What does a trading agent do when it isn’t trading?

What does a trading agent do when it isn’t trading? Most of its work. A well-built agent spends its time throwing away trade ideas that fail its rules, re-checking every open position, and moving stops one way as trades gain. Opening a position is the shortest step. In two weeks of telemetry from Senpi’s live strategies on Hyperliquid, published on Aug 11, 2026, the agent edited positions 2.8 times as often as it opened them, and ran about 118 tracking passes for each position it opened.

What the agent checks, and how often

1. It refuses

Every strategy generates signals: something moves, a scanner notices, and a candidate trade appears with an asset and a direction. Almost none of them become trades. Before anything reaches the exchange, a signal has to clear a series of gates, and each one can discard it:

• The risk gate. If the strategy’s risk budget is already used, the signal stops there. Across the two weeks, this was the single most common reason a signal never became a trade.
• Cooldown. After certain outcomes a strategy is benched on purpose. It can see the setup; it isn’t allowed to take it.
• Duplicate check. If the strategy already holds the asset, the signal is discarded rather than adding exposure to a bet it is already in.

Anyone can generate signals. A model with a chart and a prompt will produce trade ideas all day, and most of them are noise. Discarding them is the hard part, and it is what makes leaving a leveraged strategy running unattended survivable: the agent has to say no far more often than it says yes.

2. It watches

Once a position is open, the job changes. Senpi ran about 118 tracking passes for each position it opened, edited positions 2.8 times as often as it opened them (sizes, stops and exit conditions revised as the trade developed), and the protection layer alone ran more than 100 times per position. Opening the trade is the cheapest step; the work comes afterwards. A person can watch one position closely. Watching eleven across crypto, stocks and commodities, around the clock, is another matter.

How often does it check?

In the August 2026 measurement, Senpi’s agents checked open positions about every ten seconds: roughly 8,600 protection checks a day for each strategy, weekends included, because the markets they guard trade around the clock. Today the interval is a setting of each strategy’s exit engine. It is 30 seconds by default, about 2,880 checks a day per open position, and a strategy can set anything from 5 seconds to an hour. Each check asks the same short questions: has the stop’s floor moved, has the position earned the next tier, and are the risk gates still clear. Most checks change nothing; the one that matters comes at an hour when you aren’t watching.

3. It ratchets

Senpi’s exit engine works as a ladder. As an open position gains, it climbs tiers, and each tier locks in a larger share of the highest gain so far. The stop is built to follow the position up and never back down; once the first tier is reached, it rests on Hyperliquid as a stop order the runtime keeps moving up. A one-way mechanism only counts if it is one-way every time, so Senpi measured all the positions the ladder managed in the window:

• 99.3% ended at the highest tier they ever reached; 0.7% ended below it.
• 40.3% climbed at least one tier, so four in ten positions moved far enough to lock in a gain at least once.
• Tier 13 was the furthest any position climbed.
• About 98% of the positions that climbed kept their highest tier. That is our arithmetic from the figures above: only a position that climbed can end below its highest tier.

For the tiers in numbers, with a $100 entry at 10x, see the worked example in what the model does and what the harness does.

What these numbers don’t mean

• They aren’t a promise about returns. They measure the guard, not the trades: whether the checks ran and whether locked gains stayed locked.
• The agent doesn’t trade every few seconds. Orders are rare; the checking is constant.
• The ladder applies to strategies Senpi runs. Every strategy built from a rule or a template exits through DSL, and can carry daily-loss and drawdown halts (copy strategies carry at most a flat stop). On a one-off position, the agent can attach a stop.
• Halts are optional and per strategy. A daily-loss halt pauses new entries until the next UTC day; a drawdown halt waits for you to resume.

How to see it on your own strategy

Ask your Senpi agent whether a position is protected. It checks that the position is tracked with a floor price, that the monitor is running and that the stop order is resting on Hyperliquid, then answers protected, unprotected or stop not on the exchange. To see why it skipped a trade, ask what the strategy refused and why. For how Senpi keeps your funds safe while this runs, see is Senpi safe.

Frequently asked questions

Sources: Senpi, What your agent does when it isn’t trading, and Jason Goldberg, Trading on Hyperliquid? Your positions deserve 8,600 AI protection checks a day, both on X, Aug 11, 2026; fleet figures are from two weeks of Senpi telemetry published then. Current check interval and protection checks from Senpi’s open-source runtime documentation, Nov 13, 2026.