Scoring alone does not publish a trade card. Every setup — including high-conviction ones — must clear a quality gate before it reaches the daily brief. The gate is a set of deterministic checks that look for the ways a setup can be inconsistent, unrealistic, or self-contradicting even when the score looks good.
The specific numeric thresholds (risk-to-reward minimums, ATR multiples, profit-floor percentages) are proprietary and tuned continuously based on outcome data. The categories of checks, however, are stable and worth knowing.
Risk-to-reward floor
What it checks → Every setup must offer more potential reward than potential risk by a meaningful margin.
Why it matters → If the stop is closer to entry than the first target, a win barely pays for one loss. Setups that fail the R:R minimum are dropped, even when conviction is otherwise strong.
Target reachability (ATR bounds)
What it checks → The proposed targets must sit within realistic distance given the asset's recent volatility.
Why it matters → A target that asks a low-volatility stock to move 10% in a day is not a setup, it is a wish. We measure recent average true range and reject targets that exceed what the timeframe can plausibly cover.
Profit-floor minimum
What it checks → The distance from entry to the first target must exceed a minimum percentage tied to the setup's timeframe.
Why it matters → A today-timeframe setup with a 0.8% target is not worth the slippage + execution risk. Floors scale up for longer timeframes — a year-long hold demands a far larger move than an intraday trade.
Direction / target coherence
What it checks → For a LONG setup the first target must be above entry and the stop below it. For a SHORT, the reverse. Sides checked independently so neither can slip through.
Why it matters → Inverted setups (stop on the wrong side) would compute zero risk and pass naive R:R math even though they make no sense as trades. The gate checks both sides explicitly.
Indicator coherence
What it checks → When conviction is high, we require that at least one of the three core technical signals (momentum, trend, price action) agrees with the direction.
Why it matters → A high-conviction LONG that contradicts RSI, MACD, AND price-vs-trend is self-inconsistent. The gate blocks these even if outcome history or macro data loaded the score.
Stale / already-past-target
What it checks → A setup where live price has already blown through the target is rejected.
Why it matters → No actionable trade remains — the move has happened. Users would see a card that feels late. The gate drops these in favor of fresh setups.
Volatility sanity
What it checks → We compute historical volatility and reject setups whose targets or stops sit outside realistic standard-deviation bands for the timeframe.
Why it matters → A three-sigma target in one week asks the asset to do something its history says it rarely does. We prefer setups within one to two standard deviations.
Correlation filter
What it checks → Within a single session, no two published setups may be highly correlated on the same side.
Why it matters → If BTC, ETH, and SOL all publish as LONG with conviction in the 80s, readers get what is effectively one trade three times. We keep the highest-conviction expression and drop the correlated doubles.
How blocking works
When a setup fails any of the checks above, it is removed from the publishable pool. The internal pipeline records the rejection reason so we can track which gates are firing most often — this feeds back into our tuning.
On a typical session, roughly one in three candidate setups does not survive the gate. That ratio is by design — we would rather publish fewer, cleaner setups than a firehose of borderline ideas.