Retriever Forwardtest (RFT)

Health of the live WebSearch feed behind the Curator Forwardtest (CFT) — ingest quality, not market outcome. In chart 1 the bars count articles and the dotted line counts search hits (a different unit, right-hand axis): one story returned by three pulls in a day is 3 hits but 1 article. New to corpus is a strict subset of distinct articles seen — the gap between them is stories the feed already had. Bars flat at zero means the daily cron stopped. Chart 2 is the query budget and what became of it: a red band is queries that cost a round trip and returned nothing. Chart 7 is the drift chart — a search engine that keeps handing back the same pages shows up as a median age climbing past news_lookback_days, at which point the typical result is already too old for the curator to read. Non-articles are quote and ticker pages: stored, tagged, never fed to a curator. Sister page: RBS covers the bootstrap-era corpus.

Summary

2,537
articles in corpus
106
pulls logged
2026-07-22 → 2026-10-29
window
91%
with body text
73%
with an author
24
non-articles dropped
1 day(s) with no pull
missed pull days
0/20
queries errored on the last pull day
6
days with >1 pull
2d
median age at pull
20%
older than the 14d window
100%
reached the curator (509/509, 2026-10-25)