Start a fresh Claude Code review session for jeanbrianc/BullyBearAI epic #1090. Review PRs #1128, #1129, #1130, #1134, #1135, #1136, #1139, and #1140. Treat the implementation report as a claim to verify. REFRESH Read CLAUDE.md, current delivery/deployment docs, epic comments, each PR's diff, review comments, checks, head commit, and base branch. Confirm #1124's current merged state and deployment proof. Refresh all dependencies before choosing an order. Use an isolated checkout and ensure the implementation session is no longer writing the branch you review. STARTING ORDER #1130 first; then #1129, #1134, #1136, and #1139; then #1135 and its child #1140; #1128 last. #1140 was based on #1135: inspect child-only changes, and retarget/rebase after the parent merges when authorized. Refresh #1128's baselines against the resulting tree and verify that thresholds were not loosened. Adjust this order to the live GitHub state. Do not merge without separate merge authorization. INDEPENDENT REVIEW Check each issue's acceptance criteria against the actual behavior. Use read-only review subagents for code, test gaps, and data boundaries when useful; one lead owns branch fixes and deployment. Reproduce meaningful tests and the repository's required gates at the exact commit under review. Distinguish author-reported results from independently reproduced results. SEQUENTIAL AWS VALIDATION This review pass authorizes deployment and testing of applicable changes in the existing bullybearai-dev environment using the project's documented deployment workflow. Verify the current AWS identity, configured account/region/role, target stack, baseline, and rollback path without printing secrets. Do not assume earlier authentication is still valid. Resolve a missing or conflicting target before deployment; continue useful local review while access is blocked. Use one deployment owner. Package the tested commit, record its artifact identity, deploy, wait for completion, and run the issue-specific smoke/integration checks with synthetic or approved test data. Confirm the relevant entrypoint actually exercises the changed code. A successful deployment does not prove unwired functionality works. Mark build-only changes as deployment not applicable with a reason; mark absent runtime wiring or human proof as blocked, never passed. Record the code state left in the shared environment before moving on. FIX AND RECHECK If a reproducible defect appears, fix it on the existing PR branch, add regression coverage, rerun affected and required checks, and push to that same PR. Coordinate any shared-branch rewrite. Verify the new head commit, rebuild, redeploy when applicable, and repeat validation. Old results do not validate a new commit. Halt the deployment sequence and restore the documented baseline if the environment becomes unhealthy. BOUNDARIES AND OUTPUT Keep broker access, credential mutations, notifications, approval/preflight creation, trading flags, and orders outside this pass. Do not fabricate customer studies, payments, or live cost evidence. Record on each PR: reviewed commit, findings, fix commits, commands/results, artifact and deployment evidence, skipped checks, remaining HITL proof, and a ready/blocked verdict. Return the final deployment state and next review or merge action for every PR.