copy latency and slippage: what decides the quality of a copied fill

A copied fill is later than the trader's, always. The question is how much later, how much the price moves in between, and whether that gap matters for the trader you copy. This page explains the mechanics without a marketing number: hypercopy does not publish a copy latency it has not measured on live copies.

the steps

1. The trader's fill happens on hyperliquid and appears on the public trades feed, which lists both counterparties of each fill.
2. Detection. Hypercopy is subscribed to the feed for every coin and filters for the addresses it follows. Its engine runs in the region of hyperliquid's api, so the feed arrives with little network delay.
3. Coalescing. A large order fills as many pieces. Hypercopy waits a few hundred milliseconds to group the pieces into one move, so you get one order instead of many. That wait is a deliberate cost: fewer orders, fewer fees, fewer sub-minimum rejects.
4. Sizing. The copy's mode (pro rata, fixed $ or %), minimum check, leverage.
5. Order. An immediate-or-cancel limit at mid price ±1%. It fills at once against the book or not at all.
6. Notification. The fill reaches you in telegram.

The order is out within a second of the trader's fill in the ordinary case. Network hiccups, hyperliquid load and reconnects make the tail longer. We measured the feed's arrival delay from europe at about half a second at median during development; that is not end-to-end latency, and the engine now runs closer to the exchange.

why ioc at mid ±1%

Hyperliquid has no market order. A copier that wants an immediate fill sends a limit order that crosses the spread, marked immediate-or-cancel. The limit is where you stop chasing. Hypercopy's is 1% from mid: enough to fill on any liquid pair in normal conditions, tight enough that a move that already ran 2% is not chased.

You can tighten it per copy with a max slippage (advanced settings, 0.1% to 10%): if the price has already moved more than that past the trader's price, the entry is not copied, and an order never fills beyond it. Exits always go through.

A no-fill is a neutral outcome: nothing opened, nothing charged, no retry. Exits are different: a close is retried until it fills, because being stuck in a position is worse than a bad exit price.

what slippage actually costs

Slippage is the difference between the trader's price and yours. On BTC or ETH, in normal conditions, a sub-second delay is a few basis points. On a mid-cap with a thin book, the trader's own order moved the price and you arrive after it: tens of basis points, sometimes more.

Compare that with the trader's edge per trade:

• swing trader, trades that make 5% on average: 10 bp of slippage is noise;
• day trader, trades that make 0.8%: 10 bp is an eighth of the edge, on entry and again on exit;
• scalper, trades that make 0.15%: 10 bp each way is the whole edge. Not copyable, at any latency.

This is why hypercopy's profiles lead with median hold and why the roster excludes median holds under five minutes. Milliseconds matter less than who you copy.

what hypercopy does to keep fills close

• engine hosted in the exchange's region, with a static ip and a persistent websocket;
• one subscription per coin on the public feed, not one per user, so it scales without hitting hyperliquid's subscription limits;
• coalescing window of a few hundred milliseconds;
• ioc at mid ±1%, no chase, no retry on entries;
• reduce-only exits retried until flat, and a reconciliation pass every minute.

faq

why not publish a latency number?

Because the number people want (trader's fill to your fill, p50 and p95, on live copies) has not been measured over enough live volume yet. When it has, it will be here.

can i pay for faster copies?

No. Every copy goes through the same engine.

is polling positions slower than the trade feed?

Yes, by seconds. Polling sees a position after it changed; the feed sees the fill as it happens. See hypercopy vs building your own bot.