Dry-Run operations handbook

Dry-Run Operations Handbook — HonestTrend Family

This is the doc for your daily 5-minute glance. For the full technical analysis, see HONEST_TREND_REPORT.md and PHASE_B_FUTURES_SHORT.md.

2026-04-22 architecture update: Adopted Phase A (BTC→DCA, trend → ETH/BNB/SOL alts) + Phase B (futures L+S hedging). The pair list and port assignments below reflect the latest state.

---

1. Currently running bots (3 dry-runs in parallel)

Strategy	Config	Market	Direction	Pairs	Port
HonestTrend15mDry	config_dryrun_honest15m.json	spot	long-only	ETH/BNB/SOL	8082
HonestTrend1mMTF	config_dryrun_honest1mmtf.json	spot 1m+4h gate	long-only	ETH/BNB/SOL	8083
HonestTrendFutures	config_dryrun_honestfutures15m.json	futures	long+short	ETH/BNB/SOL	8084

All three share parameters:

• Parent class HonestTrendGeneric
• Simulated capital 10,000 USDT
• Max concurrent positions: 3
• Stake 1500 USDT/trade (up to 3900 after pyramid)
• SQLite DB user_data/tradesv3_<name>_dryrun.sqlite
• Log directory logs/<name>_dryrun_*.log

BTC is not in the trend pool — BTC alpha is accumulated via the Smart DCA dual-channel system (see the last section).

---

2. Strategy logic (signal rules)

Entry (buy only when all 6 conditions are met)

1. EMA golden cross: EMA(94, 15m) crosses above EMA(139, 15m)
   • 94 × 15m candles = 23.5 hours
   • 139 × 15m candles = 34.75 hours
   • Ratio 1.48x (not the classic 2x; Stage 1-3 validated this ratio as more robust)
2. +DI > −DI (trend direction confirmation)
3. ADX > 18 (sufficient trend strength)
4. Volume > past 24h average (volume confirmation)
5. Fear & Greed index FnG < 80 (the only data-validated sentiment defense rule)
6. Current bar volume > 0 (non-empty bar)

Exit (sell on any trigger)

1. EMA death cross: EMA(94) crosses below EMA(139)
   • Also requires holding time ≥ 12 hours (to avoid 15m noise triggers)
2. Manual force exit (Telegram /forcesell)
3. Stoploss:
   • Spot bots (15mDry / 1mMTF): no stoploss (stoploss = -0.99). Stage 1 proved tight stops wipe out the trend-following edge
   • Futures bot: −8% hard stoploss (futures can’t run naked like spot — must prevent liquidation)
4. Risk control: account-level drawdown kill-switch in risk_manager.py (15% pause / 20% retire)

Futures short exclusive rules

Only HonestTrendFutures goes short:

• Short entry: EMA death cross + minus_di > plus_di + ADX > 18 + FnG < 70
• FnG > 70 forbids shorts (don’t get squeezed during euphoria)
• FnG < 30 softly amplifies (panic-bottom shorts are highly efficient)
• Leverage 1x (no liquidation risk, pure directional alpha)
• Funding cost: roughly 4% annualized drag (included in backtest)

---

3. Expected metrics (what you’ll see next)

Based on 2024-01 to 2026-04 backtest (OOS segment 2025-07 to 2026-04):

Metric	Expected range
Trades per week	1–3 (approx)
Trades per month	4–10
Average hold per trade	2–3 days
Win rate	35–45% (typical 40%)
Average winning trade	+3% to +10%
Average losing trade	−1% to −4%
Worst single trade	−23% (historical extreme, 1% probability)
Longest losing streak	7 trades (historical extreme)
Monthly P&L variance	Very high — monthly losses are allowed; look at rolling 6-month PF

Typical trade flow after a trigger

[Telegram: Entry Fill] BTC/USDT at 93,250 USDT — 3,160 USDT stake (trend_15m tag)
    ↓
  (hold 2–3 days, EMA 94 stays above EMA 139)
    ↓
[Telegram: Exit Fill] BTC/USDT at 95,800 USDT — +2.7% profit (trend_exit tag)

---

4. Daily 5-minute checklist

Daily (1 minute):

• Scan Telegram — any 🚨 Risk State Change alert?
  • None → carry on. Yes → go straight to §7
• Any Entry/Exit notifications? Roughly matching expected frequency (1-3/week)?

Weekly (3 minutes):

# Check current status
python $PROJECT_DIR/scripts/risk_monitor.py status

# Or send /status to the bot on Telegram — it will reply with:
#   Current balance / Open trades / Total profit

Expected to see:

• Status: ACTIVE
• Drawdown: < 10% (normal); 10-15% → be alert; ≥15% → auto PAUSE
• Rolling PF (50 trades): > 1.2 mid/late; ≥1.0 early accumulation; <1.0 alert

Monthly (10 minutes):

• The 1st at 06:00 auto-runs walk-forward check; Telegram alerts if <50% of windows profitable
• Proactive comparison: live dry-run stats vs latest walk-forward backtest

  curl -s http://127.0.0.1:8082/api/v1/profit | jq

---

5. What messages will you see on Telegram

Freqtrade bot itself (sent by HonestTrend15mDry)

Trigger	Example
Startup	Exchange: binance / Timeframe: 15m / Strategy: HonestTrend15mDry
Entry (order placed)	Buying BTC/USDT at 93250 USDT, tag: trend
Entry filled	New trade filled BTC/USDT at 93250 USDT — 3160 USDT stake
Exit	Selling BTC/USDT at 95800 USDT — +2.7% profit, tag: trend_exit
Exit filled	Trade closed BTC/USDT: +2.7% (+86 USDT)
Warning	Dry run is enabled. All trades are simulated. (once at startup)
Protection triggered	A pair is cooled down (current strategy has no protection configured, so this won’t fire)

Interactive commands (send to the bot):

• /status — current open positions
• /profit — cumulative P&L
• /daily / /weekly / /monthly — periodic stats
• /trades — recent trades list
• /performance — per-pair performance
• /balance — simulated wallet balance
• /forceexit <trade_id> — force-close a trade (works in dry-run too)
• /stopentry — pause new entries (does not affect existing positions)
• /help — full command list

Risk Monitor (sent by scripts/risk_monitor.py every 4h)

Only sends on state change; won’t spam under normal conditions.

Trigger	Message style
DD first reaches ≥ 15%	🚨 Risk State Change: ACTIVE → PAUSED. Drawdown 15.2% ≥ 15%...
DD recovers to <10%	🚨 Risk State Change: PAUSED → ACTIVE. Drawdown recovered to 9.0%...
DD ≥ 20%	🚨 Risk State Change: ACTIVE → RETIRED. Drawdown 21.3% ≥ 20%. Manual reset required.
6 months + PF<1.2	🚨 Risk State Change: ACTIVE → RETIRED. After 180d live, rolling PF 1.10 < 1.20.
Manual pause/retire	⏸ Manual PAUSE / 🛑 Manual RETIRE

Walk-Forward (1st of each month 06:00, scripts/walk_forward_check.py)

Health	Message
≥50% windows profitable	No alert (silent pass)
<50% windows profitable	⚠️ EDGE DEGRADING: <50% windows profitable
0/4 windows profitable	🚨 EDGE DEAD: 0 windows profitable. Consider retiring.

KOL Alerts (independent of this strategy, strategies/telegram_alerts.py)

Runs every 30 minutes. New fix: every alert is now a clickable link → jumps to the original Google News article for verification.

🟢 TRUMP (+0.50)
[Trump endorses crypto at rally](https://news.google.com/rss/articles/...)

---

6. What counts as “normal”?

✅ All of these are normal (don’t panic)

• No trades for 1-2 weeks in a row — EMAs haven’t crossed; market didn’t give a signal. The strategy is designed to wait for good opportunities.
• BTC crashes 10% one day but the bot does nothing — only EMA death cross triggers exit. No stoploss is intentional.
• Single trade loses 5-10% — within the distribution, CVaR 95% = −9.13%
• A whole month with net loss — normal; look at rolling 6-month/yearly
• No entries at all when FnG > 80 — that’s the rule; extreme greed phases are more prone to drawdown
• Moves with BTC but smaller magnitude — strategy is beta-biased, not an all-weather hedge

⚠️ These need attention (not necessarily immediate action)

Observation	Reasonable explanation / action
5 consecutive losses	Within normal range (historical max 7 in a row); keep watching
DD 10-15%	Near PAUSE threshold; check risk_monitor.py status, don’t intervene
A whole month with no trades	FnG has been > 80 for a long time, or ADX < 18. Check Fear & Greed and BTC trend
A position held > 10 days without movement	Very rare; check EMA values
Slippage > 0.1% (after live startup)	Order-book depth insufficient; consider reducing the pair count

🚨 These are red lines (must investigate immediately)

Observation	Action
Telegram received a RETIRED alert	Strategy has auto-retired — do not directly reset; investigate the cause first
DD ≥ 15% and PAUSED	Wait for auto-resume (DD recovers to <10%); do not manually reset
Single trade loses >25% (beyond historical worst)	There’s a black swan event; manually pause and check the market
2+ consecutive months of walk-forward alerts	Edge is decaying; consider retiring
Bot process gone (pgrep returns nothing)	Check logs, restart the bot, check journald for crash reason
Live vs dry-run P&L gap >3%/month	Slippage/latency larger than expected; consider reducing size or switching strategy

---

7. How to intervene (emergency handbook)

Pause new entries (does not affect existing positions)

# Method 1: via risk manager (recommended, sends Telegram alert)
cd $PROJECT_DIR
python scripts/risk_monitor.py pause --note "your reason"

# Method 2: via Telegram command
# Send to the bot: /stopentry

Full retirement (requires manual reset to recover)

python scripts/risk_monitor.py retire --note "your reason"

Reset to ACTIVE (use with caution)

python scripts/risk_monitor.py reset --note "your reason"

Stop the bot process

pkill -TERM -f "config_dryrun_honest15m"
# Wait 5-10 seconds for full shutdown
pgrep -af "config_dryrun_honest15m"  # confirm exit

Restart the bot

cd $PROJECT_DIR
./scripts/start_honest_trend.sh dryrun

Force-close a specific trade

# Send on Telegram: /forceexit <trade_id>
# Or curl the API:
curl -X POST http://127.0.0.1:8082/api/v1/forceexit \
  -H "Content-Type: application/json" \
  -d '{"tradeid": "1"}'

View real-time logs

tail -f $PROJECT_DIR/logs/honest15m_dryrun_*.log

---

8. Transitioning from dry-run to live (roadmap)

Don’t rush to go live. Let dry-run run for at least 2-4 weeks to validate the following conditions:

Transition checklist

• Dry-run has accumulated ≥ 20 trades (statistical significance)
• Rolling PF ≥ 1.2
• Max DD < 12%
• No RETIRED events
• No “live vs dry-run” deviation (dry-run can only be compared to backtest)
• Monthly walk-forward check has passed at least once with 4/4

Transition steps

1. Add Binance API key to SOPS (see deployment doc)
2. Launch live: ./scripts/start_honest_trend.sh live
3. Keep dry-run running in parallel — 15m dry-run as shadow validator to compare latency/slippage
4. Live first month: only 1,000 USDT (don’t start with 10K)
5. Use risk_monitor monthly; gradually scale to 10K once stable

---

9. Common self-Q&A

Q: Why no entries after a week? A: EMA 94/139 needs a very clear uptrend. Look at the BTC 15m chart — EMA 94 and 139 need to cross from below to above. If the market is ranging/falling, no entry. That’s a feature, not a bug.

Q: How to check the current FnG? A: Dashboard http://localhost:3000 shows it in real time. Or curl https://api.alternative.me/fng/?limit=1. When current ≥ 80, the bot does not open any positions.

Q: How big will the gap between dry-run and backtest be? A: Theoretically very small (both use the same market data). Significant differences only appear in live (real fill latency, order-book depth, fee tiering). Dry-run mainly validates strategy logic bug-free, not P&L accuracy.

Q: When to start 1m live? A: Your call. Recommend starting after dry-run has run 2-4 weeks and confirmed stable. Command is ready: ./scripts/start_honest_trend.sh live (prerequisite: SOPS has Binance key).

Q: What if dry-run keeps losing money? A: Depends on the type of loss:

• Single-month loss, 6-month PF still >1.2 → normal, no action.
• 6-month PF < 1.0, trades > 50 → do not launch live. Consider retiring or switching strategy.
• DD > 15% → risk_monitor auto-pauses; check state, do not manually reset.

Q: Can I run multiple strategies in parallel? A: Yes. Currently 3 bots in parallel (15m spot 8082 / 1m MTF spot 8083 / 15m futures L+S 8084) + 2 daemons (event_reactor + event_dca_bot). Each bot has an independent port + independent SQLite DB. Telegram shares one chat (each bot distinguishes its own messages via bot_name).

Q: Why ETH/BNB/SOL? Why not BTC? A: Phase A (2026-04-22) adopted: BTC uses DCA accumulation (weekly + event), not in the trend pool. Reason: BTC’s return on trend strategy is below its own DCA baseline, and the user is long-term bullish on BTC (DCA accumulation fits better). ETH/BNB/SOL are high-beta alts and the trend strategy performs better there.

Q: SOL was said to be unsuitable before — why add it back now? A: The previous conclusion was under the old framework (BTC-dominant). After Phase A switched to spot alts-only, re-running walk-forward: SOL as a high-beta alt captured more alpha (W4 2021 +143%). Also, futures L+S can hedge SOL’s downside risk (W5 2022 LUNA: SOL spot lost 17%, futures L+S instead earned 48%).

---

10. Related file index

Main docs
├── docs/HONEST_TREND_REPORT.md     full technical analysis (why this strategy was chosen)
├── docs/DRYRUN_HANDBOOK.md         ← the one you're reading (daily ops)
├── risk_state.json                 current risk manager state (generated at runtime)
├── walk_forward_history/           monthly walk-forward archive

Active strategies
├── strategies/HonestTrendGeneric.py    base class (all HonestTrend* inherit)
├── strategies/HonestTrend15mDry.py     15m spot long-only (port 8082)
├── strategies/HonestTrend1mMTF.py      1m MTF spot long-only (port 8083)
├── strategies/HonestTrendFutures.py    15m futures L+S (port 8084, Phase B)
├── strategies/HonestTrend1mLive.py     1m live (old pair list, pending cut to A5)
├── strategies/dca_executor.py          DCA execution engine (shared by weekly/event)
├── strategies/event_dca_bot.py         event-driven DCA daemon (WebSocket)
├── strategies/risk_manager.py          state machine
├── strategies/telegram_alerts.py       alerts module

Config
├── configs/config_dryrun_honest15m.json           A5 spot long
├── configs/config_dryrun_honest1mmtf.json         A5 spot 1m MTF
├── configs/config_dryrun_honestfutures15m.json    Phase B futures L+S
├── configs/config_live_honest1m.json              old live (not cut to A5)
├── configs/backtest/config_backtest_15m_alts_a5.json     A5 backtest
├── configs/backtest/config_backtest_15m_futures_a5.json  Phase B backtest
├── configs/backtest/config_backtest_*.json        others

Scripts
├── scripts/start_honest_trend.sh       startup script
├── scripts/risk_monitor.py             monitor + CLI
├── scripts/walk_forward_check.py       monthly validation

Systemd timers (every 4h / monthly)
├── ~/.config/systemd/user/crypto-risk-monitor.{service,timer}
├── ~/.config/systemd/user/crypto-walkforward.{service,timer}

Archive (don't touch, and already .gitignored out of the public repo)
├── docs/RETIRED_STRATEGIES.md          retired strategies record (code deleted, only reasons + lessons remain)

---

11. Smart DCA dual-channel (BTC exclusive)

The trend strategy doesn’t touch BTC; BTC is accumulated by this DCA combo:

Weekly DCA (time-driven)

• Service: crypto-dca.timer → dca_executor.py
• Frequency: every Monday 00:00 SGT
• Amount: $500 base × multiplier (0.0-3.0), weighted by FnG + cycle factors + KOL
• Status: dry-run (DCA_LIVE_ENABLED not set)
• Logs: journalctl --user -u crypto-dca.service -f

Event-driven (WebSocket daemon)

• Service: crypto-event-dca.service → event_dca_bot.py
• Type: always-on, listens to Binance aggTrade WebSocket
• Triggers: 4-tier signals (FLASH 1m>3% / FAST 5m>5% / SUSTAIN 24h>10% / CAPITUL 30d DD>25%)
• Budget: $2000 reserve per month, $700/trigger (FnG<20 ×1.5, >60 ×0.7)
• Limits: 72h cooldown, max 3 per month
• State file: event_dca_state.json
• Self-test: python strategies/event_dca_bot.py --self-test
• Details: EVENT_DCA.md

The two channels have independent budgets (not shared); both invoke dca_executor.py to execute. To enable live: add DCA_LIVE_ENABLED=true + Binance spot API key in SOPS secrets.env, and both channels go live simultaneously.

---

12. Core principles (one-liner edition)

1. Dry-run to accumulate samples; don’t rush to live
2. Daily glance Telegram, weekly check status, monthly check walk-forward
3. The rules are encoded; don’t hand-tune
4. When red lines fire, trust the state machine — don’t rush to reset
5. Monthly loss is normal; 6-month PF < 1.2 is the real alarm

Remember Taleb’s words: strategies don’t die slowly — they fail off a cliff. Write the kill-switch ahead of time; when the cliff comes, the kill-switch saves you, not your judgment.

---

Last updated: 2026-04-20. When something goes wrong, read this doc first, then the background in HONEST_TREND_REPORT.md.