Sweep Optimization Plan — 10 Steps

Created: 13 Apr 2026 Context: Bybit DCA Z-VWAP bot parameter optimization via backtesting

Status Overview

• Step 1: Fetch & cache data (step1_fetch_data.py → data_cache_30d.json)
• Step 2: Sweep Z×TP×SL core R:R (step2_sweep_z_tp_sl.py → results_step2_z_tp_sl.json)
• Step 3: Sweep from cache with R:R ratios (sweep_from_cache.py)
• Step 4: Backtest core library (backtest_core.py — run_backtest())
• Step 5: Download data in pickle format (download_data.py → .pkl caches)
• Step 6: Sweep runner — all 8640 combos (sweep_runner.py → sweep_results_7d.json)
• Step 7: Analyze sweep results ← CURRENT
• Step 8: Robustness check on 14d data ✅ (step8_robustness.py)
• Step 9: DCA overlay — SKIPPED (Rick: DCA провал, не делаем)
• Step 10: Apply to live bot ✅ (zvwap-bot launched 13 Apr 2026)

Step 10: Live Bot Config (Zvwap)

• PM2: zvwap-bot (was bybit-bot)
• Deposit: $159.64
• BO: $20, Leverage 3x, Max 6 positions
• Margin: $120 max (75% of dep)
• Entry: Z > 3.0, Z < 3.5 (skip breakout)
• Exit: TP 3% / SL 1% (R:R 3:1) OR Z exit < 0.3
• Filters: NATR 0.5-2.0%, CHOP ≥ 45, Vol > $20M
• No DCA, no SOs, no Time Stop (pure MR)
• TMM tag: "Zvwap" + "L"/"S"
• TMM desc: Z, NATR, CHOP, TP/SL prices at entry; PnL, ROI, duration, R:R at close
• Risk per trade: $0.60 (SL 1% × $20 × 3x)
• Est worst DD: ~$10 (6.3% of dep)
• TODO later: analyze live stats → tune TP/SL (maybe 4.5/1.5), add Time Stop

Step 7 Sub-steps

• 7.1 — analyze_sweep.py: Top by PnL/PF/WR, best R:R, heatmaps, parameter sensitivity ✅
• 7.2 — Run analysis → analyze_report_7d.txt + candidates_7d.json ✅
• 7.3 — Interpret results, pick candidates for step 8

Step 7 Results Summary (7d data)

• 8208 valid combos, 70 "golden" (PnL>0, PF>1.3, trades≥50)
• Best R:R: 3:1 dominates (TP=4.5% SL=1.5% — top 18 of 20)
• Best absolute PnL: +$79.5 (Z=2.2/3.5, NATR 0.5-2.0, CHOP≥55, 195 trades)
• Best PF: 1.50 (Z=2.5/3.0, same TP/SL, 111 trades)
• Sensitivity: Z_entry=3.0 best avg, NATR_min=0.5 best, NATR_max=2.0 best, CHOP≥55 best
• Avoid: 5:1 R:R with TP=7.5% (worst combos, -$174), Z_entry=2.0 (too early)
• 5 candidates saved for step 8 (14d robustness)

Step 8 Results — Robustness Check (7d vs 14d)

• 10 candidates tested (5 from step 7 + 5 bonus diverse Z=3.0 + different R:R)
• Only 1 stable (PnL>0 AND PF≥1.2 on 14d): Z3.0 tighter TP/SL (3:1, TP=3%, SL=1%)
• ALL combos degrade on 14d — PF drops 0.17-0.32 (🔴 most)
• Best 14d PnL: Top1 from 7d → +$55.5 (PF 1.17, 307 trades) — still profitable but PF below 1.2
• Best 14d PF: Z3.0 TP=3%/SL=1% → PF 1.21, +$45.0, 307 trades — ONLY stable candidate
• Key insight: 7d sweep was likely biased by favorable week. 14d normalizes. Tighter TP/SL (3/1) more robust than wide (4.5/1.5)
• 4:1 and 5:1 R:R collapse on 14d (PF 1.06-1.08) — too greedy for MR
• Files: robustness_report.txt, robustness_results.json

Sweep Data

• File: sweep_results_7d.json (2.8MB)
• Date: 2026-04-13
• Valid combos: 8208 / 8640
• Elapsed: 595.6s
• Quick insight: Top = 3:1 R:R, TP=4.5%, SL=1.5%, Z=2.2, CHOP≥55

Current Live Config (known negative PnL)

• TP=1.5%, SL=8%, Z_entry=1.8, Z_max=2.5
• NATR 0.75-2.5%, CHOP≥45, MaxSO=3
• This config is NOT in the sweep grid (SL 8% not tested — too wide for no-DCA)

Files

File	Purpose
step1_fetch_data.py	Fetch 30d klines → JSON cache
step2_sweep_z_tp_sl.py	Core Z×TP×SL sweep
backtest_core.py	Library: run_backtest(symbols, params)
download_data.py	Download → pickle cache (7d/14d)
sweep_runner.py	Full 8640 combo sweep
analyze_sweep.py	Step 7: analysis script
data_cache_7d.pkl	7-day cached data
data_cache_14d.pkl	14-day cached data
sweep_results_7d.json	8208 results from sweep