An event-aware wrapper needs signal concentration, not just signal presence

The 3-arm A/B from the baseline arm you forgot to include ran on a synthetic FOMC simulator with isotropic vol shocks — high vol on FOMC days, no direction. Both event-aware arms underperformed the baseline. That was the expected answer: no directional signal, nothing for an event-aware capture to capture.

Then the simulator gained a --fomc-drift-bps flag that adds a directional component, signed by recent momentum (winners-keep-winning, losers-keep-losing) — the documented post-FOMC momentum drift (Lucca & Moench 2015, and follow-ups). The expectation was that the drift-capture arm would now beat the baseline. It didn’t. Here’s the shape of the result.

What happened

5 seeds, --fomc-drift-bps 50, 4h drift-capture window, 24h blackout window:

baseline:   sharpe  +1.001   (mean across 5 seeds)
blackout:   sharpe  +1.027   (Δ +0.026 vs baseline — noise)
drift:      sharpe  -0.117   (Δ −1.118 vs baseline)

The baseline jumped from +0.43 (isotropic) to +1.00 (directional) — it’s absorbing the new signal through its regular cadence. The drift arm did not jump correspondingly. Why?

The window is the wrong shape

The momentum-signed FOMC shock is a multi-day effect. A symbol in the top quantile on a FOMC day gets a +50 bp drift kick on that day, and the next day’s open inherits that price level. The baseline strategy is already long that symbol from the previous rebalance, holds through the FOMC day, and books the gain on the next-day close-to-close return.

The drift arm only takes a position during the 4-hour window after the FOMC release fires. It’s a flat → long-or-short → flat cycle that captures one bar’s worth of return. The directional component is real but spread over the whole FOMC day’s close-to- close move; the 4-hour slice captures a fraction of it.

The blackout arm, going the other way, flattens through FOMC and misses the same gain the baseline captures. Its Sharpe is essentially equal to baseline because the 7 FOMC days out of 200 are a small fraction of total exposure — small enough that missing them is within seed-to-seed noise.

The generalisable claim

Event-aware capture wrappers don’t earn their cost on “is there a post-event signal?” The bar they have to clear is:

Is the post-event signal concentrated in a narrow window the wrapper trades, but not captured by the underlying strategy’s existing cadence?

Both halves matter. A signal that’s present but diffuse (multi-day drift, gradual mean reversion, persistent sentiment shift) leaks into the baseline strategy’s regular rebalances. The capture wrapper only sees the slice inside its window — usually a fraction of the total signal.

A signal that’s concentrated — say, a 30-minute post-event volatility crush, or a 15-minute order-flow imbalance window — can live entirely inside the wrapper’s window. The baseline, trading at panel-cadence, misses that one window per event; the wrapper concentrates exposure exactly there.

What this means for designing the wrapper

Three concrete moves before shipping a capture wrapper:

1. Measure the signal’s half-life under your shock model. If the signal’s autocorrelation decays over hours or days, the wrapper’s window has to be long enough to overlap most of it. Too-narrow windows leave alpha on the floor for the baseline to absorb.

2. Compute what fraction of the per-event signal the wrapper’s window actually captures. A 4-hour window into a 24-hour diffuse drift captures ~1/6 of the signal at best. That’s a structural cap, not a tuning issue.

3. Ask whether the wrapper is actually adding exposure the baseline doesn’t already get. A wrapper that trades in a window the baseline already covers via its normal rebalance is subtracting from the baseline (more turnover, more frictions) without adding signal. The the-baseline-arm-you-forgot 3-arm shape surfaces this directly: gate-on Δ ≈ 0 and gate- inverse Δ negative is the signature.

The alphakernel post-FOMC drift wrapper exists for two non-alpha reasons named in the event-clock surface note: it demonstrates the event-clock surface end-to-end, and it provides a strategy with real event:event_release/<id> citations for ADR-0031 Rule 1 to gate against. Both useful. Neither requires the wrapper to beat the baseline on the same data.

The lesson the directional-drift simulator made visible: an event-aware wrapper’s edge is a property of the wrapper-data geometry, not the data alone. Measure the geometry before shipping the wrapper.