CUR2-2107: sui object event deltas followup

[tomfutago]

Summary

This PR refactors Sui object-event anchoring so tokens_sui_object_event_deltas no longer reconstructs prior object state by scanning recent history directly. Instead, it introduces tokens_sui_coin_object_latest_state as a compact helper table with one latest pre-window Coin object state per object_id, and uses that helper to stitch incremental windows safely. The downstream transfer models keep the same output contract, while the anchor reconstruction work moves into a reusable state layer.

Expected Performance

Structural reasoning, pending the first production-shaped CI/prod run:

• tokens_sui_object_event_deltas no longer scans recent month-to-date history to reconstruct anchors.
• Prior-state anchors now come from tokens_sui_coin_object_latest_state, joined to the active window’s window_first_version object set.
• The old MTD recent_history_anchors scan is eliminated from object_event_deltas.
• The added window_first_version group-by runs over the already-loaded active window rows, so it should be small relative to the removed history scan.

Rough estimate:

• Current incremental runtime is around 25m, a more realistic first expectation is “meaningfully below that,” roughly in the 5-15m range depending on merge overhead

[tomfutago]

Regression Report: PR #9599

Models tested: 1

Scope note: CI relations were built with the current bounded window 2026-01-01 to 2026-04-27, so this is a bounded-window regression, not full-history parity.

tokens_sui_object_event_deltas

• Row count: prod 512,633,861, ci 512,633,861 (0.000000%) -- pass
• Daily coverage: 0 dates with diffs -- pass
• Aggregates (7d): rows match (26,875,247); sum(balance_delta) matches (16,381,918,215,434,972,864); unique_txs match (19,248,489); unique_objects match (3,468,636) -- pass
• Key-match: ci_only 0, prod_only 0 -- pass
• Block range: bounded by CI 2026-01-01 to 2026-04-27, checkpoints 229,588,356 to 269,524,071
• Uniqueness: 0 duplicate keys, 0 duplicate rows -- pass

Verification queries:

tokens_sui_object_event_deltas

• summary: 7382427
• daily coverage: 7382428
• aggs: 7382429
• key-match: 7382430
• uniqueness: 7382431

Downstream diffs, briefly: direct_transfers has 304 prod-only rows; owner_net_transfers has 1,246,533 prod-only rows; base_transfers and transfers each have 15,020 prod-only and 1 CI-only row. Since object_event_deltas matches exactly and downstream SQL only changed the temporary start date, these are downstream stale incremental residual artifacts in current prod, not an object_event_deltas regression.

[cursor]

PR Summary

Medium Risk
Moderate risk because it changes incremental anchoring/windowing logic for tokens_sui_object_event_deltas, which can affect correctness at window boundaries and in delayed-run scenarios.

Overview
tokens_sui_object_event_deltas no longer rebuilds prior object state by scanning month-to-date history; it now joins a new incremental helper, tokens_sui_coin_object_latest_state, to fetch a single pre-window anchor row per object_id.

The old monthly tokens_sui_coin_object_anchor_state model and its seed are removed, schema/docs are updated accordingly, and the new helper enforces "strictly before the incremental window" semantics (with a bounded catch-up window) so anchors cannot overlap the active window.

^{Reviewed by Cursor Bugbot for commit c01b3b5. Configure here.}

[tomfutago]

bugbot run

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

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

^{Reviewed by Cursor Bugbot for commit c01b3b5. Configure here.}

[0xRobin]

@tomfutago this tightly couples the runs of the latest_state to the event_deltas model which is not ideal.
If they are ever out of sync they will start producing wrong data..
I'm not sure this is the right approach.