Coverage
Are we collecting evenly across the 40 seeded actors and across all four source types (arXiv / website / news / patent)? A skewed corpus weakens every downstream signalling-theory claim.
Source-kind mix
A heavy long-tail on one source signals a collection bias — e.g. “website only” means we miss what arXiv and news reveal.
Weekly throughput
Signals collected per ISO week. Looks for steady cadence and flags gaps where the cron run produced nothing.
| Week | Signals | Distribution |
|---|---|---|
| 2026-W31 | 328 | |
| 2026-W30 | 350 | |
| 2026-W29 | 493 | |
| 2026-W28 | 283 | |
| 2026-W27 | 180 |
Per-actor coverage
One row per seeded actor. Distribution column shows the source mix inside that actor’s signal pool. A row with everything in one bar is fine for arXiv-only research labs, suspicious for portfolio-style VCs.
Coverage = (actors with ≥1 signal in the window) / (actors seeded). A single ISO week is Monday–Sunday; iso_week is calculated from signals.inserted_at.