PaperBot

Autonomous · BTC and ETH · daily candles · live trading on your own exchange account, in preparation

No trade is the default.

PaperBot is a crypto trading system built to preserve capital first, avoid liquidation second, and return something third. It sizes every position by risk, refuses more trades than it takes, and writes every refusal down. Its public record runs on live Binance prices today, and you can watch it decide. Live trading on your own Binance, OKX or Bybit account is in preparation.

No return claims here, ever. The record below is a paper account; the day the first live account opens, this page says so first. What is ready, and what is not yet.

The live record Paper account

Everything it has done, and everything it declined to do.

The same numbers the operator sees, read from the bot's own event log. Equity is a curve and a number among others, never a return. Where the log is silent, so is this page.

Daily candles judgedBTC and ETH, at each 00:00 UTC close
Trades takenFills against the live order book
Refusals written downEach with its gate, its numbers and the size it would have been
LiquidationsA liquidation is treated as a system failure, never an exit

Equity Paper account

Waiting for the equity history…
Show as a table

Why it traded, or did not

Latest decisions, newest first
Waiting for the decision log…

How a decision happens

Five steps between a candle closing and a paper order, and a written reason at every one.

A candle closes

Only closed daily candles count. A look-ahead check runs in the build, so a feature can never see a price that has not happened yet.

The trend is tested

Long when the 30-day and 90-day returns agree upward, flat otherwise. One deterministic rule, the same in replay and live.

Risk sets the size

The smallest of six terms: 0.25% of equity at risk, a volatility target, a daily loss budget, a Kelly cap, a notional cap and the order book's depth. Then every throttle for drawdown and loss windows.

Nine checks must all pass

Post-fill safety state, open risk, caps, depth, spread, funding, no alert anywhere, no cooling window, above the venue minimum. One failure and the trade is refused, with the numbers.

The order is placed and logged

Today it fills against the live Binance book after a modelled delay, with fees, slippage and funding charged. On a live account it goes to your exchange with a resident stop beside it. Every step is in an append-only log that replays to the same state.

Nine things it will never do

Each of these is a test that fails the build.

Most retail bots sell at least one of them. These are written into the approved parameter set as forbidden behaviours, and a named test proves the code cannot do each one.

  • Double down after a lossNo martingale, no recovery sizing. A loss changes nothing about the next size except to make it smaller.martingale_or_recovery_sizing
  • Average down without a planScaling in is allowed only along a ladder declared before the first entry, with a hard cap.averaging_down_beyond_pre_declared_entry_ladder
  • Run a grid or DCA with no ceilingAny grid or dollar-cost plan needs a hard total notional cap or it does not run.grid_or_dca_without_hard_total_notional_cap
  • Re-enter right after a stop-outA four-hour cooling window in the same symbol. The revenge-trading guard.re_entry_within_cooling_window_after_stop_out
  • Raise leverage on an open positionLeverage is set before entry and can only fall while a position is open.leverage_increase_on_open_position
  • Risk more while in drawdownSize scales down with drawdown and never up. A review halt at 10%.risk_per_trade_increase_while_in_drawdown
  • Estimate its own edge liveThe Kelly cap uses an assumed edge fixed in the parameters, never one fitted to recent wins.live_estimation_of_kelly_numerator
  • Use cross marginIsolated margin only, so one position's loss can never reach another's collateral.cross_margin_in_pilot
  • Post volatile collateralStablecoin collateral only. Never wrapped or yield-bearing tokens.non_stablecoin_collateral

Six layers between a signal and a loss

Each layer is true when the one before it has failed.

1Structural limitstrue even when everything is dead

Position leverage at most 2× isolated on BTC and ETH, per-asset notional caps, gross exposure at most 1×, stablecoin collateral only. The worst day in Bitcoin's recorded history does not reach liquidation at these limits.

2Exchange-resident protectiontrue when the bot is dead but the exchange is alive

A reduce-only stop at the planned exit on mark price, a second catastrophe stop at twice the distance, and the exchange's own dead-man's switch so resting entries vanish when the bot dies.

3Bot-managed exitstrue when the bot is alive

Signal reversal, trailing exits, time stops, and exits on volatility doubling, depth collapse or funding cost. These fire well before the resident stops.

4Liquidation Prevention Engineindependent of the strategy

A GREEN → YELLOW → ORANGE → RED → BLACK state machine on liquidation distance, margin ratio, tier changes, ADL rank and market stress. It can only reduce exposure, whatever the strategy thinks. Sensors that need a real account are shown greyed, not assumed clear.

5Portfolio and account guardsthe pre-trade gate

Nine checks before any order, daily, weekly and monthly loss limits, a market-wide circuit breaker on a 5% BTC hourly move, and no-trade windows around exchange maintenance.

6Infrastructure fail-safesa second host in a second region

An independent watchdog that flattens the book if the heartbeat stops for 90 seconds, reconciliation on every restart and every minute, a persisted kill switch only a person can clear, and alerts by message and e-mail.

What is ready, and what is not yet

This page will never show you a return figure. Here is what it shows instead.

The record is a paper account

Fills are simulated against the live order book after a modelled delay, pessimistically. It proves the machine, not a profit. The first live account gets its own label the day it opens.

It has no proven edge yet

The strategy is one deterministic trend rule, chosen for being simple and honest, not for being clever. The rigorous backtest standard it must pass has not run yet. If no rule passes it, the project stops there. That is written into the plan.

It can stop

The road to real money has four places where the plan says stop, not proceed. A stage that makes money and hides a defect is worse than one that loses a little and surfaces it.

Plans

Start by watching. Pay only for what is actually ready.

Pricing goes to the waitlist first. Each plan opens only when the stage behind it is done, and the roadmap below says which stage that is.

Watch free · now

  • The live record on this page
  • Every decision and refusal, with reasons
  • A weekly e-mail of what happened and why

No account needed.

Paper next

  • Your own paper bot on our infrastructure
  • Your symbols and starting equity
  • Alerts, monthly statements, "why it did not trade"
  • Read-only API

Opens after the current soak completes.

Guard later

  • Everything in Paper
  • A read-only connection to your Binance, OKX or Bybit account
  • The safety engine watches your real positions: liquidation distance, funding drag, ADL rank
  • Alerts at YELLOW, ORANGE and RED

Read-only keys. It never places an order.

Trade in preparation

  • Live execution on your own Binance, OKX or Bybit account with the approved parameter set
  • Trade-only keys, withdrawals disabled, IP-bound
  • Your own kill switch and watchdog

Opens after the live pilot, and only where the law allows it. Your money stays on your exchange.

The road to real money

Slow on purpose. Every stage has a gate, and some gates say stop.

Sept 2026
Build

Event log, data layer, risk engine, paper broker, safety engine, strategy, dashboard. 829 tests. Nine forbidden behaviours each with a failing test.

16 Sept → 7 Oct 2026 at the earliest
Soak · running now

21 days of continuous paper trading on the live market, checked daily, recomputed by hand weekly.

Zero paper liquidations in the final 14 days, every restart reconciles clean, replay matches live.
Oct–Nov 2026
Backtests to the institutional standard

Purged k-fold with embargo, deflated Sharpe, cost sensitivity at twice the fees, seven stress replays.

At least one rule passes in full. If none does, the project stops here.
Stage 0.5 · 6–8 weeks
Demo key, no money

The first exchange key. Exchange-resident stops, private streams, real rate limits, every sensor the paper stage could not see.

All 28 failure-injection tests pass. Weekly host-kill drill passes twice.
Stage 1 · 8+ weeks
Pilot with the operator's own US$1,000

Spot only, long or flat. Expected result: roughly flat. The product of this stage is a proven system, not a return.

30 trades, slippage within model, drawdown under 5%, no failure caused by our code.
Stages 2–3
US$3,000, then US$5,000

Perpetuals at 1× isolated, then up to 2×. The configuration the whole research phase was built around.

Never scale up within 30 days of a drawdown over 5%. Never add capital to make back a loss.

Waitlist

Be there when the first plan opens.

One e-mail when a plan opens, and the weekly record if you want it. No marketing sequence. Tell us which exchange you use so we build the connection you need first.

Stored only to contact you about PaperBot. Nothing is shared or sold. Privacy.