# Binance Copytrader Replication Handoff Created: 2026-06-01 UTC For: laptop-side successor agent repeating the Veronika HYPE path on other Binance copytrader channels. ## Read First Governance and environment: ```bash /var/www/vps2.happyuser.info/top/.codex_CLI_laptop_local_philosofy/PHILOSOPHY.md /var/www/vps2.happyuser.info/top/.codex_CLI_laptop_local_philosofy/AGENT_GOVERNANCE.md /var/www/vps2.happyuser.info/top/.codex_CLI_laptop_local_philosofy/ENVIRONMENT_POLICY.md /var/www/vps2.happyuser.info/top/.codex_CLI_laptop_local_philosofy/VERONIKA_TO_VERONIKA2_HANDOFF_20260526.md /var/www/vps2.happyuser.info/top/veronika/docs/AGENT_GIT_WORKTREE_POLICY.md ``` On VPS before Python/tools/tests: ```bash source /var/www/vps2.happyuser.info/top/.codex_CLI_laptop_local_philosofy/profiles/top_1.server.sh python --version ``` Laptop agents must use the laptop profile/runtime, not copied VPS venvs or VPS agent resume ids. Hard safety boundary: - Do not place live orders without explicit human approval for that exact step. - Do not read or print secrets or `.env` values unless the human explicitly approves and the task requires it. - Do not restart live, paper-live, or monitor processes from this handoff. - Do not delete runtime state. - Keep raw loops, NPZs, reports, live/paper-live state, and private runtime artifacts out of `dev`. ## What Veronika Achieved The HYPE path moved from source discovery to a guarded live canary: 1. Public Binance copytrader source positions were collected and normalized from public lead endpoints. 2. HYPE source-position state was converted into explicit strategy intents: open from source open, DCA from champion policy, close from validated source close/history. 3. Candidate 189 became a reusable DCA policy module instead of hardcoded runner behavior. 4. Paper-only/live-disabled canary proved the policy loop could run without an order path and with absolute capital guards. 5. Exchange live canary wrapper added a broker boundary: auth probe, stable client order IDs, precision/min-order preflight, open/close submission, exchange position sync, session DB, telemetry, status files, hot-stop/restart snapshots. 6. Monitor/alert tests covered stale status, rejected-order clusters, repeated close/open attempts, historical rejected-order dedupe, exchange mismatch, stale DB row filtering, and notional mismatch. 7. Responsibility was clarified: strategy is the trader; live runner is only broker/market-maker execution machinery. Observed HYPE live result from Veronika handoff: the runner detected the Binance copy source close via history and closed the BingX long with `FILLED` status at `2026-05-26T12:31:16.594828Z`. Treat that as evidence to inspect, not a reusable launch permission. ## File Map Source ingestion and canonicalization: - `obw_platform/meta_strategies/telegram_signal_dca/paper_live_binance_copy_public_positions.py` - public Binance copy open positions/history polling; - normalizes symbol, side, entry, mark, open/close timestamps, avg cost/avg close; - read-only; no order submission. - `obw_platform/meta_strategies/binance_online_copytrading/binance_online_copytrading.py` - broader public copytrading paper/shadow runner for multiple leads. - `reports/binance_copy_4751838302089254401_20260519/REPORT.md` - early contrarian-on-close evidence: 47 normalized/tested positions; DCA variants improved net and DD versus plain. Meta-strategy from source state to intent: - `obw_platform/meta_strategies/telegram_signal_dca/hype_copy_signal_meta_strategy.py` - filters source positions; - validates source history matches with open/close tolerances; - emits explicit `open_entry`, `dca_entry`, and `close_position` intents; - close policy is explicit: `source_close_closes_follower`. Champion DCA policy: - `obw_platform/meta_strategies/telegram_signal_dca/hype_cap100_champion_dca_strategy.py` - candidate 189 params; - DCA drops, multipliers, base/add notionals, levels; - owns trading policy and emits order-intent dictionaries. - `reports/hype_canary_smoke_20260525/RUNBOOK.md` - candidate 189 smoke selection and safety interpretation. - `docs/optimization/HYPE_VERONICA_OPTIMIZATION_ACCELERATION_PLAN.md` - acceleration and search policy; Python replay remains the reference until Numba equivalence passes. Paper/live-disabled and live canary boundary: - `obw_platform/meta_strategies/telegram_signal_dca/hype_cap100_live_disabled_canary.py` - paper-only runner by construction; - no exchange order submission path; - applies strategy intents with paper fills, slippage, caps, daily loss guard, deadline, and max order rate. - `obw_platform/meta_strategies/telegram_signal_dca/hype_cap100_bingx_live_canary.py` - exchange canary wrapper; - broker duties: auth probe, exchange profile/defaults, preflight, stable client order IDs, submit open/close, safe order logging, exchange position sync, state/status/telemetry/session DB, hot-stop/restart snapshot. - do not generalize to new capital until the same gates pass for the new channel/symbols. Runner/broker execution layer: - `obw_platform/runners/live_runner_dual_responsibility_map.md` - canonical statement of responsibility split. - `obw_platform/runners/strategy_intents.py` - typed `ExecutionIntent`, strategy-owned rejection backoff, entry/manage intent normalization. - `obw_platform/runners/live_runner_dual.py` - production broker/execution adapter for the dual-strategy path: symbol resolution, CCXT orders, precision, fill polling, cancellation, rollback, reconciliation, telemetry, heartbeat, stop guard. - `obw_platform/runners/virtual_exchange.py` - replay/preflight exchange with OHLCV NPZ/DB input, fills, rejections, state, fees, slippage, and broker-like interface. Monitor/telemetry: - `obw_platform/meta_strategies/telegram_signal_dca/live_session_telegram_monitor.py` - reads live dir status/session DB/process state and sends deduped Telegram alerts. - Runtime artifacts to inspect, not commit: - `RUN_STATUS.json` - `run_state_*.json` - `run_telemetry_*.jsonl` - `session.sqlite` - `live_equity.csv`, `live_candles.csv`, `live_chart_events.*`, `live_strategy_params.json` Key tests: ```bash python -m unittest obw_platform.tests.test_live_runner_strategy_intents python -m unittest obw_platform.meta_strategies.telegram_signal_dca.test_hype_strategy_runner_split python -m unittest obw_platform.meta_strategies.telegram_signal_dca.test_hype_cap100_bingx_live_canary python -m unittest obw_platform.meta_strategies.telegram_signal_dca.test_hype_cap100_live_disabled_canary python -m unittest obw_platform.meta_strategies.telegram_signal_dca.test_live_session_telegram_monitor ``` ## Philosophy Constraint The strategy is the trader. The live runner is the broker/market-maker executor. It may: - resolve symbols; - normalize precision/min order; - apply venue parameters; - submit/cancel/poll orders; - reconcile open orders, fills, and positions; - persist state and telemetry; - block unsafe execution because of venue/control constraints. It must not invent trading policy: - no runner-created open signals; - no runner-created TP/SL/DCA/close rules; - no hidden retry storm logic that bypasses strategy backoff; - no fallback from limit to market unless the strategy intent/policy explicitly allows it; - no close on source disappearance unless the meta-strategy emitted an explicit close intent. Tests that encode this split: - `test_strategy_entry_signal_is_the_only_open_source` - `test_live_runner_does_not_close_without_strategy_close_intent` - `test_live_executor_stops_dca_intents_after_first_blocked_dca` - `test_strategy_owned_backoff_blocks_retry_storms_and_resets_on_fill` ## Replication Sequence For A New Binance Copytrader Use a new owner branch/worktree for non-trivial work. Start with analysis-only and paper-only modes. 1. Collect and canonicalize signals. - Identify the Binance `portfolioId`. - Poll public open positions and position history. - Normalize to stable rows: `key`, source position id, symbol, side, entry, mark, opened/closed UTC, avg cost, avg close. - Save raw pages plus normalized CSV/JSON under a report path, not runtime state. - Reject rows with missing symbol/side/entry, unknown side, impossible timestamps, or no executable market mapping. 2. Build source-signal evidence. - Produce a compact report like `reports/binance_copy_4751838302089254401_20260519/REPORT.md`. - Separate signal modes: follow open, contrarian on close, long-only/short-only, source-close-following, autonomous exits. - Do not use Binance lead `avgCost`/`avgClosePrice` as promotion fills. They are artifact baselines only. 3. Paper/replay with executable fills. - Use first/next bar open, live paper snapshot entries, or virtual exchange fills. - Include slippage, fees, min-order gate, unlevered capital grounding, drawdown, MTM tail risk, and notional > equity checks. - Keep Python/exact replay as reference. Accelerated engines are usable only after equivalence passes. 4. Prove backtest/live equivalence. - Strategy code must be the same policy surface used by paper and live wrappers. - Meta-strategy emits explicit intents. - Runner only consumes intents. - Compare paper theoretical fills vs exchange/virtual fills. 5. Preflight with virtual exchange and restart evidence. - Run against `obw_platform/runners/virtual_exchange.py`. - Inject rejects/timeouts/open-order restart scenarios. - Verify stable client order IDs, no duplicate entries after restart, pending orders reconciled, and state snapshots recover. - Confirm `STOP_NEW_ORDERS`, `KILL`, and hot-stop controls are recognized. 6. Live-disabled canary. - Use a paper-only wrapper with no `create_order` path. - Enforce absolute caps before every base/DCA entry: - max gross notional; - max one-side notional; - max daily realized + unrealized loss; - max new orders per hour; - runtime deadline; - source freshness/delay. - Exits/reduces may remain allowed when they reduce risk. 7. Shadow/dry-order exchange validation. - Auth probe may confirm credentials presence/permissions without printing secrets. - Validate exchange, symbol, precision, min qty/notional, position mode, reduce-only/hedged params, and client order ID constraints. - Write desired orders/shadow orders; do not submit. 8. Small canary live. - Requires explicit human approval. - Use non-withdrawal API key, IP whitelist, hard caps, kill switch, no auto-restart, and dedicated runtime dir. - Start with tiny capital and single source/symbol policy. - Monitor from the first minute. 9. Monitor and telemetry gates. - `RUN_STATUS.json` fresh within threshold. - No rejected-order cluster. - No repeated open/close attempts for the same action. - No stale DB rows from other bot/exchange/path included in live notional. - No exchange mismatch. - No source/history mismatch beyond tolerance. - No unreviewed control flags/backoff active. 10. Scale-up criteria. - Several weeks profitable paper/shadow/live-small evidence. - Clean restart evidence. - Clean monitor history. - Slippage within configured bp gates. - Drawdown and daily loss within contract. - No duplicate entries, missed closes, or policy/runner responsibility violations. - Human-approved capital step with rollback plan. ## Failure Modes And Safeguards Added Repeated order POST storm after rejects: - Strategy-owned `StrategyRejectionBackoff` in `obw_platform/runners/strategy_intents.py`. - Live canary tracks order error backoff/circuit and entry failure cooldown. - Monitor detects rejected-order clusters and repeated attempts. Close/open exchange parameter errors: - `live_order_params()` handles position mode/exchange differences. - BingX hedge mode uses CCXT `hedged=True`; Gate.io uses safe `text` client field. - Close handles no-position/reduce-only errors as sync-only where appropriate. Stale DB rows causing false alarms: - Monitor filters position/notional checks by current live dir bot id and expected exchange. - Tests cover stale other-exchange and other-bot rows. Historical rejected orders causing repeated alerts: - Monitor baseline/seen-order state dedupes old fills/rejects. - `--init-baseline` marks existing orders without alerting. Source close vs own TP/DCA policy mismatch: - Source close is now an explicit meta-strategy close intent, not runner inference. - HYPE champion DCA policy remains in `hype_cap100_champion_dca_strategy.py`. - Future channels must decide whether source close always wins, whether autonomous TP/SL/TTL wins, or whether those are separate strategies. Do not mix them silently. Live/backtest responsibility split: - `live_runner_dual_responsibility_map.md` documents broker vs strategy ownership. - Tests enforce no open/close without explicit strategy/meta-strategy intent. ## Checklist Before New Live Capital - Read the governance files and activate the correct local environment. - Confirm branch/worktree owner and keep runtime artifacts out of `dev`. - Identify source `portfolioId`, symbols, and signal mode. - Save raw and normalized source evidence. - Produce a report with counts, period, trade count, net, PF, max DD, skips, and assumptions. - Use executable fill sources only for promotion evidence. - Split train/validation/holdout by trades/signals; do not tune on holdout after failure. - Verify exact replay and any accelerated replay equivalence. - Extract trader policy into a module; do not hardcode strategy policy in runner. - Add tests proving paper/live wrappers consume the same meta-strategy and policy modules. - Run virtual exchange preflight with rejects, restarts, no-position close, and pending-order scenarios. - Run live-disabled canary with hard caps and no order path. - Run monitor dry-run and baseline initialization. - Verify status/session DB/telemetry are fresh and scoped to the new runtime dir. - Get explicit human approval before any real exchange submit path. - For live-small: use tiny capital, non-withdrawal key, IP whitelist, kill switch, no auto-restart, and a rollback/stop plan. ## Open Questions And Risks Before Scaling - Which source-close semantics should each new channel use: source-close-follows, autonomous exit, or separate strategy families? - How stable are Binance public copy endpoints and fields across lead pages? - Are source position IDs stable enough for deterministic client order IDs and history matching on every lead? - Can multi-symbol channels preserve the same notional, DCA, and daily loss gates without hidden correlation risk? - `live_runner_dual.py` remains large; exchange adapter, telemetry, and bar loop are not physically split. - Pending-order restart lifecycle is improved but still needs richer DB-backed accounting before larger capital. - Slippage calibration from HYPE does not automatically transfer to other symbols/exchanges. - Monitor alerts are necessary but not sufficient; a human must review early live-small incidents before scaling. - Laptop/server parity must be proven per OS. Do not assume VPS paths, venvs, or tmux/session state exist on laptop. ## Minimal Command Shapes Read-only source poll/backfill shape: ```bash python obw_platform/meta_strategies/telegram_signal_dca/paper_live_binance_copy_public_positions.py \ --portfolio-id \ --report-dir reports/binance_copy__ \ --once ``` Smoke the responsibility split: ```bash python -m unittest \ obw_platform.tests.test_live_runner_strategy_intents \ obw_platform.meta_strategies.telegram_signal_dca.test_hype_strategy_runner_split ``` Smoke the HYPE canary and monitor surfaces before copying patterns: ```bash python -m unittest \ obw_platform.meta_strategies.telegram_signal_dca.test_hype_cap100_bingx_live_canary \ obw_platform.meta_strategies.telegram_signal_dca.test_hype_cap100_live_disabled_canary \ obw_platform.meta_strategies.telegram_signal_dca.test_live_session_telegram_monitor ``` Virtual exchange preflight shape: ```bash VIRTUAL_EXCHANGE_NPZ= \ VIRTUAL_EXCHANGE_SYMBOLS= \ VIRTUAL_EXCHANGE_ERRORS_JSON='{"reject_next_n_orders":1}' \ python ``` Do not convert any of these command shapes into live order submission without a new, explicit human instruction.