Action Plan · Living Document
DCC Trading Governor Checklist
Every step from overnight results to live trading. Check off as we go. Updated each session.
2 / 10
Phases Done
P1
Current Phase
~40
Total Tasks
+4.16%
Best Edge So Far
Phase 0
Foundation — Build the Engine
✓
v0.1: Basic MDL arena with 7 generators
Direction prediction on single TF. DCC sensor (6-bit, 64-slot). Checkpoint/resume. CSV streaming.
BD_MDL_DCC_Predictor.py (single file, 1060 lines)
Result: 50.01% accuracy on 7,009 bars BTC 15m. REVERT-MU +0.88%. DCC pegged at u=1.0 (bug).
✓
v0.2: Fix DCC sensor bugs
Identify: time-based DCC never fires in fast backtest. TREND-S poisons at -4.6%. Thompson too slow to learn.
Result: Diagnosis complete. u=1.0 frozen. Need bar-based updates.
✓
v0.3: Governor architecture
Split into BD_DCC_engine.py (computation) + BD_MDL_DCC_Predictor.py (CLI). Vectorized numpy (1000x speedup). Thompson sampling with decay. 8-bit symbols, 128-slot buffer. Bar-based DCC and escalation. Dynamic window. Contrarian mode. 8 generators.
BD_DCC_engine.py (634 lines) + BD_MDL_DCC_Predictor.py (367 lines) + BD_MDL_DCC_Test.bat
Result: 2000 bars/s. DCC moves (u: 10→18). Escalation fires. 2 nuclear resets on real data. Contrarian on 15m: +0.54%.
✓
Governor architecture paper v1.0
12-chapter paper: 5 actuators, S-metric, full auto mode, CCH bridge.
BD_8Z_DCC_Trading_Governor.html (v1.0, 674 lines)
Phase 0.5
Overnight Validation — 924K Bars
✓
Build overnight multi-TF analysis suite
Load 1m BTC data (924K bars, 21 months). Resample to 12 TFs. Run NORMAL + CONTRARIAN on each = 24 runs. Cross-TF coherence. 2-TF combo search.
BD_MDL_DCC_Overnight.py + BD_MDL_DCC_Overnight.bat
✓
Run overnight and analyze results
4/12 TFs positive contrarian (8h +0.68%, 15m +0.54%, 30m +0.52%, 4h +0.23%). 0/12 normal positive. 10m+1m combo = +4.16% across 3,355 signals. MM contrarian thesis validated.
✓
Fee impact analysis
Per-trade EV = 0.02%. Any fee above zero kills standalone edge. MEXC zero-fee = +40%/yr (1x), +400%/yr (10x). Or use as ZZ/SM confirmation layer (fees don't matter).
✓
Governor paper v2.0
17 chapters. Added: contrarian discovery, overnight results, fee analysis, search space, three-engine architecture, updated phases.
BD_8Z_DCC_Trading_Governor.html (v2.0, 621 lines)
Gate: P0.5 → P1
✓ Contrarian edge exists on real data. ✓ Multi-TF combos amplify signal. ✓ 1m appears in all winning combos. Proceed to fine-grained search.
Phase 1
L1→L2 Zoom Scanner — Find the Optimal TF Combo
▶
Build BD_MDL_DCC_Scanner.py
Single script: L1 broad scan + L2 auto-zoom. Reuses BD_DCC_engine.py. Reads existing 1m data for BTC/ETH/SOL.
BD_MDL_DCC_Scanner.py + BD_MDL_DCC_Scanner.bat
○
L1: Broad scan — 30 TFs × 3 assets × contrarian
TFs: 1m, 2m, 3m, 4m, 5m, 7m, 8m, 10m, 12m, 15m, 18m, 20m, 25m, 30m, 35m, 40m, 45m, 1h, 75m, 90m, 2h, 3h, 4h, 5h, 6h, 8h, 10h, 12h, 16h, 1D. Assets: BTC, ETH, SOL. Total: 90 runs. Expect ~30 min.
Depends: ETH + SOL 1m data files in data/ folder
○
L1 analysis: rank all 90 single-TF results by edge
Find which TFs and which assets have positive contrarian edge. Do ETH and SOL show same pattern as BTC? Is 15m–30m the universal sweet spot or BTC-specific?
○
L1 combos: test all 2-TF pairs from L1 winners
Take top 10 single-TF results. Test all 45 pairwise combinations. Also test top cross-asset combos (BTC 10m + ETH 5m, etc).
○
L2: Zoom around winners — ±2 min steps
If 10m+1m is best: test 8m+1m, 9m+1m, 10m+1m, 11m+1m, 12m+1m. Also 10m+2m, 10m+3m. Find the peak. Maybe 9m+2m at +5.1% is better. Auto-zoom: script picks top 5 combos and generates zoom grid automatically.
○
L2 results: identify THE combo
The best TF × asset × combo with highest edge AND sufficient signal count (>1000 signals).
Gate: P1 → P1s
Need: best combo identified with >+2% edge on >1000 signals across >12 months data. If no combo clears this bar, expand generator catalogue (jump to P2) before continuing.
Phase 1s
Seconds-Scale Analysis
○
Run engine on 1s BTC data
86,400 bars per day. Resample to 2s, 3s, 5s, 10s, 15s, 30s. Run contrarian on each. MM stop hunts are 1-second events — DCC might see them as highly compressible templates.
Depends: BTCUSDT_1s data files (already have 1 day in data/)
○
Test seconds + minutes cross-scale combos
1s+10m, 2s+15m, 5s+30m, etc. Does adding seconds-scale confirmation to the minute-scale signal amplify edge? The MM propagation delay (stop hunt starts in 1s, plays out over 1m–5m) should be visible here.
○
Fetch more 1s data if needed
Currently 1 day of 1s data. May need 7–30 days for statistical significance at seconds scale.
Gate: P1s → P1x
Need: at least one seconds-scale combo with >+1% edge, OR determination that seconds scale adds nothing (also useful — focus on minutes).
Phase 1x
Cross-Asset Combinations
○
Run L1 scan on ETH and SOL (if not done in P1)
Same 30 TFs, contrarian mode. Do alts show same contrarian pattern as BTC? Different TFs might dominate (ETH more volatile, SOL more manipulated).
○
Cross-asset combo search
BTC 10m + ETH 5m. BTC 1m + SOL 15m. All pairwise cross-asset combos from L1 winners. MMs move correlated assets together — BTC dump triggers ETH/SOL dumps 2–5 seconds later. Cross-asset coherence detects the propagation delay.
○
3-TF combo search (if 2-TF results warrant)
Maybe BTC 1m + BTC 10m + ETH 5m is the ultimate signal. Combinatorial explosion: use L1→L2 zoom, not brute force. Only test 3-TF combos built from top-10 single results.
Gate: P1x → P2
Need: final “best known combo” with edge, signal count, and robustness documented. This becomes the target configuration for the live system.
Phase 2
Generator Expansion + Walk-Forward
○
Add VOL-ABSORB generator
Big volume + small candle = absorption = reversal. Uses volume data already in CSV. Competes in MDL arena alongside existing generators. Does not duplicate SM logic (DCC uses it for direction prediction, SM uses it for trade gating).
○
Add STOP-HUNT generator
Detect sweep below recent low / above recent high followed by reversal. Classic MM pattern. Should compress well because it repeats on a template.
○
Add SESSION generator
Time-of-day pattern: Asian range, London breakout, US session. Encode hour-of-day as part of the prediction signal. Requires timestamp in data.
○
Multi-source fetcher upgrade
Download volume from Binance + Bybit + OKX + MEXC simultaneously. Sum volumes. Cross-exchange volume divergence is itself an MM signal.
○
L3: Walk-forward robustness
Take top 10 combos. Test on Jun–Dec 2024 separately from Jan–Mar 2026. Does 10m+1m work in BOTH periods? This is the overfitting test. If it only works in one period, the edge is spurious.
Gate: P2 → P3
Need: best combo passes walk-forward (positive edge in both sub-periods). New generators improve edge. If walk-forward fails, the edge is overfit — return to P1 with expanded generators.
Phase 3
Multi-TF Governor — Real-Time System
○
Build real-time multi-TF arena
Run 2–4 arenas simultaneously on live WebSocket data. Each arena on its TF. Governor computes cross-TF coherence in real time. Outputs: direction, confidence, coherence score, S-metric.
○
Live signal logging (no trading)
Run for 7–14 days. Log every signal with timestamp. Compare to what actually happened. Is live accuracy consistent with backtest accuracy?
○
Signal quality monitor implementation
ACTIVE/DARK mode switching. Track rolling accuracy. Auto-disable when no edge detected. Auto-re-enable when structure returns.
Phase 4
ZZ + SM + DCC Hybrid
○
DCC as ZZ confirmation filter
Boolean signal: “DCC agrees with ZZ entry direction” or “DCC says MM trap, reduce size.” ZZ makes the trading decision. DCC filters. Measure: does DCC improve ZZ win rate?
○
DCC as ZZ breakout detector
When ZZ says leg exceeds average length (potential breakout), DCC says whether this is a real breakout or MM fake-out. Test: DCC contrarian on breakout bars vs non-breakout bars.
○
SM volume confirmation layer
SM detects absorption (big vol, small candle) or momentum confirmation (big vol, big candle). DCC detects regime + direction. ZZ detects structure. Triple agreement = max size. Any disagreement = reduce.
○
Backtest the hybrid on ZZ paper trader data
Run ZZ paper trader with and without DCC filter on same data. Measure: PnL improvement, drawdown reduction, win rate change.
Phase 5
MEXC Zero-Fee Standalone Bot
○
Playwright browser automation for MEXC
No API (retail-locked). Browser automation opens/closes positions on MEXC zero-fee perpetuals. DCC signal drives entries directly.
Depends: P3 (validated live signals)
○
Paper trade on MEXC with real UI, fake money
Run the bot for 14 days. Monitor: fills, slippage, latency, position management. Compare to backtest expectations.
○
Live with minimum capital ($100–$200)
Real money, real fills, real slippage. Start at 5x leverage. Measure: actual vs expected PnL. If positive after 30 days, scale up.
Phase 6
Full Auto Mode
○
Single CLI command: auto mode
python BD_MDL_DCC_Governor.py auto BTCUSDT --max-hours 24 — DCC decides timeframes, generators, windows, thresholds, sizing, when to trade, when to stop.○
HTML live paper trader
Same design system as ZZ and SM paper traders. WebSocket feeds. Real-time equity curve, trade log, DCC dashboard. Multi-TF coherence visualization.
○
Compounding + skimming protocol
Auto-skim at 2x equity. Cold wallet transfers. System runs with minimum human oversight. The endgame.
Reference
Dependency Map
P0Engine built
→
P0.5Overnight validated
→
P1L1→L2 zoom
P1Best minutes combo
→
P1sSeconds scale
→
P1xCross-asset
P1xFinal combo
→
P2Generators + L3
→
P3Live multi-TF
P3Validated live
→
P4Hybrid ZZ+SM+DCC
P3Validated live
→
P5MEXC bot
→
P6Full auto
P4 and P5 are parallel paths after P3. P4 (hybrid) doesn’t need zero fees because DCC is a filter, not a standalone trader. P5 (MEXC bot) is the standalone zero-fee path. Both lead to P6 (full auto) where the hybrid runs on MEXC with zero fees for maximum edge.
Key files across all phases
BD_DCC_engine.py — Pure computation. Used by everything. Never touches I/O.
BD_MDL_DCC_Predictor.py — Single-TF CLI with resume/logging. Used for testing.
BD_MDL_DCC_Overnight.py — Multi-TF batch analysis. P0.5 tool.
BD_MDL_DCC_Scanner.py — L1→L2 zoom search. P1 tool. (to build)
BD_MDL_DCC_Governor.py — Real-time multi-TF governor. P3 tool. (to build)
BD_8Z_DCC_Trading_Governor.html — Architecture paper. Updated each phase.
BD_8Z_DCC_ActionPlan.html — This checklist. Updated each session.