Agent objective
- Interact with the supplied stateful environment.
- Produce verifier-checkable actions or artifacts.
- Maximise scalar reward under the package contract.
A time-boxed PostgreSQL optimization environment where agents rewrite one analytical cohort query and select up to three constrained indexes. Hidden evaluation requires exact business semantics before scoring p95 speedup across shifted data regimes alongside worst-case latency, index footprint, build cost, and write overhead—rewarding robust query-and-index co-design rather than narrow benchmark speed.
A time-boxed PostgreSQL optimization environment where agents rewrite one analytical cohort query and select up to three constrained indexes. Hidden evaluation requires exact business semantics before scoring p95 speedup across shifted data regimes alongside worst-case latency, index footprint, build cost, and write overhead—rewarding robust query-and-index co-design rather than narrow benchmark speed.
Enough detail to understand the intellectual terrain; generated instances, hidden mechanisms, and solution paths remain inside the private package.
| Environment | Mathematical or technical frontier | Adaptive research problem |
|---|---|---|
| Cohort semantics | Activation, retention, purchasing, revenue | Preserve exact windows, distinct-user logic, nulls, negative amounts, UTC grouping, output types, and ordering. |
| Workload diversity | Broad, narrow, filtered, and empty cohorts | Generalize across whale tenants, recent windows, country/channel selectivity, joint filters, and empty ranges. |
| Query rewriting | PostgreSQL analytical planning | Replace correlated subqueries with a parameterized read-only SELECT while retaining the full nine-column contract. |
| Index co-design | B-tree/BRIN selection under budgets | Choose at most three declarative indexes within size and build-time limits rather than emitting arbitrary DDL. |
| Operational robustness | Tail latency and write amplification | Improve p95 performance without material worst-case regression, excessive bytes, build cost, or write-path overhead. |
Two independently restarted candidates both preserved exact results across three hidden profiles and six workloads each. In the available SQLite proxy, the stronger reached 0.3623 reward and 2.08× geometric-mean speedup, yet still regressed slightly on one workload and earned zero tail/write components. That makes the task an honest optimization tradeoff. Crucially, these are proxy results—not official PostgreSQL scores—because planner and physical-index behavior may differ.
We publish aggregate behavior and task structure, while withholding generated instances, hidden labels, exact successful probes, private checks, and solution trajectories.
Shown with its provenance and limitations; it is not a performance guarantee.
Two independently restarted agents were correct. This is a deterministic SQLite proxy result, explicitly not the official PostgreSQL score.
As identified by the supplied artifact.
2 reported runs.
DATA-SQLDATA-0188-v2-local-report.md
Machine-readable provenance and the exact displayed metric are available in results.json.
The paid ZIP will live in a private R2 bucket. Vercel authorizes the buyer and issues a 2–5 minute object URL; R2 serves the bytes directly.
Authenticated buyer + entitlement check
+ private R2 object + 2–5 minute signed URL
= direct, auditable download
Package SHA-256
6f53f94e13134b46adfd725fe1786e1a7ae451c8f7c5fe00b524af9f713e82eeOne purchase licenses this identified item to one legal organisation for worldwide, perpetual commercial model training, evaluation, research and development. Redistribution and resale of the package are not permitted.