Trend classification requires elevated volatility and session participation. Either alone is not a partial answer.

The gate

Four combinations, one of which is Trend and three of which are Normal.

The two inputs produce four states and the mapping is deliberately lopsided. Elevated volatility in session is the only combination that authorises trend handling. Elevated volatility out of session defaults to Normal, because range without participation is not continuation. Ordinary volatility in session is Normal by the volatility condition. Ordinary volatility out of session is Normal twice over. Three of four states route to the static architecture, which means the switch's default is the conservative branch and trend handling is the exception that has to be earned by both conditions simultaneously.

Why not a score

A weighted combination would let a strong volatility reading purchase the missing session.

The alternative design is obvious and worse. Score the volatility reading, score the session participation, combine them, and route on the total. Under that scheme a sufficiently extreme overnight print outscores a moderate in-session reading and receives trend handling — which is the precise outcome the two conditions were separated to prevent. Scoring is the right structure when inputs are substitutable, and these are not: session participation is a precondition for continuation rather than a contributor to its likelihood. An AND gate is what encodes a precondition. A score encodes a preference.

What the session condition is measuring

Not time of day — whether there are participants behind the movement.

The session key is often read as a schedule and it is a proxy for something else. Continuation requires that the move be carried by flow that persists, and flow that persists comes from participation. A range produced during thin hours can be identical in magnitude and entirely different in composition — a few orders moving price through a shallow book, with no reason for the movement to continue once they are done. The trend architecture holds positions on the assumption that what pushed price will keep pushing it. Out of session that assumption is unsupported, and the magnitude of the reading does nothing to support it.

• Magnitude and persistence are different properties; ATR measures only the first.
• Thin-book range can match in-session range and mean the opposite thing.
• The session key is a proxy for participation, which is what continuation needs.

Deliberately binary

The switch has two states because the architectures it selects between are two.

There is a recurring suggestion that the switch should have intermediate states — a partial trend, a semi-session. It should not, and the reason is downstream rather than upstream. The switch's output selects a branch family, and there are two families with two distinct management architectures. An intermediate switch state would have to select an intermediate architecture, which does not exist and could not be defined without inventing a third branch. Resolution beyond what the downstream can act on is not extra precision; it is a number that has to be rounded by the operator, at which point the rounding is the actual decision and it is undocumented.

The residual judgement

The gate is mechanical; whether the setup deserves a trade at all is not.

Worth being clear about the boundary. The switch answers which management architecture applies given that a trade is being taken. It does not answer whether the setup qualifies, whether the structure supports continuation, or whether the account may carry the position — those are the plan's, the operator's and the gate's questions respectively, and all three can refuse a trade the switch has classified as Trend. The two-key lock removes one specific judgement from the operator's hands, cleanly and completely, and leaves the rest exactly where it was.

The key idea

When one input is a precondition rather than a contributor, the structure has to be a gate.

The design question worth carrying elsewhere is whether two inputs are substitutable. If more of one genuinely compensates for less of the other, a score is correct and an AND gate throws away information. If one is a precondition — a state without which the other's meaning changes rather than diminishes — then a score is a mechanism for talking yourself past it. The test is to ask what an extreme reading on the present input would mean if the absent one stayed absent. If the answer is not much, the structure is a gate.