Order-ACK Coalescing & Residual-Illusion Slippage Playbook

Why this matters

Many execution stacks assume acknowledgment timing ≈ market timing. That breaks when broker/exchange gateways coalesce order ACKs into bursts (for throughput protection, queue flush cycles, or session-level shaping). During coalescing windows, your parent thinks child intents are still outstanding, then suddenly receives a burst of late ACKs/fills. The result is a residual-illusion loop:

• You believe you are underfilled and accelerate.
• Real fills were already happening.
• Late ACK burst arrives, revealing overfill + adverse markout.

This is a hidden but recurring slippage tax in fragmented and high-message-rate regimes.

Failure mode in one line

State-estimation lag (from ACK coalescing) causes false residuals, which causes unnecessary urgency, which worsens impact and markout.

Observable signatures

1) ACK burstiness without matching intent burst

• Child-send cadence is stable, but ACK arrivals cluster.
• Inter-ACK times show bi-modality: long silence + microburst.

2) Fill-before-ACK behavior spikes

• Exchange/drop-copy execution timestamps precede client ACK visibility by large deltas.
• Parent inventory estimate lags true inventory.

3) Aggression overshoot episodes

• Temporary participation-rate spikes right after ACK catch-up.
• Residual flips sign (positive -> negative) immediately after burst reconciliation.

Core model: latent residual vs observed residual

Define:

• R_true(t): true residual quantity at time t
• R_obs(t): residual seen by parent using ACK-visible state
• L_ack(t): ACK visibility lag random variable
• u(t): urgency / aggression control action

Then:

R_obs(t) = R_true(t) + ε_ack(t)

where ε_ack(t) increases with ACK coalescing lag and burst depth.

Control law must use de-biased residual:

R_hat(t) = R_obs(t) - E[ε_ack | regime, venue, msg_rate, session_state]

and cap urgency when ACK uncertainty is high.

Practical feature set

ACK-path features

• ack_lag_ms_p50/p90/p99
• ack_burst_size_mean/p95
• ack_silence_window_ms
• ack_coalescing_index = p90(inter_ack) / p10(inter_ack)

Execution-path features

• fill_to_ack_gap_ms
• dropcopy_to_ack_gap_ms
• late_ack_fraction (ACK later than control-cycle horizon)

Control-risk features

• residual_flip_rate_1m
• post_burst_participation_spike
• overshoot_notional_bps

Regime state machine

NORMAL

• ACK lag stable, burst index low.
• Use standard residual-tracking controller.

ACK_LAGGED

Trigger:

• ack_lag_ms_p95 above threshold OR ack_coalescing_index jump.

Actions:

• Switch from R_obs to R_hat.
• Reduce aggression gain k_u.
• Tighten child-order size cap.

BURST_RECONCILE

Trigger:

• ACK burst arrival (ack_burst_size > burst_threshold).

Actions:

• Freeze urgency escalation for T_freeze.
• Recompute inventory from authoritative fills/drop-copy.
• Cancel stale duplicate intents before re-optimizing.

SAFE_CONTAIN

Trigger:

• Repeated residual sign flips + growing overshoot.

Actions:

• Enforce hard participation cap.
• Prefer passive/price-improved tactics.
• Optionally pause child creation until state confidence recovers.

Online calibration loop

1. Estimate lag distribution per venue/session bucket

   • Open, midday, close, high-vol bursts, and gateway throttle periods.

2. Fit residual-bias model

   • Predict ε_ack from ACK-path + message-rate covariates.

3. Backtest with delayed-information replay

   • Reconstruct what controller would have seen in real time.

4. Policy guardrail tuning

   • Choose k_u attenuation, freeze windows, and burst thresholds via tail-risk objective (p95/p99 IS).

Metrics that should be on your dashboard

• ACK_LAG_BPS_TAX: incremental IS bps attributable to ACK visibility lag
• RESIDUAL_ILLUSION_RATE: % control cycles where sign(R_obs) != sign(R_true)
• BURST_RECONCILE_COUNT: episodes per symbol/session
• POST_BURST_MARKOUT_5S: short-horizon toxicity after burst catch-up
• SAFE_CONTAIN_DWELL: time spent in protective mode

Incident runbook (fast)

1. Confirm ACK-path degradation (not market-data gap first).
2. Cross-check with drop-copy/fill channel; rebuild authoritative position.
3. Enter ACK_LAGGED profile immediately.
4. If sign-flip loop persists, move to SAFE_CONTAIN.
5. Post-incident: update lag priors and threshold tables by venue/session slice.

What usually goes wrong in production

• Treating ACK timestamps as execution truth.
• Using one global threshold across venues.
• Ignoring session-phase effects (open/close have different coalescing patterns).
• Replaying fills but not control-loop observability (you miss policy error source).

Minimal implementation checklist

• Split execution truth channel vs ACK visibility channel.
• Compute de-biased residual R_hat online.
• Add ACK_LAGGED/BURST_RECONCILE/SAFE_CONTAIN states.
• Add urgency freeze after burst catch-up.
• Track p99 tail metrics, not only average IS.
• Recalibrate per venue + session phase weekly.

Bottom line

ACK coalescing is not just an ops nuisance; it is a state-estimation error source that directly changes execution behavior. If you model and control for residual illusion explicitly, you can cut avoidable urgency spikes and recover a meaningful chunk of tail slippage.