gap |
|---|---|---|---|
| German | 20,024 | Best-served; extend to Handelsblatt / Manager Magazin | |
| Asian (CN/HK/TW/JP/KR) | 0 | Nikkei / Caixin / SCMP absent | |
| Hindi | 1,632 | hindi.moneycontrol only | |
| Arabic | 1,637 | tradearabia only; Zawya / MEED absent | |
| Italian | 665 | ilsole24ore only | |
| Spanish | 408 | Expansion / El Economista absent | |
| French | 0 | Les Échos absent |
Task #61 (GDELT multilingual ingest via native-language detector + translate-on-pass) is the P0 unlock for the international sales conversation. Qwen translate-on-pass is $0-marginal.