How every published signal is scored, then the strategy the paper record tests: what the bot does, what it no longer does, and what the evidence says. Updated Oct 29, 2026. Member accounts are paper only; access is by invitation.
On the operator's own machine, only ledger lanes buy (weekly floor, DCA lane, crash-DCA bids, allocation policy, lab-promoted), and every position exits on its ladder.
The scorecard: beat a weekly BTC buy, net of real fees, or say we do not. A September 2026 audit found that the book's return was roughly BTC times its crypto share: no alpha either way.
Where it runs. Real money trades on one account: the operator's own Coinbase account. The cloud bot at bot.magicmoneymachine.app runs a paper fleet, published at the public paper ledger. Hosted accounts, sign-ups, trials and desktop downloads have been paused since Sep 5, 2026 (why).
Nothing buys unless it has a named owner in the bot's authorization ledger, a written record of which lanes the operator has approved to spend money. As of v3.63.352 those lanes are:
| Lane | What it does |
|---|---|
| Weekly BTC floor | A fixed BTC buy every Monday, in every market regime. This is also the benchmark the bot is scored against. |
| DCA lane | A fear leg (buys when the Fear & Greed index is low) and a dip leg, which buys BTC only since v3.63.349. |
| Crash-DCA bids | Resting buy orders below entry (−15 / −25 / −35%) on coins already held, so a crash adds to a position instead of selling it. |
| Allocation policy | Moves the book toward a BTC target in small steps (at most $50 per 6-hour pass). The target leans into BTC when BTC is above its 200-day average and toward cash when it is below. Since v3.63.349 it never sells an alt at a loss to fund a BTC buy. |
| Lab-promoted | Lets a research engine trade only after the operator names it in the ledger. As of Oct 2026 none is named, so this lane buys nothing. |
15% of book value is held as free cash. Automated buys refuse any order that would dip below it, and a 6-hourly refill trims profitable alt slices (never BTC, never at a loss) to rebuild it. Only the operator can spend it, one coin at a time, and those positions get a wide ladder. Until the reserve fills, automated buys pause.
Every position exits on its sell ladder. The bot does not try to pick a top: it sells slices at preset profit levels.
| Rung | Above entry | Slice sold |
|---|---|---|
| 1 | +5% | 20% |
| 2 | +10% | 25% |
| 3 | +20% | 25% |
| 4 | +35% | 20% |
| 5 | +50% | 10% |
Runners: a position that runs far enough is promoted to a wider ladder (+75 / +100 / +150%) with a small tail left open. BTC is never sold by the bot.
The trailing stop, the profit-lock trail, time stops and exchange-side stop orders are gone. Each had a losing record: hard stops lost $81 net on 56 of 92 trades, and a conditional stop sold the bottom 15 times out of 16. The cost is real: a position can fall a long way and simply be held.
Honest caveat on the ladder itself: in the bot's own replays the +5…+50% ladder was the worst exit at 90 and 180 days in three replays (v3.63.327). Wider exits are being replayed. The ladder is what runs today, not a proven edge.
The eight CEX entry engines (mean reversion, VSA breakout, momentum breakout, pump fade, RSI divergence, Bollinger squeeze, orderbook imbalance, funding skew) run in shadow: they log signals and place no orders.
No entry engine has a proven edge. The walk-forward test found about zero gross edge over 13,045 signals, −9.8% per position, and 0 of 4 walk-forward quarters positive. That is why they stay in shadow. The mean-reversion readings on the home page are shadow readings, not trading signals.
A rule reaches real money one way: a replay shows it beats the weekly BTC buy net of real fees, out of sample, and the operator writes it into the ledger.
On Oct 15, 2026, v3.63.325 and v3.63.326 deleted about 30 systems and cut the bot's main loop from 75 sections to 33. Among them:
A follow-up audit (v3.63.327) found that about 18 of the ~30 had never been fairly tested, so some may come back, but only through the same replay-then-ledger route.
Never built: maker-first orders (try a limit price, then fall back to market). Earlier versions of this site described them as running; they were researched, not built.
From the evidence table in the bot repo (Oct 2026). These are replays, not live results, and not a forecast.
| Question | Verdict |
|---|---|
| CEX entry engines | About zero gross edge (13,045 signals); −9.8% per position; 0 of 4 walk-forward quarters |
| Weekly BTC buy vs fear-only DCA | The weekly buy wins per $1 at 365 days (1.778 vs 1.725); fear works as a size multiplier, not a better rule |
| Resting crash ladder (−15 / −22 / −30%) | Fails: 1.107 per $1 at 90 days vs 1.143 holding BTC. Stood down Oct 13, 2026 |
| Buying fear (Fear & Greed ≤ 35 / ≤ 20) | Worse than buying on an ordinary day, in both halves of the data |
| Weekly rotation into alts | Every rule lost to holding BTC out of sample |
| DEX launch sniping | Negative per $10 bet; no DEX paper book beats holding SOL |
Earlier versions of this page showed a crash-DCA backtest (+40.48%, 93.6% win rate). It is removed: the resting crash ladder failed its own replay and was stood down.
| Rail | What it does |
|---|---|
| Circuit breaker | Pauses new buys on a portfolio drawdown; sells still run |
| Portfolio risk-off | A risk exit that stays after the stops were deleted |
| Blacklist | A never-enter list of coins |
| Authorization ledger | No lane spends money without a named owner and the operator's written approval |
| Conviction reserve | 15% of book held as cash that automated buys cannot touch (v3.63.352) |
| Key scope check | A Coinbase key that can transfer funds out is refused at connect (v3.63.303) |
Funds stay on the Coinbase account; the bot never takes custody.
The DEX work runs on the operator's research machine as paper practice books. As of Oct 2026 no DEX lane beats holding SOL, the research wallet is unfunded, and DEX trading is not authorized for any account. Earlier versions of this site sold DEX as live on paid tiers; that was wrong, and those tiers are not for sale.
The paper fleet's fills are published as they happen. Access is private and by invitation; members sign in.
Public paper ledger