perf(dex_solana.trades test): bound accepted_range scan to a recent window

[a-monteiro]

The dbt_utils.accepted_range test on dex_solana.trades.amount_usd ran with no where config, so every run full-scanned the entire table — 17.2B rows / 236.8 GB, ~1.34M cpu_ms, ~62s wall — just to flag trades above $1b. It runs 4 heavy times/day = ~0.95 TB IO and ~1.49 CPU-hrs/day, purely re-validating immutable history.

The test exists to catch outlier amount_usd as data lands, so bounding it to a recent window matches its intent. This adds a config.where of block_time >= now() - interval '7' day. block_time is a regular column with Delta min/max stats and block_month is the partition key, so the bound prunes to recent files (pushdown confirmed live).

Measured on prod data (spellbook-hourly, faithful full-column scan, 3 warm runs, medians):

axis	full history (prod)	7-day bound	factor
IO	236.8 GB	13.7 GB	17.2x
rows scanned	17.20B	100.3M	~171x
cpu_ms	1,338,645	85,556	15.6x
wall	62.3s	4.8s	~13x

Per day (4 heavy runs): ~947 → ~55 GB IO and ~1.49 → ~0.10 CPU-hrs.

No equivalence proof here on purpose: this changes the test's coverage from all-history to the recent window. An outlier landing more than 7 days ago (e.g. a late backfill) would no longer be flagged — wanting reviewer sign-off on the window length (happy to bump to 30d for a wider safety net; ~30d still scans far less than full history).

Compiled test confirms the bound wraps the relation:

from (select * from dex_solana.trades where block_time >= now() - interval '7' day) dbt_subquery

Fixes CUR2-2800

[a-monteiro]

• perf(dex_solana.trades test): bound accepted_range scan to a recent window #9779 👈 (View in Graphite)
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[cursor]

PR Summary

Low Risk
Test-only change that narrows data-quality coverage to the last seven days; no model SQL or production pipeline logic is modified.

Overview
Limits the dbt_utils.accepted_range check on dex_solana.trades.amount_usd so it only evaluates rows with block_time >= now() - interval '7' day, instead of full-table scans on every run.

The $1b cap is unchanged; only test scope and cost change. Outliers in amount_usd are still caught for recent ingests, but trades older than seven days (e.g. late backfills) would no longer fail this test.

^{Reviewed by Cursor Bugbot for commit e723d21. Configure here.}