Avoid a full miner_predictions scan when loading scores

[stas-synth]

get_miner_scores never bounds validator_requests in time, so the planner can't use the asset/time_length predicates to restrict the join and falls back to a full scan of miner_predictions before filtering. A prediction is scored exactly when its window closes — scored_time = vr.start_time + time_length — so the added bound is implied by the existing scored_time filter and removes no rows. Plan cost 5.59M → 2.09M; 110s → 30s on a 10-day window; verified identical over a fixed window: 491,520 rows both sides, zero rows in either EXCEPT, matching md5 across the full result set including computed window values. The invariant was checked separately: 0 violations across 3.7M rows over 14 days.

[claude]

Claude Code Review

This repository is configured for manual code reviews. Comment @claude review for a one-time review, or @claude review always to subscribe this PR to a review on every future push.

_{Tip: disable this comment in your organization's Code Review settings.}

[Copilot]

🟢 Approval recommended

The added predicate matches the score timestamp construction and preserves existing query results.

0 open findings

What changed in this PR

Optimizes score loading by adding a logically equivalent time bound that enables indexed request and prediction lookups.

Changes:

• Adds an indexed validator_requests.start_time constraint.
• Documents the query-plan improvement and validation results.

File	Description
synth/​validator/​miner_data_handler.py	Avoids a full miner_predictions scan when loading scores.

🧠 Review effort: Balanced

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.