feat(dex): wire 1inch-LOP aggregator trades into dex_aggregator.trades (CUR2-2693, 2/2)

[thevaizman]

Part 2 of 2 re-landing #9747 (CUR2-2693). Stacked on #9758 — merge only after it.

One line: adds ref('oneinch_lop_aggregator_trades') to the as_is_models list, so venue-settled 1inch-LOP fills (excluded from dex.trades by #9758) get their intent-layer record in dex_aggregator.trades — exactly how oneinch_ar_trades rows are served today.

Why this can't reproduce the #9752 revert

The revert's failure mode was generic erc20-decimals enrichment applied to lineage raw amounts. The as_is path skips enrichment entirely: rows carry 1inch's own enriched amounts and guarded amount_usd. Full-history validation across all 13 chains: 0 rows with amount_usd > $1B, 0 negative, max $38.3M (top-5 all plausible large fills). Additionally, this PR's own CI rebuilds dex_aggregator_trades with these rows and runs the exact dbt_utils_accepted_range test that failed in prod — pre-merge proof, not post-deploy discovery.

CI expectations

While #9758 is unmerged, this PR's CI builds the full stack (CI compares against the prod S3 manifest, not the base branch): venue_settled (~9 min) + views + dex_aggregator_trades full build (28.6–75 min observed variance for the same build) + tests. After #9758 merges and this branch rebases, the diff collapses to one line and CI builds only dex_aggregator_trades + its tests.

Post-merge

Full refresh of dex_aggregator_trades (backfills historical '1inch-LOP' rows), then re-run the audit query — the full-history gate says it passes. Note open PRs #9690/#9357 also touch this file (trivial append rebase for whoever lands second).

🤖 Generated with Claude Code

[cursor]

PR Summary

High Risk
Changes core DEX trade attribution and incremental merge behavior across dex.trades and dex_aggregator.trades; misclassification would affect volume/USD metrics at scale.

Overview
Fixes double-counting for 1inch limit-order fills that a resolver settles on an underlying DEX in the same transaction: the venue row stays in dex.trades, while the intent-layer fill moves to dex_aggregator.trades.

A shared oneinch_lop_evt_index() macro assigns evt_index over all limits fills before exclusions, so merge keys stay stable when rows are reclassified (merges upsert only). oneinch_lop_own_trades drops venue-settled fills via oneinch_lop_venue_settled_fills, applies a 6-hour maturity delay so venue decoded events land first, and adds a seed check. New oneinch_lop_venue_settled_fills (incremental) flags fills that share a tx with dex_*_base_trades, with keep-guards for nested classic/fusion/cross-chain and direct fills. oneinch_lop_aggregator_trades exposes those fills for aggregators (as-is amounts, like oneinch_ar_trades). dex_aggregator_trades unions ref('oneinch_lop_aggregator_trades') into as_is_models, skipping generic ERC20 enrichment. Seeds and a regression test guard the split.

^{Reviewed by Cursor Bugbot for commit e2bec45. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 2 potential issues.

Bugbot Autofix prepared a fix for the issue found in the latest run.

• ✅ Fixed: Six-hour merge race double-count
  • Delayed oneinch_lop_own_trades until fills age out of the venue classifier window and extended the passthrough incremental overlap so newly mature rows still merge once.

Or push these changes by commenting:

@cursor push 3659104cd6

Preview (3659104cd6)

diff --git a/dbt_subprojects/dex/macros/models/oneinch_lop_dex_trades_passthrough.sql b/dbt_subprojects/dex/macros/models/oneinch_lop_dex_trades_passthrough.sql
--- a/dbt_subprojects/dex/macros/models/oneinch_lop_dex_trades_passthrough.sql
+++ b/dbt_subprojects/dex/macros/models/oneinch_lop_dex_trades_passthrough.sql
@@ -46,7 +46,9 @@
 	AND o.block_date > current_date - interval '3' day -- dev_dates mode for dev, to prevent full scan
 {%- else -%}
 	{% if is_incremental() %}
-	AND {{ incremental_predicate('o.block_time') }}
+	-- oneinch_lop_own_trades only emits after classifier finality; keep a one-unit
+	-- overlap so newly mature rows are still picked up by dex_<blockchain>_trades.
+	AND {{ incremental_predicate("o.block_time + interval '1' " ~ var('DBT_ENV_INCREMENTAL_TIME_UNIT')) }}
 	{% endif %}
 {%- endif %}
 

diff --git a/dbt_subprojects/dex/models/_projects/oneinch/outside/oneinch_lop_own_trades.sql b/dbt_subprojects/dex/models/_projects/oneinch/outside/oneinch_lop_own_trades.sql
--- a/dbt_subprojects/dex/models/_projects/oneinch/outside/oneinch_lop_own_trades.sql
+++ b/dbt_subprojects/dex/models/_projects/oneinch/outside/oneinch_lop_own_trades.sql
@@ -10,7 +10,6 @@
 {% set src_symbol = "coalesce(src_executed_symbol, '')" %}
 {% set dst_symbol = "coalesce(dst_executed_symbol, '')" %}
 {% set placeholder_tokens = oneinch_cross_chain_placeholder_tokens_cfg_macro() | join(', ') %}
-{% set maturity_delay = "interval '6' hour" %}
 
 
 
@@ -70,6 +69,6 @@
     )
     -- maturity delay: the 1inch lineage builds from raw traces while venue base trades
     -- build from decoded events; a fill merged into dex_<blockchain>_trades before its
-    -- venue's event decodes would never be removed (merge can't delete), so fills only
-    -- pass through once old enough for the venue side to have landed
-    and f.block_time <= now() - {{ maturity_delay }}
+    -- venue classification finalizes would never be removed (merge can't delete), so
+    -- fills only pass through once they age out of the classifier's incremental window
+    and not ({{ incremental_predicate('f.block_time') }})

_{You can send follow-ups to the cloud agent here.}

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit e2bec45. Configure here.}

[tomfutago]

Regression Report: PR #9759

Overall status: PASS / no blocker

Scope: targeted regression for dex_aggregator_trades, focused on the new 1inch-LOP as-is hookup and the prior amount_usd accepted-range failure.

Accepted Range Check

dbt_utils_accepted_range_dex_aggregator_trades_amount_usd__1000000000 passed in CI.

Independent Dune scan of CI relation dune.dune_spellbook_ci__tmp_pr9759_27368340570_1.dex_aggregator_trades:

• Rows with amount_usd > 1B: 0
• Rows with negative amount_usd: 0
• Max amount_usd overall: $714.17M
• 1inch-LOP rows with amount_usd > 1B: 0
• 1inch-LOP max amount_usd: $38.35M

1inch-LOP Regression

Relations checked:

• Source: dune.dune_spellbook_ci__tmp_pr9759_27368340570_1.oneinch_lop_aggregator_trades
• Final: dune.dune_spellbook_ci__tmp_pr9759_27368340570_1.dex_aggregator_trades
• Prod: dex_aggregator.trades

Results:

• CI source 1inch-LOP rows: 3,881,477
• CI final 1inch-LOP rows: 3,881,477
• Prod 1inch-LOP rows: 0 as expected pre-merge
• Source-only keys: 0
• Final-only keys: 0
• Source duplicate keys: 0
• Final duplicate keys: 0

Note

There are 191 matched rows on 2026-06-11 where the live source view now has amount_usd populated but the materialized CI final table has NULL, totaling $380,378 of source-vs-CI drift. Since keys match exactly and the drift is confined to fresh rows after the CI snapshot, this looks like upstream/view enrichment drift rather than a PR regression.

Conclusion: safe to proceed from regression perspective.