Dispatch-Cadence Resonance Slippage Playbook

Date: 2026-03-13
Category: research (execution microstructure)
Scope: intraday child-order scheduling for fragmented lit venues

TL;DR

A fixed child-order dispatch cadence (e.g., every 200ms/500ms/1s) can unintentionally phase-lock with venue-level liquidity replenishment cycles. When that happens, orders repeatedly arrive at the same unfavorable micro-phase (thin queue, stale refill, high cancel hazard), creating a hidden slippage tax.

This playbook treats the problem as a timing-resonance control loop:

• detect cadence-lock risk in real time,
• measure phase-conditioned execution quality,
• break harmful synchronization with bounded jitter and multi-timescale cadence switching,
• protect completion reliability and p95 slippage together.

1) Problem Statement

Most execution schedulers optimize size/urgency/venue but keep dispatch timing mechanically regular. That regularity is usually considered harmless. In practice, it can become expensive when:

1. venue micro-liquidity is quasi-periodic (refill/cancel pulses), and
2. our scheduler repeatedly arrives at the same phase offset.

Result: we keep paying the same microstructure penalty branch, then misattribute cost to “market noise.”

2) Mechanism: How Resonance Creates Hidden Cost

Let:

• τ_d = dispatch interval (our scheduler),
• τ_r = dominant replenishment period (venue-local),
• φ_t = phase of order arrival vs replenishment cycle.

When τ_d / τ_r is near a small rational ratio (1, 1/2, 2/3, 3/2, ...), arrivals concentrate in a narrow phase band.

If that phase band corresponds to:

• low executable depth,
• high queue fade probability,
• elevated adverse markout,

then expected slippage inflates persistently.

This is not one bad print; it is a structural timing tax.

3) Data Contract (Minimal)

Per child order:

• decisionTs, sendTs, ackTs, fillTs
• venue, side, limit/marketability flags
• intended size, executed size
• local quote/depth snapshot at sendTs
• short-horizon markout (e.g., 1s/5s/30s)

Per venue stream:

• L1/L2 updates with precise timestamps
• refill events at touch and near-touch depth bands
• cancel-intensity estimates in rolling windows

Clock discipline is mandatory: compare phase only after timestamp quality filters.

4) Feature Engineering

4.1 Replenishment Period Estimate

Estimate dominant refill periodicity τ_r via short rolling spectral/auto-correlation methods on touch-depth refill signals.

Use confidence weight w_r to avoid over-trusting noisy estimates.

4.2 Phase of Arrival

For each child:

φ_t = 2π * ((sendTs - t0) mod τ_r) / τ_r

where t0 is rolling cycle anchor.

4.3 Resonance Proximity

Define

RP = min_k |τ_d/τ_r - q_k|

where q_k are low-order rational anchors (1, 1/2, 2/3, 3/2, ...).

Lower RP = higher resonance risk.

4.4 Phase-Conditioned Cost Spread

Bucket arrivals by phase decile and compute:

• fill probability by phase,
• realized slippage by phase,
• markout by phase.

Large between-phase dispersion indicates phase-sensitive microstructure.

5) Model Stack

5.1 Baseline Cost Model

Standard slippage forecaster with spread, depth, volatility, urgency, queue/toxicity features.

5.2 Resonance Overlay

Add timing features:

• RP (resonance proximity)
• PhaseQuality(φ_t) (learned expected cost by phase)
• interaction: RP × PhaseQuality

Predictive target should include tail-aware objectives (q90/q95), not only mean bps.

5.3 Resonance Tax Metric

Define Cadence Resonance Tax (CRT):

CRT = E[cost | current cadence policy] - E[cost | phase-randomized counterfactual]

Counterfactual can be estimated with replay + jittered timing simulation under same market path.

6) Real-Time Control State Machine

State A: FREE_RUNNING

• No meaningful resonance signal.
• Use default cadence policy.

State B: WATCH_RESONANCE

Trigger examples:

• RP below threshold,
• phase-cost dispersion rising,
• CRT estimate positive and persistent.

Action:

• activate bounded cadence jitter (small),
• diversify phase footprint across venues.

State C: LOCK_RISK

Trigger examples:

• repeated adverse executions in narrow phase bins,
• CRT breach over rolling horizon,
• p95 slippage drift with stable non-timing covariates.

Action:

• switch to multi-timescale cadence ladder,
• increase randomized offset envelope,
• enforce per-venue anti-lock cooldown.

State D: SAFE_MODE

Trigger examples:

• severe tail-cost escalation with low completion slack.

Action:

• prioritize completion reliability,
• cap experimental timing changes,
• route to safer liquidity profile until resonance signal normalizes.

Use hysteresis to avoid flapping between B/C states.

7) Control Levers

1. Bounded Jitter Injection
   Add small random perturbation to dispatch interval to decorrelate phase lock.

2. Cadence Ladder Switching
   Rotate among vetted intervals (e.g., 180ms/260ms/410ms) using policy rules.

3. Phase-Aware Venue Allocation
   Shift marginal flow toward venues currently in higher phase quality.

4. Deadline Coupling Guardrail
   As completion deadline tightens, reduce exploratory jitter amplitude.

5. Anti-Harmonic Filter
   Block intervals that repeatedly sit near harmful low-order harmonics for a venue regime.

8) Validation Protocol

Offline

• Replay historical sessions with fixed cadence vs jittered/ladder policies.
• Compare mean + q95 slippage, completion rate, and retry/reject burden.
• Perform regime splits: calm / stressed / open / close / event windows.

Shadow Online

• Emit policy decisions without acting.
• Verify CRT signal stability and no timestamp-quality artifacts.

Canary

• Small flow fraction, strict rollback gates:
  • q95 slippage must improve or stay neutral,
  • completion degradation bounded,
  • no significant increase in control-plane rejects.

9) Monitoring Dashboard (Recommended)

• CRT (overall + venue)
• resonance proximity histogram (RP)
• phase occupancy entropy (are arrivals too concentrated?)
• phase-conditioned fill/cost heatmap
• cadence mode share (free/jitter/ladder/safe)
• completion ratio + deadline miss rate
• p50/p90/p95 slippage trend

10) Common Failure Modes

1. Fake resonance from bad clocks
   Timestamp drift can manufacture phase patterns.

2. Over-jitter under tight deadlines
   Decorrelation helps cost but can hurt completion if unconstrained.

3. One-venue overfit
   Harmful harmonics differ by venue and regime.

4. Mean-only optimization
   Resonance mostly hurts tails; mean-only dashboards miss it.

5. No hysteresis
   Controller thrashes between modes and leaks execution quality.

11) Practical Deployment Checklist

• Timestamp-quality guardrails enabled
• Refill-period estimator with confidence score in production
• Phase-conditioned diagnostics validated on replay
• CRT estimator calibrated with counterfactual replay
• Jitter amplitude bounded by deadline-aware policy
• Canary gates for q95 + completion + reject burden
• Rollback automation tested
• Post-launch weekly harmonic blacklist refresh

12) Key Takeaway

In fast microstructure, when you dispatch can matter almost as much as how much you dispatch.

A deterministic cadence can quietly synchronize with unfavorable liquidity phases and create recurring slippage damage. Treat timing resonance as a first-class risk factor, and execution policy becomes materially more robust—especially in p95 tail regimes where hidden costs usually accumulate.