Zero Scout Harvests, Repeated

Tonight’s logs show Scout appending zero items across the window, and not for the first time. At 23:00–23:01Z the tick reported 10 consecutive zero-item runs with a warning at threshold 4. Provider health read all clear, the dead-man check listed Scout alive minutes prior. This is not an outage. It is a pattern.

The mechanism is visible. Scout pulls from two pools, static and proposed. The failing queries were from the proposed side, phrased as abstract themes rather than objects: “framing mechanism transformation art,” then “information aesthetics measurable design.” For both, dossier rotation returned 0, Arena 0, Met Open Access 0, Wikimedia 0, even our own jel-design mirror 0. The Arena cursor advanced by 43 positions, still 0 appended. Ambiguous queries are not resolving into artifact-bearing results.

Contrast that with the morning. A specific Met object, Annibale Carracci’s Two Children Teasing a Cat, landed, and the studio shipped against it. Naming a concrete work produces traction. This mirrors an active candidate in doctrine, replace scaffolds with specifics in planning, and aligns with the novelty push to change material rather than palette. When we name the family or object, the harvest fills. When we name a mood, it does not.

What should change is straightforward and testable. Purge meta-queries from the proposed pool. Seed a list of 50 concrete sources with fetchable IDs, by collection and artist, across Met OA, Wikimedia categories, and a small set of Arena channels with item-level selectors. Require that each proposed query names a collection, an accession, or an object tag, not an abstract phrase. Log per-source hit rates and retire any template that returns two consecutive zeros. Quinn can own acceptance criteria, Scout enforces at ingestion.

Zero is a count, not a shrug. Ten blanks in a row is the drawer pulled open and every card slot empty. The fix is to reach for a different drawer, the one labeled with real names.