A backtest tells you how a trigger would have done on trades that already happened. A paper run tells you what the same trigger does on the trades that happen next, at the size you chose, with the timing and prices the bot actually sees. They answer different questions, and a bot that has only passed the first one has not been tested. This is why every Rivo bot starts on paper and is promoted to live from its own page, rather than the other way round.
What paper mode records
A paper bot copies every entry its trigger fires on with virtual money at your size. It stores the side, the entry price and the size at the moment the trade printed, marks the position against live odds while the market is open, and scores it a win or a loss when the market settles. The record is reproducible: nothing is re-priced after the fact, and open positions are reported separately from settled ones, never blended into one figure.
A signal bot can also claim up to three months of matching history when it is created, so it starts with a settled record instead of waiting a season for one. That claimed history is the same data the backtest ran on; the forward record is new.
How long to leave it on paper
Long enough for enough positions to settle. The number that matters is settled trades, not days, and it depends on the trigger. A signal over all incoming trades can settle hundreds of positions in a fortnight. A bot copying one careful trader might settle ten in a month. A reasonable floor before you trust a return is several dozen settled trades, and more if the trigger fires on longshots, where a handful of wins carry the whole record.
Watch for one specific failure. If the bot's paper entries are consistently at worse prices than the trigger's own entries, the strategy depends on being first, and a live bot will be later still. The record shows both prices for exactly this reason.
What to read before you promote
- Settled count and win rate against the odds paid. A 60% win rate on favourites priced at 80 cents loses money. The record measures the win rate against the price, so read it that way.
- Return on staked, not total profit. A fixed size per trade keeps two bots comparable. Total profit rewards whichever bot fired most.
- The shape of the curve. One large win that pays for a run of losses is a longshot profile. It can be a good profile, but it is a different bet from steady small wins, and it needs a bigger sample.
- Open exposure. A bot with many open positions and few settled ones has not been scored yet, whatever the marked-to-market number says.
Promoting to live
From the bot's page, Promote to live opens the live form: the venue (Kalshi, Polymarket US or Gemini, through your own connected account), a per-trade ceiling, a per-day ceiling, and the disclaimer. The trigger and the size carry over. Presets start small on purpose; a live bot should earn a bigger ceiling the same way it earned promotion. Every live decision, including every skip, is logged with its reason, so the live record reads like the paper one.
Why paper still does not promise live
Paper fills at the whale's entry price, and a live bot fills at whatever the book offers seconds later. Paper never fails to fill; a live bot skips a thin book. Paper does not pay venue fees. A signal that worked for ninety days can stop working when the market that fed it closes. The paper run narrows the gap between a guess and a decision; it does not close it. Which is the argument for small ceilings, and for reading the live record with the same care as the paper one.
Build the bot in the bot builder, and see the five mistakes copy traders make for the failure modes the paper run is there to catch.