Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Guides Qdrant query volume scaling. Use when someone asks 'query returns too many results', 'scroll performance', 'large limit values', 'paginating search results', 'fetching many vectors', or 'high cardinality results'.
.claude/skills/qdrant-qdrant-scaling-query-volume/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-13 | ✗→✓ | ▲ Improved | 4% | 0% |
| case-01 | ✗→✓ | ▲ Improved | -29% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -34% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -40% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -45% | 0% |
Problem: When a query has a large limit (e.g. 1000) and there are multiple shards (e.g. 10), naively each shard must return the full 1000 results — totaling 10,000 scored points transferred and merged. This is wasteful since data is randomly distributed across auto-shards.
Instead of asking every shard for the full limit, ask each shard for a smaller limit computed via Poisson distribution statistics, then merge. This is safe because auto-sharding guarantees random, independent data distribution.
The strategy trades a small probability of slightly incomplete results for a large reduction in inter-shard data transfer, especially for high-limit queries across many shards. The 1.2x safety factor and the 99.9% Poisson threshold keep the error rate very low — comparable to inaccuracies already introduced by approximate vector indices like HNSW.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-13 | fail→pass | 18,120 | 17,690 | -2% | 1 | 1 | 0% | 2,573 | 2,686 | +4% | 0 | 0 | — |
case-01 | fail→pass | 24,616 | 15,302 | -38% | 1 | 1 | 0% | 3,234 | 2,292 | -29% | 0 | 0 | — |
case-02 | fail→pass | 22,083 | 13,432 | -39% | 1 | 1 | 0% | 4,056 | 2,657 | -34% | 0 | 0 | — |
case-03 | fail→pass | 32,065 | 16,362 | -49% | 1 | 1 | 0% | 4,256 | 2,560 | -40% | 0 | 0 | — |
case-04 | fail→pass | 21,291 | 8,433 | -60% | 1 | 1 | 0% | 2,855 | 1,583 | -45% | 0 | 0 | — |
case-05 | pass→pass | 16,491 | 5,080 | -69% | 1 | 1 | 0% | 2,642 | 939 | -64% | 0 | 0 | — |
case-06 | pass→pass | 11,447 | 6,579 | -43% | 1 | 1 | 0% | 2,055 | 1,286 | -37% | 0 | 0 | — |
case-07 | fail→fail | 11,656 | 13,842 | +19% | 1 | 1 | 0% | 1,996 | 2,185 | +9% | 0 | 0 | — |
case-08 | pass→pass | 19,219 | 14,966 | -22% | 1 | 1 | 0% | 3,076 | 2,750 | -11% | 0 | 0 | — |
case-09 | fail→pass | 20,863 | 8,411 | -60% | 1 | 1 | 0% | 3,451 | 1,836 | -47% | 0 | 0 | — |
case-10 | fail→fail | 13,216 | 5,303 | -60% | 1 | 1 | 0% | 1,199 | 1,113 | -7% | 0 | 0 | — |
case-11 | fail→fail | 22,120 | 13,961 | -37% | 1 | 1 | 0% | 3,406 | 2,464 | -28% | 0 | 0 | — |
case-12 | pass→pass | 13,644 | 12,168 | -11% | 1 | 1 | 0% | 2,504 | 2,078 | -17% | 0 | 0 | — |
case-14 | fail→pass | 15,429 | 4,353 | -72% | 1 | 1 | 0% | 2,548 | 1,026 | -60% | 0 | 0 | — |
case-15 | fail→pass | 15,623 | 2,497 | -84% | 1 | 1 | 0% | 2,807 | 574 | -80% | 0 | 0 | — |
case-16 | fail→pass | 20,684 | 14,197 | -31% | 1 | 1 | 0% | 2,967 | 2,800 | -6% | 0 | 0 | — |
case-17 | fail→fail | 12,637 | 9,197 | -27% | 1 | 1 | 0% | 1,724 | 1,504 | -13% | 0 | 0 | — |
case-18 | pass→pass | 18,161 | 2,022 | -89% | 1 | 1 | 0% | 2,656 | 567 | -79% | 0 | 0 | — |
case-19 | pass→pass | 15,509 | 12,460 | -20% | 1 | 1 | 0% | 2,437 | 2,220 | -9% | 0 | 0 | — |
case-20 | fail→pass | 27,991 | 4,438 | -84% | 1 | 1 | 0% | 1,618 | 551 | -66% | 0 | 0 | — |
case-21 | fail→pass | 39,969 | 1,785 | -96% | 1 | 1 | 0% | 3,320 | 468 | -86% | 0 | 0 | — |
case-22 | pass→pass | 13,301 | 12,629 | -5% | 1 | 1 | 0% | 1,974 | 2,127 | +8% | 0 | 0 | — |
case-23 | pass→fail | 21,963 | 15,596 | -29% | 1 | 1 | 0% | 3,213 | 3,144 | -2% | 0 | 0 | — |
case-24 | pass→pass | 20,667 | 18,552 | -10% | 1 | 1 | 0% | 2,856 | 3,212 | +12% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 24 cases were attempted, and 22 counted toward the lift figure. The other 2 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +42 percentage points is the difference between those two pass rates over the 22 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.