2026-04-25 to 2026-10-29
7 backtest-tail pools (biweekly) + 98 forward cron pulls (daily)
The news-coverage bridge across the backtest→forward handoff (2026-07-22): the backtest's last ~3 months of ranked top-100 pools (GKG discovery + look-ahead-clean Wayback ledes) spliced onto the forward cron's daily WebSearch pulls. Both Wayback and WebSearch ledes are look-ahead-safe (WebSearch because the forward news is genuinely current); the backtest-only live-fallback is the single look-ahead-biased provenance. In chart 4, read the step down at the handoff as a change of units before reading it as a change of supply: the blue points are ranked pools truncated at max_articles = 100, so they sit on that cap by construction, while the gold points are whatever the forward corpus held inside the news window.
The user-set investor_profile.md knobs relevant to news retrieval.
| Backtest-tail cadence | biweekly (top-100 pools, Wayback ledes) |
| Forward cadence | daily cron (WebSearch pulls) |
| News lookback (backtest) | 14 days |
| Max articles / backtest pool | 100 |
The full per-wave search-term lists behind plot 8 live in retrieval_config.json (wave_keywords), which is gitignored, so this link resolves only when you are viewing a local copy of the dashboard. The forward cron queries the same waves.
65% of the 3,179 unique articles (2,074) carry a byline (native capture on both sides: Wayback/live extraction on the backtest tail, the WebSearch pull on the forward side).