Multi-Signal Exit Confirmation for Market-Regime Withdrawal

Sequenced leg gates, the fake-reversal filter, and the confirmation-lag discipline.

A regime is easy to enter and hard to leave. Why should the exit ever trust a single print?

The detector that named the withdrawal fired on one threshold crossing. The exit asks a harder question, and a single crossing is the wrong way to answer it. Today is not a hypothetical: the live WITHDRAWAL episode reached Day 14 with the $60K shelf held by default, and the first formal exit gate fires on this morning's ETF flow publication.

The 06-19 lesson built the detector that labels the market state. This lesson governs the moment that label should change back. The discipline below separates a true regime turn from the fake reversal the tape throws up three times before the real one.

§ IFrame

Entry and exit are not symmetric. Entry can afford to be eager because a slightly-early entry into a confirmed regime costs little. Exit cannot, because the cost of a false exit is the worst trade in the book: closing the protective position one day before the regime that made it correct resumes with force.

So the exit runs on a different architecture. The entry was one signal over one threshold. The exit is several signals, ordered, each gating the next, with a deliberate lag built into the sequence. Call the structure the sequenced leg gate: a chain where no leg counts until the leg before it has held, and the regime does not flip until every leg has cleared. The sequenced leg gate is the architecture. The fake-reversal filter rejects single-print head-fakes. The confirmation-lag discipline is the patience that makes the first two work.

§ IIFoundations

Why one signal is never enough to call an exit

A regime is a persistent state, not a daily reading. Aldridge's treatment of statistical-arbitrage strategies (Ch 8, pp. 217–221) carries the operational point: the tradeable entity is the relationship, and a relationship is confirmed by repeated behavior, not one observation. One positive ETF print is one observation. The regime turning is a change in the relationship between flows, price, and structure that has to show in more than one place before it is real.

Mean-reversion testing supplies the second half. Halls-Moore frames the question (Successful Algorithmic Trading, IV Modelling, p. 96) as whether a series returns to its mean reliably enough to trade. A withdrawal regime is a low-mean state, and the exit question is whether the series has genuinely left it or merely printed one value near the boundary. A boundary-touch is not a state change.

The fake-reversal filter

A fake reversal is a single signal that crosses the exit threshold and then fails to hold. June 2026 produced one on June 4, when a lone net-positive ETF print broke a multi-day outflow streak and reversed the next session. A protocol firing on that print would have closed into the continuation. The fake-reversal filter is the rule that no single print initializes an exit: a leg requires either multiple instruments agreeing on the same session or the same instrument agreeing across consecutive sessions before it counts as cleared.

The confirmation-lag discipline

The third idea is the one operators resist, because it feels like leaving money on the table. The confirmation-lag discipline holds that the exit should lag the first positive signal by design. The lag is not slowness; it is the price paid to convert a noisy signal into a confirmed one. A regime fourteen days old is no more expensive to exit on Day 16 than on Day 14, and the two extra days buy the multi-session agreement the filter requires.

§ IIIMechanism

The exit protocol is a four-leg sequence. Each leg has a threshold, a confirmation rule, and a position in the order. A leg does not begin evaluating until the leg before it has cleared.

The regime flips from WITHDRAWAL to its successor only when all four legs have cleared in order. A leg that arms and then fails its confirmation resets to unarmed, and the sequence does not advance past it. Order enforced, confirmation required at every leg, no skipping.

§ IVWorked Example

Take the live episode. Day 14, $60K shelf held as resistance, this morning's ETF publication covering Monday's flows is the Leg-1 input.

Suppose the print is a single fund net-positive $40M, every other fund flat or negative. Under a naive exit, that is a green print and a reason to lighten. Under the sequenced leg gate it is nothing: it fails the multi-fund threshold, so Leg 1 does not even arm. WITHDRAWAL holds into Day 15. Now suppose instead two funds print net-positive, combined $90M. Leg 1 arms but does not clear — the fake-reversal filter requires a second confirming publication. Tomorrow either confirms or resets. Quarter-end Tuesday makes the single-print read especially treacherous, because rebalancing flows can manufacture exactly the one-day positive the filter is built to reject.

The structure the detector supplies sharpens the picture. SOL has led BTC and ETH on relative strength for three sessions inside the withdrawal — the alt-relative-strength-rotation-within-WITHDRAWAL pattern. One asset leading the regime out raises attention, but leadership-in-one-name is not a leg. It is context, not a signal that clears a gate.

def evaluate_exit(legs: list[Leg], signals: SignalBundle) -> ExitState:
    state = ExitState(regime="WITHDRAWAL", cleared=[])
    for leg in legs:
        outcome = leg.evaluate(signals)
        if outcome is Outcome.UNARMED:
            break
        if outcome is Outcome.ARMED:
            state.armed = leg.name
            break
        state.cleared.append(leg.name)
    if len(state.cleared) == len(legs):
        state.regime = "EXIT_CONFIRMED"
    return state

The loop reads as the discipline reads. It stops at the first leg that is not cleared. It promotes the regime only when every leg has cleared. No path through this function exits on a single signal, because the structure forbids it.

§ VConnection to Prior Lessons

06-19 regime detection. That lesson introduced the hysteresis buffer — the rule against flipping state inside the first days of a transition. The exit protocol is that same principle expressed as a multi-leg sequence. Detection answers which regime; this lesson answers when it ends.

06-26 cross-asset pair signals. The pair trade is sized differently inside a confirmed-withdrawal regime than during an exit. The exit gate tells the pair book when the regime tailwind is fading, so positions can unwind before the regime that justified them turns.

06-12 strategy attribution. Attribution measured whether a strategy still earns its capital. The exit protocol is the regime-level analog — it measures whether the regime that justified the whole defensive posture still holds. Both are retirement gates: one retires a strategy, this retires a regime.

§ VIConnection to Today's Dev Lesson

The Python lesson implements the sequenced leg gate as a guarded state machine. Each leg is a state; each confirmation rule is a guard on the transition out of that state. The enum defines the legal states, the transition table defines the allowed moves, and a confirmation counter on a deque enforces the consecutive-print rule the fake-reversal filter requires.

This lesson defines what each leg must check. The Python lesson encodes the legality of the moves so no caller can advance the machine out of order.

§ VIIClosing

The exit is harder than the entry because the cost of error is asymmetric, and the discipline answers the asymmetry with sequence, confirmation, and patience. The sequenced leg gate refuses to advance out of order. The fake-reversal filter refuses the single print. The confirmation-lag discipline accepts a small, known cost in timing to reject a large, unknown cost in false exits.

Today the protocol gets its first live test. Whatever the ETF print says this morning, the correct response is to read it through the gate, not around it.

Examine the print when it lands. Run it through Leg 1 before you run it through your hopes.