Fix copy-engine full scans, retry-inflated paper exit rate, review page timeouts

[yonixnetwork]

Follow-up to the server outputs after deploying 8ad1d33 (db_health, parity_report --days 7, pumpswap_window_check).

What the server showed

• Load 9.48 on 2 vCPU, 392 MB free, 1.17 GB swap. evm_trades 3.5 GB / 4.4M rows. The review aggregates still ran past 120 s.
• Copy trading: TOO_LATE BSC 270,217 and Robinhood 6,648. Median detection was 300 s on BSC and 217 s on Robinhood.
• Paper exit failure rate was 50% (the cap), "measured from 6553 live SELL orders". 6,450 of those were retries of the one stuck 7wmm sell.

Changes

• Index ix_evm_trades_chain_at (chain, at): migration 0042, built CONCURRENTLY.
  • The copy engine's once-a-second poll compared lower(trader), which no index serves, so every tick read the whole table on each chain.
  • The 14-day prune and the market-regime windows also scanned the whole table.
  • The poll is now target_trades(). A test checks that the query plan uses the new index; it fails without it.
• measured_live_rates: counts one result per position and side (the first final attempt). Retries of a stuck exit no longer count as separate trials. There is a new test with 300 retries.
• Review endpoints (ledger review, comparison, EVM ML):
  • They are computed in a background task, one at a time, on their own pool of 2 connections with a 240 s statement limit.
  • Requests get an answer straight away from the last result, marked stale while a refresh runs.
  • Before the first result exists, the answer is 503 REVIEW_COMPUTING.
  • The pages show when the numbers were computed. Covered by new tests.
• db_health: also times the copy poll and lists the background review results (compute time, last error).
• Audit report: new section V with the server findings, what was fixed, and what still needs the operator (server resize; 7wmm second-RPC re-simulation and backoff, not implemented).

Not changed

• Live execution is unchanged. No live transaction was sent. Paper stays the default. Tasks #138 and #178 stay blocked or locked.

Tests

• ruff clean.
• core 881, api 171, data-solana 10, data-evm 14, copy-engine 13, discovery 18, migration 11, momentum 11, decision-engine 58, ml 43, paper-trading 81.
• alembic upgrade + check clean. Web tsc, lint and build OK.

Not verified

• Production copy latency and review compute times. Re-run parity_report and db_health after deploy.

🤖 Generated with Claude Code

https://claude.ai/code/session_015vpuZ1UX4MXz7NCF7wrtDv

---

Generated by Claude Code