Equity people inherit a habit that does not transfer. In listed equities there is a consolidated tape, so a backtest run on one data vendor's prints and an execution routed somewhere else still refer to a common reference. Crypto has no consolidated tape. Every candle in a crypto backtest belongs to one venue, and a strategy validated on that venue's history is a claim about that venue's microstructure, not about the asset.
The good news is that this engine does not hide it. The venue is the first field on every signal.
The venue is printed on the signal, and so is the instrument type
The crypto strategy feed presents each signal as a card with a fixed payload. At capture the feed header read 50 of 442, with a template control and a search over keyword, ticker or symbol, and each card carried a configuration name, a deploy action and a timestamp.
Two adjacent cards make the point better than any argument. Configuration 19 reads a trading pair of BYBIT:SOLUSDT.P, a sell action, prior position long, current position flat, leverage 2, price 76.93, time period 1D. Configuration 20 reads a trading pair of BINANCE:CRVUSDT, a sell action, prior long, current flat, leverage 0, price 0.2289, time period 360m.
Read those two lines as an operations person rather than a researcher. The first is a perpetual contract on one venue, carrying leverage and therefore funding. The second is a spot pair on a different venue at zero leverage. Same engine, same signal grammar, two entirely different execution problems, two different sets of fees, two different liquidation regimes and two different sources of truth for what the price was at the moment the rule triggered.
What actually differs once you leave the reference venue
Four things move, and they move independently of each other, which is why a single slippage assumption does not cover them.
The price level itself. Two venues quoting the same spot pair are arbitraged to within a few basis points most of the time and not during the ten minutes you care about. The dispersion widens exactly when a momentum rule triggers, because the trigger is a fast move and fast moves are when cross-venue latency and inventory constraints bind. A backtest that filled at one venue's print during a move is describing a fill that may not have been available anywhere else.
The instrument. A perpetual is not the spot pair with more leverage. It carries a funding payment that accrues periodically, so a strategy holding a perpetual through many funding intervals has a running cost the spot backtest does not contain, and the sign of that cost is systematically against a crowded direction. If the reference series was the perpetual and you execute spot, or the reverse, you have introduced a basis you never modelled.
The bar boundary. The 360m period on that second card is a six hour bar, and a six hour bar has to be anchored somewhere. Whether it starts at 00:00 UTC, at the venue's own listing epoch, or at the data provider's convention determines which trades exist at all. A rule that fires on a bar close fires at a different moment, and therefore at a different price, under a different anchor. This is the difference nobody checks and it is fully deterministic, so it is also the easiest one to eliminate.
The fee schedule. Maker and taker rates differ by venue and by tier, and a perpetual's taker fee on a leveraged notional is charged on the full contract value rather than on your margin. Backtests are frequently computed at a single blended rate, and the engine's own profit factor tile is labelled net of fees across all, which tells you fees were applied and does not tell you at which schedule.
What that does to the trade you are actually evaluating
Take the Curve round trip that is visible in the feed and price the sensitivity, because the abstract version of this argument never persuades anyone.
The buy card reads a price of 0.2266 on the 360m configuration, and the sell card on the same configuration reads 0.2289 about ten hours later. That is 0.0023 of gross move, a little over 100 basis points. Now assume your executable venue prints two ticks away from the reference on each side, which is 0.0002 in and 0.0002 out. You have given up 0.0004 against a gross of 0.0023, which is 17 percent of the trade.
Repeat that across a high frequency of small winners and the strategy's published edge does not survive the substitution. This is the specific reason a venue mismatch is more dangerous on short holding periods than on long ones. The absolute price discrepancy between venues is roughly constant in basis points and the gross profit per trade shrinks as the bar shortens, so the ratio between them, which is the only thing that matters, deteriorates as you move from a 1D configuration to a 360m one.
Notice also that the card's own notes advertise a 99 percent win rate and a 16 percent drawdown on that configuration, and 92 percent with 15 percent drawdown on the other. Those are the reference venue's numbers under the reference venue's fills. They are not a forecast for a different venue and they were never presented as one.
The questions to settle before a deployment is funded
Put these in the diligence pack and require written answers, because every one of them has a factual answer that someone knows.
Which venue's series was the strategy estimated on, and is that the venue you will execute on? If not, name the substitute in the funding memo rather than leaving it to the trader on the day.
Which price does the rule reference, and which price does the fill assume? Last trade, mid, mark price and index price are four different series, and on a leveraged venue the index price governs liquidation while the last trade governs your fill. A backtest that triggers on last and fills on last is optimistic in a way that only shows up in fast markets.
Where is the bar anchored, and in which timezone? For anything that is not a whole day, get the anchor in writing.
Is funding included for perpetual configurations, and at what assumption? If the answer is that funding was not modelled, then the leveraged configurations are not comparable to the spot ones and should not sit in the same ranking.
What fee tier was assumed, and does your actual volume qualify for it? A backtest run at the best tier while you trade at the standard one is a fixed, knowable haircut you can compute today.
And the one people forget: can you legally and operationally trade the reference venue at all? Venue access is a compliance question with a jurisdictional answer, and if the answer is no for a fund domiciled where yours is, then every published figure for that configuration is a number about a market you cannot participate in.
The substitution test, and what to do when it cannot be run
The clean resolution is to re-run the strategy on the venue you will actually execute on, over the same window, with your fee tier, and compare the two records side by side. Ask for that before funding. The delta between the two is the venue haircut, and it belongs in the funding memo as a named cost rather than appearing later as unexplained tracking error.
If the re-run cannot be produced, you still have a defensible fallback. Measure the cross-venue price dispersion yourself over a representative window, take an upper quantile rather than the mean, apply it to both sides of every round trip, and multiply by the strategy's trade count. Then require the strategy to clear its hurdle after that haircut. It is a crude adjustment and it is honest about being crude, which is a better position than adopting a number from a venue you do not trade.
Whatever you conclude, write the reference venue into the deployment record next to the strategy name. Twelve months from now, when a review asks why the sleeve tracked below the published figures, the answer needs to be a documented substitution with a priced haircut, not a discovery.