Formerly EdgePilot.
AI-powered quant strategies, automatically selected and deployed as markets change. Systematic trading, made simpler. Not financial advice.
Same fills, different P&L?
For an inverse contract, 50 contracts at 10k and 50 at 15k give a 12k entry, not 12.5k.
Check contract units and the venue's averaging rule before comparing netted and hedged results.
#AlgoRover#Backtesting
Bid and ask look fine. Why is the strategy still waiting?
Inspect a complete snapshot, including its snapshot and end-of-batch flags. Consumers waiting for those signals can stay blocked even while the book updates.
#AlgoRover#MarketData
No auction trade, but the opening bar has a price?
Compare it with the first actual trade and check how missing values are filled. A carried-forward close is not evidence of an executable open.
#AlgoRover#MarketData
New lookback. New pairs?
Changing 12 months to 10 can change which stocks pair up. Save each rebalance's pair list, align test dates and costs, then check each pair's contribution.
A window change can change both model inputs and what you trade.
#AlgoRover#Backtesting
Rate limiter on. Still getting throttled?
If a request costs 40 units but your code charges 1, counting requests won't respect a weighted limit.
Compare the endpoint's documented weight with the delay actually applied.
#AlgoRover#QuantTrading
The table has prices. Why does the backtest return NaN?
Compare the same contract and minute after listing across the source table, contract metadata, subscriptions and replay input.
Trace the missing link before filling gaps with zero.
#AlgoRover#DataQuality
Off the shortlist. Still holding a position?
A market's closing date changes. It drops out of discovery, but you still hold it and the local date stays stale.
Track open positions separately from new candidates.
#AlgoRover#RiskManagement
Same 100-point rule. Same relative move?
At an index level of 10,000, 100 points is 1%. At 30,000, it's about 0.33%.
Decide whether your rule fixes points, percentages or volatility. Changing that basis creates a new rule to test.
#AlgoRover#Backtesting
Look away, and the signal is gone?
Showing the current bar isn't the same as reviewing the last 60 bars. Store each bar's signal, then filter by bar count.
A date cutoff doesn't define a 60-bar window.
#AlgoRover#QuantTrading
Late trade data? Check where it waited.
Compare three timestamps: trade, publication and local receipt.
Split publication delay from delivery delay. A faster server can't recover time a message spent waiting to be sent.
#AlgoRover#MarketData
Same quote currency. Same kind of asset?
A /USDT filter can pull stock tokens into a crypto-only universe.
Check instrument metadata, then inspect the final list. The quote suffix tells you how it's priced, not what you're trading.
#AlgoRover#TradingUniverse
A continuous chart rolls. Did the position?
Moving from the near contract to the next means closing one and opening another.
Check both fills, the price gap and fees. A smooth chart doesn't make the roll free.
#AlgoRover#Futures
One lot traded. How did the backtest fill two?
In one backtest, a resting order fills one lot. An unrelated order triggers another match at the same time. The original order fills again off the same print.
Check fills by order and market event.
#AlgoRover#Backtesting
25 runs, but the estimate says 20?
From 0.10 to 0.30 in steps of 0.05, both ends included, gives five values. Floating-point rounding can make the estimate count four. With five values on the other axis, that's 20 vs. 25.
Check which sets ran.
#AlgoRover#Backtesting
Index way off. Bad quote?
Say a futures contract goes from 100 to 110 while a return-based index goes from 1,000 to 1,100. Both rose 10%.
Check how the index is built. To compare moves, use returns over the same window.
#AlgoRover#Futures
New ticker. Broken lookup?
A reported case: adding OPEN broke a working SPY open-price lookup. The ticker collided with the column name.
Select the symbol and column separately; check against the raw data.
#AlgoRover#DataQuality
Order ID rounded off?
Example: 30000000000000002 becomes 30000000000000000 after a float round-trip. A cancel request now carries the wrong ID.
Keep IDs exact, using strings or integers as the API requires.
#AlgoRover#OrderExecution
Live price. Live open interest?
Example: the page refreshes at 10:00, but open interest is yesterday's snapshot. That's not a new intraday reading.
Check each field's as-of time and update cadence.
#AlgoRover#DataQuality
Stop updated. At the venue too?
Your code moves a trigger from 95 to 98, but the request drops the new price.
Check the outgoing field and the trigger returned by the venue. A local edit isn't proof the stop moved.
#AlgoRover#OrderManagement
No open order. No fill?
An order fills at the venue, but your bot still shows SUBMITTED. An empty open-order list won't settle it.
Match the order ID against history, fills and positions before sending another order.
#AlgoRover#OrderTracking