A quiet strategy is never a mystery.
Every limit is a ceiling, not a target, and every refusal is written to the decision that was refused. A strategy that found no edge and one that is out of bankroll look nothing alike, which is the whole point.
The order of the checks is the design
They run cheapest first. By the time a venue is contacted, six other things have already agreed the order should exist.
-
First
Strategy state
Paused or halted strategies stop here. Halting is sticky: it is written to the strategy, survives a restart, and carries the reason it happened.
-
Then
Plan entitlement
Automated execution needs a plan that includes it. A lapsed subscription stops placing orders rather than continuing on the assumption the card recovers.
-
Then
Daily loss limit
Measured on realized cost from filled orders since midnight UTC, not on marked-to-market swings. A limit that trips on an unrealized move halts a strategy over a price change that has cost nothing.
-
Then
Open positions
A hard count of orders still live or filled and unsettled. It is the cap on how many things can go wrong at once.
-
Then
Bankroll
Committed notional plus this order must fit inside the bankroll you declared. The declared number is the ceiling, never a target to fill.
-
Then
Venue health
The account must exist and be active. Credentials the venue has rejected mark the account revoked, so the next pass stops here instead of burning a round trip per decision.
-
Last
The order itself
Only now does anything leave the process. The venue is asked last because it is the only check that costs a network call.
Sizing is deliberately timid
Quarter-Kelly, scaled again by confidence. It gives up some growth in exchange for a lot of tolerance to being wrong about fair value, which is the right trade when the probability is an estimate.
One decision, one order
Decisions are unique per strategy and market, so a re-run that still sees the same edge updates its opinion instead of stacking a second order. Orders carry an idempotency key, so a retry cannot fill twice.
Failures do not cascade
A venue rejecting one order does not stop the rest, and one strategy failing does not stop another user's pass. The alternative is a single bad market silently disabling everything.
What PredictHQ does not claim
A modelled edge is an estimate from a trade tape, and it can be wrong. These limits bound how much any single mistake costs; they do not make the mistakes less likely, and they cannot make a losing strategy profitable. PredictHQ does not know your circumstances and nothing it produces is financial advice.