Position-Staleness & Inventory-Hallucination Slippage Playbook

Why this matters

Many execution stacks optimize slippage assuming position state is close to real-time truth.

In production, that assumption fails when:

• drop-copy lags the primary execution channel,
• venue confirms arrive out-of-order,
• partial fills are batched,
• reconciliation applies late correction bursts.

Then the router trades against a hallucinated inventory.

Result: you pay hidden slippage through avoidable hedge churn, over-correction, and deadline panic decisions.

Core failure mechanism

Let:

• I_true(t): true net inventory,
• I_est(t): strategy-estimated inventory used for routing/hedging,
• e_I(t) = I_est(t) - I_true(t): inventory hallucination error.

When event latency widens, e_I becomes persistent, not noise.

Two expensive branches appear:

1. Phantom-neutral branch (I_est ≈ 0, I_true != 0)

   • strategy believes risk is neutral and delays corrective flow,
   • later catches up in worse liquidity.

2. Phantom-exposed branch (I_est != 0, I_true ≈ 0)

   • unnecessary hedges/offsets are fired,
   • you pay spread/fees/impact for non-risk.

Near cutoff times, both branches become convex in cost.

Branch-cost decomposition

E[Cost] = C_base + C_phantom_hedge + C_late_catchup + C_flip_churn + C_reconcile_shock

Where:

• C_base: normal spread/impact/fees under synchronized inventory,
• C_phantom_hedge: unnecessary hedging from phantom exposure,
• C_late_catchup: delayed correction when stale inventory hid true risk,
• C_flip_churn: repeated side switching from noisy state updates,
• C_reconcile_shock: large forced adjustments after delayed reconciliation.

The key is not just average lag, but lag interacting with urgency windows.

Metric stack

1) Position Freshness Gap (PFG)

PFG = now - last_confirmed_inventory_event_ts

• High PFG means the controller is acting on stale state.

2) Inventory Uncertainty Band (IUB)

IUB = q90(I_true | event stream) - q10(...)

• Wide IUB means brittle one-shot hedge decisions should be suppressed.

3) Shadow Exposure at Risk (SEaR)

Expected notional error from inventory uncertainty:

SEaR = E[ |I_true - I_est| * mid_price ]

• Converts stale-state risk into risk-budget units.

4) Hedge Mismatch Pressure (HMP)

HMP = E[ |H*(I_est) - H*(I_true)| ]

Where H* is target hedge function.

• High HMP implies likely phantom hedges or delayed genuine hedges.

5) Reconciliation Shock Rate (RSR)

Frequency of inventory corrections above threshold k shares/contracts after delay.

• High RSR indicates hidden state debt being repaid in bursts.

Control policy state machine

STATE 1 — LOCKED

Conditions:

• low PFG,
• narrow IUB,
• low RSR.

Policy:

• normal execution and hedge logic,
• standard urgency ramp.

STATE 2 — DRIFT_WATCH

Conditions:

• PFG rising,
• moderate IUB,
• early RSR uptick.

Policy:

• reduce hedge flip frequency,
• require stronger signal before side reversal,
• cap incremental aggression per cycle.

STATE 3 — SHADOW_RISK

Conditions:

• high IUB/SEaR,
• persistent stale events,
• rising HMP.

Policy:

• switch to uncertainty-aware hedge sizing,
• prioritize reliability venues/paths for state-critical child orders,
• throttle cancel/replace churn,
• reserve explicit catch-up budget.

STATE 4 — SAFE

Conditions:

• deadline proximity + high SEaR,
• reconciliation instability,
• branch-cost projection exceeds risk budget.

Policy:

• freeze non-essential micro-optimizations,
• execute pre-approved bounded completion ladder,
• prefer state-certainty over local spread optics.

Use asymmetric hysteresis to prevent flapping.

Modeling pattern (production)

1. Dual-ledger inventory model

   • keep I_confirmed and I_provisional in parallel,
   • decision logic consumes posterior over I_true, not single point state.

2. Event-latency regime model

   • estimate latency by venue × channel × session segment,
   • detect sudden lag-regime shifts and widen uncertainty immediately.

3. Uncertainty-aware action objective

   • optimize mean + q95 cost with explicit penalties on phantom hedge and late catch-up branches.

4. Reconciliation-safe ingestion contract

   • idempotent event apply,
   • monotonic sequencing where available,
   • bounded correction protocol for late corrections.

5. Delay-injection replay

   • replay historical sessions with synthetic drop-copy and confirm delays,
   • validate state transitions and p95 protection before policy promotion.

Practical rollout checklist

• Add PFG/IUB/SEaR/HMP/RSR dashboards by strategy and symbol bucket.
• Introduce DRIFT_WATCH/SHADOW_RISK states with manual override.
• Add side-flip cooldown + minimum conviction threshold.
• Canary uncertainty-aware hedge sizing on one strategy.
• Weekly attribution split: synchronized windows vs stale-state windows.

Bottom line

A large share of avoidable slippage is not market impact—it is state impact from trading on stale inventory truth.

Treat inventory as a latency-conditioned belief state, not an exact register. Once uncertainty is priced directly into control decisions, phantom hedges and late catch-up convexity drop materially, especially in deadline-heavy execution windows.