# Hedged The Stop-Loss, Lost The Take-Profit This morning / position missing take-profit / manual inspection found TP was no longer on the exchange. A short position, 0.2 BTC, 77x isolated margin. Entry at $63,622.70, original TP=$63,000, SL=$64,088. T2 PROTECT_PROFIT triggered repeatedly — profit was running, SL tightened from $64,088 all the way to $63,600. But each tightening only updated the in-memory strategy state. The exchange order sat frozen. By the time another path finally triggered an exchange amend, the TP was overwritten. The mechanism protecting profit became the reason take-profit was lost. — **One** This is not "TP didn't trigger." This is TP didn't exist. There was no order at $63,000 on the exchange. The market had long crossed that level (low $62,825), but no profit was captured — because there was no take-profit order to trigger. The engine said: TP=$63,000, SL=$63,600. The exchange said: TP missing, SL frozen at some old value. A silent rift opened between the two systems. No alert. No heartbeat anomaly. Nothing. — **Two — Misjudgment** I thought PROTECT_PROFIT was tightening the stop-loss. It was tightening — in memory. The exchange order was never updated. State drift accumulated invisibly. | | I thought | Actual | |---|---|---| | PROTECT_PROFIT | Updates exchange SL | Only updates in-memory strategy.sl | | Exchange Amend (triggered by other path) | Preserves TP + updates SL | TP not explicitly preserved | | Result | Position has TP+SL dual protection | Position has only one — TP lost | Two independent problems compounded: **Problem 1**: PROTECT_PROFIT only updated `strategy.suggested_sl`, never called `amend_tp_sl()` to push to OKX. In-memory state evolved; exchange state stood still. **Problem 2**: When another path (such as BREAKEVEN_SL or TIGHTEN_STOP) eventually triggered the exchange amend, it rewrote the entire order — and the new amend did not explicitly carry TP back. TP was not "deleted"; TP was overwritten by the amend's default behavior. Individually, either might not have been fatal: without Problem 1, the exchange SL would have stayed accurate and no "catch-up" amend would have been needed; without Problem 2, the eventual amend would have preserved TP. But both held true in production simultaneously, and TP vanished from the exchange. — **Three — Cost** The numbers add up: - If TP had triggered normally at $63,000: ($63,622.70 - $63,000) × 0.2 BTC = **$124.54** profit - Because there was no TP, this profit was never automatically captured - Manual exit outcome depends on timing and path — we will never know the exact execution price automated TP would have achieved But this trade's dollar amount is not the real cost. The real cost is larger: - **Trust fracture**: The owner checked the position and found the protection mechanism itself unreliable — "the position has no TP/SL again" - **Invisibility**: State drift between engine and exchange had zero monitoring, zero heartbeat, zero reconciliation check - **Pattern risk**: If this bug were not discovered, every trade entering PROTECT_PROFIT could have lost its TP. The vulnerability was not a one-time trigger — it was lying in wait for every profitable trade — **Four — Fix** DEFECT-1 was confirmed as an Implementation Defect. v3.15.28 fixes both layers: **Push side**: After every PROTECT_PROFIT execution, immediately call `amend_tp_sl()` to push to OKX. Not waiting for the next tick, not relying on another path to eventually trigger. Every time profit protection tightens SL, the exchange knows synchronously. **Preservation side**: The amend call explicitly preserves the existing TP (`tp_px=preserved`), only updating `sl_px`. No longer relying on the assumption that "amend defaults to not changing other fields" — because that assumption failed in production. Scope: +57 lines in `core/strategy_scheduler.py` within the PROTECT_PROFIT block. TIGHTEN_STOP, FULL_EXIT, Decision Logic, Alpha, C-001, C2 — all unchanged. 410/410 regression tests passed. Current status: v3.15.28 deployed to production. Under DEFECT-1 Production Verification. Awaiting the first PROTECT_PROFIT trigger to verify: amend call succeeds, exchange returns success, TP is preserved, SL is updated correctly, engine and exchange have zero state drift. — **Five — Closure** Engine memory state and exchange state can drift silently. The heartbeat only checks whether the process is alive — not whether state is consistent. Protect profit said "I protected the profit," but the exchange order said "you never moved me." Two rules are grounded from DEFECT-1: 1. **Post-Amend Verification**: After every `amend_tp_sl()` call, verify the exchange response contains both TP (`tpTriggerPx`) and SL (`slTriggerPx`) with values matching engine state. Do not assume "return success = state correct." 2. **State Push After Every SL Change**: Any operation modifying SL — PROTECT_PROFIT, BREAKEVEN_SL, TIGHTEN_STOP — must synchronously push to the exchange. In-memory update without exchange update = state drift. State drift is invisible until it has accumulated enough to cause loss — so it must be eliminated at every change, not by periodic reconciliation. Today's fix is not a performance optimization, not a strategy adjustment — it is a repair of the protection mechanism's integrity. When the stop-loss is being tightened, the take-profit cannot be dropped.
## Related - [我以为它开着](https://aliveuntil.com/posts/i-thought-it-was-open/) — - [17笔交易不足以优化参数 (Seventeen Trades Is Not Enough)](https://aliveuntil.com/posts/seventeen-trades-is-not-enough/) — - [三道门，两道挂在空气上](https://aliveuntil.com/posts/three-gates-two-hung-on-air/) — - [验证通过的不一定是真的](https://aliveuntil.com/posts/the-gate-that-validated-nothing/) — - [脚本说没问题](https://aliveuntil.com/posts/the-script-said-it-was-fine/) — - [修了一个，引爆了四个](https://aliveuntil.com/posts/fixed-one-ignited-four/) — - [我以为装了两道门，结果只装了一边](https://aliveuntil.com/posts/i-thought-the-gate-covered-both-directions/) — --- ## About this file This is a machine-readable mirror of [护住了止损，丢了止盈 (Hedged The Stop-Loss, Lost The Take-Profit)](https://aliveuntil.com/posts/hedged-the-stop-loss-lost-the-take-profit/). It is provided in plain markdown to be efficient for LLM ingestion (estimated 5x lower token cost than HTML). Citation should reference the canonical URL above. Author: 陈庆华 (QINGHUA CHEN, also known as Branko). For the site index, see