← Back to MagicMoneyMachine
Changelog
What's new in each release of MagicMoneyMachine. Older entries are kept as a historical record; they describe the product as it was then, not offers.
LATEST v3.63.352 Oct 21, 2026
Cloud and operator bot · v3.63.293 → v3.63.352, the public-relevant parts
A plain-English summary of the bot repo's own changelog since v3.63.292. Sales, sign-ups and
trials stay paused throughout. Nothing here is an offer, and nothing here is a performance claim.
Paused-sales and security hardening (v3.63.302 – v3.63.306)
- Web login no longer signs tokens with the key that encrypts users' exchange credentials (v3.63.302).
- A Coinbase API key that can move money out is refused at
/connect; before, “View + Trade only” was just instruction text (v3.63.303).
- The bot's content-security policy drops
'unsafe-eval' and unpkg (v3.63.304).
- The public
/api/signals feed no longer leaks coins on the do-not-trade list (v3.63.305).
- Marketing email turned off while sales are paused; lead-capture, newsletter and lead-nurture functions retired (v3.63.306). These edge-function changes ship on the operator's go, not on merge.
App redesign, sales behind one switch (v3.63.320, Oct 13)
- The bot app is down to five screens. Its Home screen answers “is it working”: book value, the scorecard against a weekly BTC buy, cash, and lane health.
- Every sales surface in the app (Lifetime banner, pricing, “Start Free”) sits behind one switch, which is off. Logged-out visitors see the public record and “new accounts are paused”.
Crash ladder stood down (v3.63.319, v3.63.322, Oct 13–14)
- The resting crash ladder failed its own replay (1.107 per $1 at 90 days vs 1.143 holding BTC) and the ledger stood it down. The allocation policy was armed instead.
The big subtraction (v3.63.325 and v3.63.326, Oct 15)
- About 30 systems deleted; the operator bot's main loop went from 75 sections to 33.
- Deleted: every stop exit (trailing, profit-lock, time, exchange-side), pair trading, CapitalRotation, the portfolio rebalancer, the autonomous allocator, smart accumulation, and the AI-brain chain.
- What is left: only ledger lanes buy (weekly BTC floor, DCA lane, crash-DCA bids, allocation policy, lab-promoted), and every position exits on its sell ladder. The eight CEX entry engines stay in shadow, with no measured edge.
- v3.63.327 then audited the cut and found about 18 of the ~30 systems had never been fairly tested. Anything that returns goes through a replay against a weekly BTC buy first.
The real book (v3.63.321, v3.63.352)
- The operator's own real Coinbase fills are published to the bot's database. v3.63.352 corrected 356 buy rows that had stored dollars as coin quantity (originals archived) and makes buys carry their order id, so each fill is attributed to the lane that placed it.
- The bot's own verdict: the real book's edge is not measurable yet from fills alone.
A 15% conviction reserve (v3.63.352, Oct 21)
- 15% of book value is held as cash. Automated buys refuse any order that would dip below it; a 6-hourly refill trims profitable alt slices (never BTC, never at a loss). Only the operator can spend it. Automated buys pause until it fills.
Also in this range
- Research on the operator's separate research machine: DEX paper practice books and replays. DEX trading stays research only; no DEX lane beats holding SOL, and its wallet is unfunded.
- Replays that said no: resting limit bids for BTC buys (+0.1% at best), per-coin dip depths (worse), a pullback ladder (failed out of sample, not authorized).
CLOUD v3.63.292 Oct 8, 2026
Cloud bot · release notes for v3.49.15 → v3.63.292 not yet published here
The hosted bot has shipped continuously since May; this page stopped being updated at
v3.49.14. Rather than leave that number looking current, this entry marked the real latest
version on Oct 8 and says plainly that the notes in between are missing. The desktop download is
still v3.49.14 (and downloads are paused).
STABLE · DESKTOP v3.49.14 May 8, 2026
Paid-tier repricing (historical)
A pricing change to the paid tiers. Nothing here is for sale: the project is private and
invite-only, and no tier, trial or payment exists. The sales detail is removed.
STABLE v3.49.6 May 8, 2026
Dashboard Audit · DEX Tier Gate · Native Cancel
Big polish pass on the user-facing dashboard at bot.magicmoneymachine.app — clearing every silent-error path and inconsistency the deep audit surfaced.
Native cancel-membership flow
- New
POST /api/user/subscription/cancel + /reactivate endpoints. One-click on the Account page; was previously a 4-tap Telegram detour.
- Cancel keeps full access until current_period_end. Nothing refunded mid-period.
- Free + Lifetime cleanly blocked (nothing to cancel). Comped users (no Stripe subscription) get a clean error pointing to support.
DEX tier gate
Server-side gating added to the DEX wallet, positions and trades endpoints (they were auth-only). DEX trading is research and paper only today, on no account.
Tier-limits drift swept across 10 sites
- Marketing site, account page, /upgrade Telegram copy, payment-confirmation message, license-redeem reveal, email drip — every place that named coin counts or capital caps was re-synced to the engine's TIER_LIMITS constant.
- Root: engine bumped to 25/75/75/300 coins long ago; UI surfaces lagged at 15/50/50/100. Paid users were undersold their own limits.
Silent errors surfaced
- Dashboard + Portfolio pages now show a red banner above the proof gate when the API can't be reached. Pre-v3.49.6 they silently rendered stale data.
- Chart oracle button now communicates "data unavailable" with reason instead of doing nothing.
- /api/license/tier no longer a hardcoded "demo" stub; returns the user's real tier + limits + subscription status.
STABLE v3.46.0 May 6, 2026
Profit Overhaul: Trend/Range Classifier + Counterfactual Engine + Phase 4 LIVE
The biggest fix: ZEC bug class is structurally closed
- Forensic April 29: bot bought ZEC 37 times at $325, sold all 37 at $325, lost $281 in fees while ZEC ran to $550+ (+69% trend)
- Root cause: mean-reversion engine fired on a trending coin → signal_reversal exits on every pullback → bot fought a 67% move and lost
- Fix:
trend_range_classifier.py — ADX + slope + ATR ratio per coin per cycle. MR engines blocked in TRENDING regimes; momentum blocked in RANGING. Wired into both local and cloud entry pipelines (LIVE).
Data-driven Fib targets (replaces theoretical constants)
mfe_mae_distribution.py computes per-coin MFE percentiles (p25/p50/p75/p90) from real closed trades
- HYPE peaks at 2.2% max — old +5% target was unreachable; now uses actual distribution
- FET caps at 3.6%; KNC at 4.9%; IRYS at 11.5% — every coin gets a ladder calibrated to its real behavior
- Per_coin_stats v2 schema migration (043) added engine, regime, mfe/mae percentile columns
Counterfactual replay (every doctrine change measured in dollars)
- New
counterfactual_engine.py replays closed trades under alternate rules
- Quantifies: "removing signal_reversal exits would have saved $553 in 30d"; "constant sizing would have made +$1,100 vs the dial chain"
- Frozen baseline contract: every Phase 4 activation must beat baseline or revert
- Weekly digest fires Mondays 8am UTC to operator Telegram
Phase 4 LIVE wiring (operator-approved May 6)
- unified_sizing LIVE — single additive formula replaces 3-dial multiplicative chain (DVOL × intel × aggressive). Bounded ±40%. Counterfactual: +$1,100/30d available.
- entry_score LIVE — single composite gate replaces 6-gate cascade (per_coin_confidence + dynamic_allowlist + engine_router + blocklist + ...)
- regime_gate LIVE — blocks MR in trends, momentum in ranges
- catalyst_sizing LIVE — 1.0×–1.2× boost on news/macro days via
catalyst_score
Architectural cleanup
- SWING lane retired (0/150 active positions on it; tripwire active)
- Dead ensemble code removed (0 cases in 30d, replaced by best_opportunity_ranker)
- Cosmic signals disabled by default (no correlation evidence)
not_buying_low reject key renamed to entry_above_dip_zone (clarity)
- Trailing stop widened ATR×2 → ATR×3 (HODL) / ATR×4 (HODL_EXTENDED)
hodl_invalidation_gate wired alert-only — fundamental thesis-broken triggers (BTC 200d break, NUPL capitulation, Pi-cycle, position drawdown >45%)
Doctrine update (CLAUDE.md)
- "Default to HODL. Ladder + Promote on runners. Don't panic."
- Pillar 1 (Capital Preservation) always wins on conflict — explicit examples documented
- HODL conviction threshold lowered 0.70 → 0.55 (most MR signals route to HODL not SWING)
- Runner-lock at MFE +30% force-promotes any position to HODL_EXTENDED ladder (+75/+100/+150)
USD/USDC routing fix (Option A — symmetric)
- Both BUY and SELL now prefer USDC pair (was: BUY tried USDC first, SELL tried USD first)
- Effect: operator's USDC stack is no longer drained over time
DEX fix
- HTTP 400 every minute since deploy (schema mismatch:
created_at vs entered_at) — fixed
- Bot can now re-load DEX positions across restarts
v3.45.0 May 6, 2026
PROFIT OVERHAUL Phases 1-3 — foundations + activations
- 4 new modules:
trend_range_classifier, mfe_mae_distribution, counterfactual_engine, best_opportunity_ranker
- 9 sections wiring data-driven Fib targets, regime gate, weekly counterfactual cron, schema migration v2
- 5 sections Phase 3 prep: catalyst_score module, cash-drag tracker, entry-gates audit, counterfactual baseline frozen
- 6 sections Phase 3 partial: 3 env-flag activations (MFE_LADDER, REGIME_GATE, CATALYST_SIZING) + shadow-mode wirings + dashboards
- 33 sections shipped in single day across phases 1+2+3 prep+3 partial
v3.44.x series May 5-6, 2026
Runner protection — stop fighting trends
- signal_reversal exits now require pnl<0 AND held >=2h (the ZEC bug fix)
- Runner-lock at MFE +30% force-promotes to HODL_EXTENDED (+75/+100/+150 ladder)
- HODL conviction threshold 0.70 → 0.55
- per_coin_stats filter signal_reversal exits (premature exits were polluting WR data, gating the bot from re-entering winners)
v3.27 → v3.43.x April-May 2026
Continuous improvement — 50+ versions
- Per-coin Fibonacci profile system (
coin_fib_profile.json)
- Engine_router with per-(engine, coin) edge scoring
- Regime overrides (per-regime config caps)
- Self-audit cron + daily digest
- Oracle integration → EnginePromoter (auto-promote shadow → live)
- Token unlocks manual override file
- Dynamic momentum allowlist (data-driven from per_coin_stats)
- Per-coin confidence floor (PF→entry confidence mapping)
- Intel sizing + dynamic blocklist
- A/B paper test infrastructure
- Trade attribution doctrine (
bot_decision / manual_user / unknown)
- BotConfig admin alerts on aggressive features
- For the full development history: the public journey timeline (the bot repo is private)
v3.26.2 April 29, 2026
Edge Discovery + Path-Independence Test + Fee-Sensitivity Analysis
Path-Independence Test (skill vs luck)
- New
scripts/path_independence_test.py runs Monte Carlo bootstrap (1000 samples WITH REPLACEMENT) on the last 60 days of trades
- Reports observed-return percentile rank within the bootstrap distribution → answers "are the bot's wins skill or luck?"
- Verdicts: STRONG SKILL (95th percentile+), MODEST EDGE (75-95th), MIXED, LIKELY LUCK, etc.
- Cron: every Sunday 05:00 UTC, logs to
logs/path_independence.log
Fee-Sensitivity Backtest Flag
continuous_backtest.py --fee-mult 1.5 simulates +50% fee shock
- Tests adversarial robustness — "does the edge survive if Coinbase doubles fees?"
- Honored via
MMM_BACKTEST_FEE_MULT env var read by simplified_engine.py
v3.26.1 April 29, 2026
UnifiedSniper Integration + Fleet Metrics + Edge-Weighted Sizing
Capital-Efficiency Weighted Sizing
- Edge-classified coins (BTC, XRP, RAVE, SOL, AVAX) now get +50% position size on entries
- Conservative cap: max 1.5× lift (capped from per_coin_stats up to 2.0×)
- MacroBottomEngine exempt to prevent compounding its already-2× sizing
BTC-HODL Benchmark + Per-Engine PF Table
- New
fleet_metrics block in /api/admin/auto_loop:
- Fleet 7d/30d net P&L vs BTC HODL %
- Per-engine PF/WR/net breakdown over 30 days
- Compounding rate vs frequency: trades/day + implied daily compound %
/loop Telegram command shows fleet 30d net + trades/day inline
UnifiedSniper Wired
- Sniper service now fires alongside the 10 strategy engines
- Cross-detector dedupe (5-min in-memory cache) prevents duplicate signals
v3.26.0 April 29, 2026
Five New Trading Engines + Sniper Consolidation + Auto-Discovery Starter
Five New High-Fit Engines (10 total now)
- BollingerSqueezeEngine — BB(20,2σ) bandwidth at 60-bar low + breakout + volume confirmation
- MAPullbackEngine — 50-EMA bounce in confirmed uptrend (50-EMA > 200-EMA, RSI > 40, ADX > 20)
- RSIDivergenceEngine — bullish divergence (price LL + RSI HL) — predicts MR signals one bar early
- LiquidityHuntEngine — 10-bar low pierce + immediate reversal + volume×1.5 (stop-hunt fade)
- MacroBottomEngine — NUPL < 0.30 + F&G < 25 + BTC > 200d-MA confluence (HIGH conviction, 2× sizing)
Strategic Redundancy Achieved
- Pre-v3.26.0:
mean_reversion was 86% of trades — single-strategy concentration risk
- Post-v3.26.0: 10 engines covering chop, breakout, trend-pullback, divergence, microstructure, macro
- EnginePromoter gates each at 50+ samples + 55% WR + PF 1.2 before live promotion
UnifiedSniper Service
- Consolidates 4 fragmented sniper modules behind one
scan() facade
- NewListingSniper (Coinbase) + VolumeSpikeSniper (cloud V2) + SolanaListingWatcher (opt-in)
- Cross-detector dedupe with 5-minute window + status reporting
Pattern Miner — Auto-Discovery Starter
scripts/pattern_miner.py reads last 60d of closed sells and computes lift for each (feature, bucket) pair
- Outputs HYPOTHESES (not auto-deployed engines) for operator review — "code an engine that does X"
- Cron: every Sunday 04:00 UTC
v3.25 series April 29, 2026
TradePlan Lanes + HODL Invalidation Gate + Edge Tiering
TradePlan Lanes (HODL / SWING / SCALP)
- Per-strategy hold times: HODL = no time stop, SWING = 4h, SCALP = 30m
- Blue chips (BTC, ETH, SOL, AVAX, LINK, AAVE, XRP) → HODL automatically
- Edge-tier non-blue-chips (RAVE, etc.) → HODL via per_coin_stats classification
- Mean-reversion on uncategorized non-blue → SCALP (bounded damage during learning)
HODL Invalidation Gate (Advisory)
- Five macro triggers per architectural plan Phase 1.3:
- BTC closed below 200d MA for 7+ consecutive days
- risk_regime = 'risk_off' for 5+ consecutive days
- NUPL fell from >0.5 to <0.25 within 14 days (capitulation)
- Pi-cycle top fired for 3+ consecutive days
- Per-position 45% drawdown in non-bull regime
- Default: advisory mode (Telegram
[REVIEW-REQUIRED] alert, NO auto-close)
- 14-day observation period required before any auto-exit flip
Trail Break-Even Floor
- Trailing stop now NEVER triggers a close below entry × 1.001
- Pre-fix: 26 of 26 trails over 7d closed at a loss because cushion (1 ATR) < trail width (1.8 ATR)
- SCALP positions now skip the trail entirely (target ladder + 30m time stop are sufficient)
Edge Discipline (with course-correction)
- Per-coin tier classification refresh every 6h (cron)
- Anti-edge gate is LIVE-MODE ONLY — paper users keep trading all coins to accumulate data
- Anti-edge watchlist deep backtest (60d) every Wednesday — looks for profitable regimes on currently-blacklisted coins
- Operator philosophy preserved: "buy low hold until profit — don't blacklist, find the profitable exit"
v3.24.4 + v3.24.5 April 29, 2026
Wash-Trade Prevention (Defense + Root Cause)
Wash-Trade Defense (consumer guards)
min_hold_minutes guard now applies to signal-reversal exits, not just standard exit loop
- Strategy-match guard: mean-reversion positions only close on mean-reversion sell signals (or generic
sell_signal) — never on contradictory momentum_breakout or vsa_breakout signals
- Combined: positions opened < 30 min ago can no longer be closed by an opposing-strategy signal
Wash-Trade Root Cause (signal coordination)
- Same-cycle contradiction filter in
shared_signals._compute_signals
- When mean-reversion fires BUY on a coin AND momentum_breakout fires SELL on the same coin in the same scan cycle, the SELL is filtered (BUY conviction wins)
- Prevents the contradiction from ever reaching the consumer — root cause closure
Forensic Data Driving the Fixes
- Pre-fix: 93 wash trades over 7 days fleet-wide, $477.51 in fee losses
- 11 users hit (every active live trader); ZEC alone = 60% of damage (37 washes, -$280.98)
- Conservative annualized recovery: ~$24,800/yr fleet-wide
v3.1.0 April 2026
Security Hardening + Exchange Protection + Risk Audit
Full Systems Security Audit
- Supabase RLS hardened — all 7 bot tables locked to service_role only (was open to anon key)
- Backup encryption upgraded — PBKDF2-SHA256 with 600K iterations (was single SHA-256 pass)
- JWT secret isolation — separate signing secret, no longer shared with Fernet master key
- FastAPI docs disabled — /docs, /redoc, /openapi.json removed from production
- CSP + Permissions-Policy headers — added to nginx config for XSS protection
- Stripe replay protection — 5-minute timestamp window on webhook signatures
- Error messages sanitized — 58 raw exception strings replaced with type names only
- Auth token masked in logs — no longer printed in full at startup
Exchange-Side Protection
- Fibonacci limit sells — GTC orders on Coinbase fill automatically even if bot is offline
- Crash DCA limit buys — -15%/-25%/-35% below entry for HODL positions, auto-merge on fill
- 2-tier DCA sizing — BTC/ETH: 50/40/30%, growth coins: 30/25/20% of initial cost
- 9 recovery layers — exchange orders, systemd, watchdog, remote commands, leader election
HODL Stop-Loss Corruption Fix
- 5 guards added to prevent stop_loss corruption on blue chip positions (BTC, ETH, SOL, AVAX, LINK, AAVE, XRP)
- Exit paths E (stale), F (profit lock), G (profit fade) now skip blue chips entirely
- Safety net resets corrupted stops before every save, startup repair at cycle 1
Risk Management Overhaul
- Sizing simplified — reduced from 8 multiplicative factors to 3 (conviction, TA gate, regime)
- Combined sizing clamped — 0.3x to 2.0x of base amount (was 0.006x to 14.2x)
- HODL risk budget fixed — blue chips use 35% stop for sizing (honest risk, ~4.7x smaller positions)
- MR stops widened — 1.5x to 2.0x ATR for mean reversion trades
- Aggregate entry cap — risk_off: 1 entry/cycle, caution: 2, default: 5
Diagnostics + Alerts
- Fear & Greed staleness alert after 6h cache age
- Regime transition Telegram alerts on market phase changes
- Position completion alerts with entry/exit prices and P&L
- Supabase consecutive failure counter with threshold alert
- WARNING severity now properly routed to Telegram (was silently dropped)
- Telegram 429 retry with exponential backoff
31 engines | 19 edge functions | 52 commands | 6 phases | 9 recovery layers
STABLE v2.3.4 April 2026
Channel Integration + Security Hardening + Email Automation
Telegram Channel (@MagicMoneyMachine)
- Public channel — automated phase change alerts, weekly digest cross-posts
- Admin bot channel control — post, pin, edit, delete, ban, update bio from Telegram
- Security — individual bots never post to channel (prevents strategy leaks + spam)
Email Automation Consolidated
- 5 email sequences in one edge function (delivery-retry v8): failed retries, onboarding drips, lead nurture, trial nurture, customer referral
- Supabase pg_cron: delivery-retry every 15min, weekly digest Mondays 9am, exec report Mondays 9am
Security
- RLS enabled on all 9 Supabase tables (was 5/9)
- Fixed SECURITY DEFINER view (performance_stats) → SECURITY INVOKER
- Channel posting: admin token only, never falls back to user's personal bot
- Stripe: archived 4 legacy products, new Growth payment link on correct product (31 engines)
Fixes
- Coinbase One fee detection (0.1% vs 0.6% overestimate)
- DCA duplicate position race condition (cycle-local set + position lock)
- Observer event dedup (fingerprint-based)
- Telegram rate limiter (2/min global, INFO 1/30s in rapid)
- Long message splitting (chunks on newlines, not truncation)
31 engines | 19 edge functions | 52 commands | 6 phases | 4 recovery layers
STABLE v2.3.0 April 2026
Strategy R&D Pipeline + Self-Improving Observer
New Strategy Engines
- VSA Breakout Engine — Detects institutional accumulation (smart money) via Volume Spread Analysis, enters before the breakout. Confirmed by OBV + ADX.
- Golden Hours Engine — Data shows 03-07 UTC has 94-100% win rate. This engine generates 2x conviction signals during optimal hours.
- Smart DCA — 3-tier inverse pyramid on blue chips: -5% (0.5x), -10% (1.0x), -15% (1.5x) from local high. Fibonacci exit targets from DCA average.
- Regime Switch Enforcement — Market phase (F&G + SMA200) now controls which engines run. Wrong-phase engines pushed to shadow mode.
Observer: Continuous Improvement Loop
- Runs every hour, reads event stream, generates actionable recommendations
- Detects: golden/dead hours, alpha/avoid coins, exit leaks, engine promotions, market opportunities
- Strategy R&D pipeline: Shadow → Evaluate → Promote → Demote (auto)
- Live Observer widget on landing page shows the bot thinking in real-time
Community & Lead Capture
- Download form now captures Name + Email (personalized delivery email)
- "Got an Idea?" community suggestions section with 3-layer bot detection
- Admin reviews all suggestions before any testing (safety gate)
Remote Recovery System
- 4-layer recovery: systemd → watchdog → remote commands → auto-reset
- Admin bot:
/remote resume works even when bot Telegram is down
- Supabase remote_commands table + edge function for out-of-band control
Fixes
- Pair trading alert spam: dedup per entry/exit
- DIP/BOUNCE/COIN alerts: 1-hour cooldown per alert type
- Health ping: hardcoded 12-cycle (not affected by rapid mode)
- Daily report: hardcoded 288-cycle (not 12x faster in rapid mode)
- Backtest v2: blue chip HODL strategy + tier comparison mode
31 engines | 16 edge functions | 49 commands | 5 phases | 4 recovery layers
STABLE v2.2.0 April 2026
Clean Data Pipeline + Observer System
- Unified event stream (events.jsonl) — every trade, decision, and error in one append-only log
- Position reconciliation — verifies bot state vs Coinbase account every hour
- Observer agent — detects patterns from event stream (hard stop clusters, drift, stale trades)
- /observe command shows real-time system health and anomalies
Blue Chip HODL Strategy
- No hard stops on BTC, ETH, SOL, AVAX, LINK, AAVE, XRP — hold through drawdowns
- Fibonacci sell targets (0.236, 0.382, 0.618, 1.0 extensions)
- Data proved: 95% of hard-stopped trades were green before the stop hit
- Hard stops cost $46.71 — more than all other exits combined earned
Profit Optimization
- Coin blacklist: negative P&L coins auto-skipped (30-day rolling window, coins get fresh chances)
- Correlation pause: all buys paused when cross-asset correlation > 0.80
- Pyramid into winners: add 50% to green positions (+2% threshold)
- Adaptive sizing: profitable coins get 1.5x, losing coins get 0.3x
Advanced Technical Analysis
- Volume Spread Analysis — detects smart money accumulation before price moves
- On-Balance Volume — bullish/bearish divergence detection
- ADX trend strength — per-coin (strong trend = wider trail, weak = mean reversion)
- DXY dollar index — crypto inverse correlation leading indicator
- /ta command for live analysis
29 Trading Engines
- 10 new engines since v2.0: Local Intelligence, Advanced TA, AI CEO, Coin Blacklist, Correlation Pause, Pyramid, Reconciliation, Observer, Signal Watcher, Volume Sniper
- Engine promotion cooldown: 200-signal minimum between mode changes (fixes spam loop)
- Event-driven AI: ~$1/year instead of $84/year (calls only on regime shifts, loss streaks, drawdown)
- /rapid mode: x12 local intelligence, zero API cost
STABLE v2.0.0 March 2026
Claude AI CEO Brain
- 3-layer AI assistant: hourly trading analyst, daily strategy director, daily business manager
- Auto-adjusts risk, screener weights, and engine modes based on Claude analysis
- AI self-improvement: tracks its own recommendation accuracy over time
- ~$7/month AI cost, fully transparent via /ai_cost command
Market Intelligence (10 Free Sources)
- Stablecoin supply tracking (capital flow signal from USDT + USDC)
- Persistent correlation matrix (crash detection — all coins moving together)
- Macro event calendar (FOMC, CPI, options expiry with risk multipliers)
- New Coinbase listing detection (alerts on new coins added)
- Execution verification (slippage tracking, orphaned order cleanup)
- Spread check gate (skip coins with wide bid-ask spreads)
Usage-Based Pricing (historical)
Tiers capped by capital and coin count, checked every cycle. Retired: nothing is sold today.
Collective Intelligence
- Network signal: aggregates all users' bot data into consensus
- Weekly performance digest emailed to all subscribers
- /network command shows collective intelligence in real time
Dev Mode + Intelligence Dashboard
- /dev toggles x12 speed on all AI/intelligence cycles
- Strategy tracker logs every intelligence loop decision
- /loops, /intel, /ai commands for live intelligence
- AI brain section in dashboard settings with live status
- Admin daily digest via separate admin bot
STABLE v1.7.1 March 2026
Elite Quant Practices
- ATR-based trailing stops: width adapts to each coin's volatility (BTC ~6%, AVAX ~14%)
- ATR-based profit targets: ladder at 0.5/1/1.5/3 ATR multiples instead of fixed percentages
- ATR-based initial stop: 3 ATR below entry, position sized by stop distance
- Breakeven lock: stop moves to entry after position gains 1 ATR (risk-free trade)
- Equal dollar risk per trade regardless of stop width
Learning Loop Fix
- Fixed critical bug: 191 of 218 signals couldn't resolve (untracked coin prices)
- 3-source price merge: WebSocket + market data cache + screener results
- Strategy engines now filter to 15 liquid coins only
- Weight optimizer now uses prediction-time dimension scores (not stale snapshots)
Intelligent Exits
- Scale-out promoted to LIVE (was shadow mode) — cleared the promotion gate on its shadow track record
- Signal reversal promoted to LIVE — real intelligence, not noise
- Regime-adaptive screener: bear market needs 0.70+ to BUY (was 0.62)
- Fixed prediction bias: 43 up / 0 down → now generates bearish signals in bear markets
Screener Optimization
- Weights updated from MFE data: trend 25%, regime 20%, fundamentals 5%
- Predictions save dimension scores at signal time for accurate correlation
STABLE v1.6.0 March 2026
Mean Reversion Gate
- Mean-reversion gate controls ALL entries — nothing bought without MR confirmation
- 280-day backtest proved: MR-gated +0.0% vs screener-only -15.2% vs BTC -35.7%
- MFE-optimized targets from 807 real entries
- Signal consensus (Meta-Brain): 3+ engines agree = 2x position size
- Kelly criterion half-Kelly position sizing from proven edge
5-Phase Market Deployment
- Automatic scaling: 10% capital in fear → 80% in euphoria → 20% in distribution
- Based on Fear & Greed index + BTC 200-day SMA position
- Replaces old risk_on/caution/risk_off with 5 granular phases
Pair Trading Engine
- Market-neutral Z-score arbitrage across 3 pairs (BTC/AVAX, LINK/AAVE, ETH/SOL)
- Fee gate: expected profit must exceed 2x trading fees
- Removed BTC/ETH pair (0.037% profit < 0.20% fee = net loss)
STABLE v1.5.0 March 2026
Business Infrastructure
- Health ping API: bots report status hourly to aggregated proof page
- Live performance page at magicmoneymachine.app/proof
- GitHub Actions release pipeline: tag → build → upload → deliver
- Auto-update checker: bots check for new versions every 24 hours
- Onboarding email drip: Day 1, 3, 7, 30 follow-ups
Coinbase Commerce
- Crypto payment webhook with full delivery pipeline
- Dynamic download URLs: always serves latest version from storage
STABLE v1.4.0 March 2026
Data-Driven Precision
- Sell targets optimized from 807 real trades: 1.5%, 3%, 5%, 10% (was 5-50% — unrealistic)
- Regime-specific engines: mean reversion only trades when proven (88% win rate)
- Regime transition detection catches market shifts 2-6 hours early
- Time-based exit: stale positions auto-exit after 6 hours
- Entry time filter: only trades during optimal market hours
Intelligence Dashboard
- 5 new Telegram commands: /accumulation, /fg, /engines, /fees, /report
- Live accumulation P&L panel on dashboard
- Engine accuracy table with rolling win rates
- Fear & Greed momentum tracking with transition alerts
Stability
- Fixed critical bug breaking exit engine every 7 minutes
- Fixed 9 API connection issues across the codebase
- Trimmed data files from 11MB to under 1MB
- Customer broadcast email system
STABLE v1.3.0 March 2026
Self-Learning Brain
- Weight optimizer auto-adjusts screener based on real trading results
- Mean reversion engine auto-promoted to LIVE (73% win rate proven in shadow)
- Prediction resolution tracks forecast accuracy over 24h windows
- Engine Brain learns which strategies work in which market conditions
Smart Accumulation Engine
- Buys during extreme fear (F&G < 25) on every -2% dip
- Scales with fear: lower fear index = bigger buys
- Tracks cost basis and average entry per coin
- Targets: BTC, ETH, SOL, AVAX, LINK, AAVE
Micro Scalp Engine
- Detects 30-minute price oscillation ranges
- Buys at support, sells at resistance, repeats
- Breakout detection switches to trend-hold mode
- Every cycle recorded for learning
Trade Intelligence
- Every alert explains WHY the trade was made
- Daily learning report emailed with engine stats, weights, predictions
- Coinbase fee calculator with all 10 volume tiers + Coinbase One support
- Restart resilience: bot resumes where it left off, no data loss
STABLE v1.2.1 March 2026
HODL Strategy Overhaul
- Trailing stop now activates at 2% profit (was 5% — never triggering)
- New profit-fade protection: auto-sells when position gives back over half its peak gain
- Simplified sell targets: fixed ladders at 5%, 10%, 20%, 35%, 50%
- Hard stops widened for disaster protection only (15% BTC/ETH, 12% alts)
- Signal reversal and time stop exits disabled — hold until targets hit
Smarter Entry Filters
- Zero new entries in risk-off markets (capital preservation mode)
- Downtrend filter blocks entries when moving averages are declining
- 24-hour re-entry cooldown after stop-outs
- Re-entry signal throttle: max 1 per coin per day
Bug Fixes
- Fixed license tier detection for legacy keys
- Fixed trading mode initialization on startup
- Tighter alt coin hard stops (2.5x ATR vs 3.0x)
STABLE v1.2.0 March 2026
License Tier Enforcement
- 3-tier system: Starter (paper only), Pro (live trading), Lifetime (permanent)
- Engine gating: Starter gets 6 core engines, Pro/Lifetime gets all 15
- Dashboard shows tier badge with gated sections for non-Pro users
- Telegram command access control by tier
- 3 layers of anti-bypass protection
Automated Sales
- Stripe payment integration with instant license delivery
- Automated email with license key and download link
- Telegram admin notifications on every sale
- Promo code support
STABLE v1.1.0 March 2026
Closed-Loop Learning System
- Strategy engines make explicit 24h price predictions
- Predictions auto-resolve against real prices
- Accuracy tracked per engine, per coin, per market regime
- Weight optimizer adjusts screener dimensions based on actual returns
Strategy Transformation
- 300s trading cycles (was 60s), 48-hour minimum hold time
- Fee-aware entry gate blocks trades where expected profit is less than 3x fees
- 12 curated liquid majors (was 99 coins including micro-caps)
- 40% cash reserve, $25 minimum buy
Production Readiness
- Landing page at magicmoneymachine.app
- License system with 3-tier validation
- PyInstaller standalone builds (no Python required)
- Telegram emergency controls: /killswitch, /reset, /resume
BETA v1.0.0 March 2026
Initial Release
- 21 trading engines with intelligent allocation
- 4 strategy engines: mean reversion, momentum, dip buyer, trend follower
- Engine Brain self-promotion system
- 7-type exit engine
- Real-time dashboard with 13 tabs
- Telegram bot with 27 commands