[droplister]

Summary

• aggregate collector holdings before joining asset metadata and collection evidence
• drive collector selection through the existing positive-holdings partial index
• evaluate the collector persona only for chosen acquirers in stage two
• pass all 150 seed asset IDs through one JSON bind, eliminating five repeated collector scans without crossing D1's SQL-variable ceiling
• add an end-to-end query fixture covering the five-holder floor, known-collection and low-quality exclusions, dominant-airdrop suppression, and chosen acquisition evidence

Production proof

The cold collection-candidate seed query was capable of saturating D1 and causing unrelated routes to return D1 DB is overloaded.

Exact before/after runs over the live database with deterministic tie ordering returned the same 150 rows:

• before: 3,265,649 rows read; 7,465 ms
• after: 1,561,486 rows read; 2,329 ms
• after with the existing positive-holdings partial index explicitly selected: 1,381,322 rows read; 2,143 ms
• result parity: exact equality across every returned field

Stage two previously materialized the complete collector population five times in 30-asset batches. One 150-asset JSON batch measured 3,104,700 rows versus roughly 10M across five batches; evaluating the persona after the target acquisition join reduced that further to 2,492,062 rows and 3,405 ms.

No schema or response-shape changes.

Verification

pm test -w xcp-api: 469 passed, 0 failed (34 live-contract tests skipped locally by design)

pm run check: typecheck, ESLint, Prettier, and structure checks passed