Fee-Table Version Drift Routing Slippage Playbook

Date: 2026-03-15
Category: research
Focus: Modeling and controlling slippage when execution routers trade on stale fee/rebate assumptions while exchange economics have already changed.

1) Why this failure mode matters

Most execution stacks optimize a net-cost objective like:

expected impact + expected spread + expected fee/rebate

That objective is only valid if fee inputs are correct at decision time.

In live environments, fee truth changes faster than many systems assume:

• exchange fee schedule revisions,
• tier transitions during the month/day,
• flag-level charging differences (displayed/passive/hidden/post-only midpoint, etc.),
• venue-specific pass-through changes at broker level,
• stale config rollout across router vs TCA vs risk engines.

When fee truth drifts but routing logic does not, desk behavior can look “impact-smart” while silently leaking basis points through wrong venue ranking.

2) Mechanism map

2.1 Drift sources

Typical entry paths:

1. Version skew: router reads fee table v_k, settlement/TCA uses v_{k+1}.
2. Tier-state mismatch: projected tier notional is stale, so marginal fee assumptions are wrong.
3. Flag misclassification: child-order type predicts one fee bucket but actual venue execution prints in another.
4. Partial rollout: one region/host updates fee config, another does not.
5. Calendar boundary bugs: effective-time cutover (month/day/session) applied late or in wrong timezone.

2.2 Cost branch dynamics

Drift rarely appears as one big error. It compounds through routing behavior:

• venue score is mis-ranked,
• order flow concentrates where net edge is overestimated,
• realized fees disappoint,
• residual schedule falls behind,
• late catch-up aggression increases impact cost.

So fee drift becomes both direct fee leakage and indirect impact leakage.

3) Cost decomposition

Let realized total execution cost per parent be:

[ C_{total} = C_{impact} + C_{spread} + C_{fees} + C_{delay} + C_{opportunity} ]

Under fee-drift episodes, separate fee-induced error:

[ C_{total} = C_{baseline} + C_{fee_mispricing} + C_{reroute_distortion} + C_{catchup} ]

Where:

• (C_{fee_mispricing}): direct add/remove fee-rebate miss,
• (C_{reroute_distortion}): wrong venue choice due to stale net-cost ranking,
• (C_{catchup}): urgency tax from schedule deficit created by earlier misrouting.

Expected branch framing:

[ \mathbb{E}[C] = p_A C_A + p_D C_D ]

• (A): aligned-fee regime,
• (D): drift regime.

Goal is reducing both drift probability (p_D) and drift-severity gap ((C_D - C_A)).

4) Feature set for modeling

4.1 Configuration consistency features

• router_fee_table_version
• canonical_fee_table_version
• version_age_seconds
• host_version_entropy (cross-host version disagreement)
• effective_time_offset_ms (config effective timestamp vs canonical clock)

4.2 Tier-state features

• projected_monthly_add_volume
• realized_monthly_add_volume
• tier_distance_bps (distance to next fee tier in expected bps benefit)
• tier_cross_probability_1h
• tier_projection_error_abs

4.3 Routing behavior features

• expected_net_cost_bps(route) vs realized_net_cost_bps(route)
• fee_component_weight_in_score
• venue_flow_concentration_hhi
• route_flip_rate_after_fee_refresh

4.4 Market interaction features

• local spread/depth/volatility,
• urgency state,
• time-to-close,
• participation deficit.

Fee-drift damage is highest when urgency is already high and route flexibility is low.

5) Operational metrics

5.1 FVG — Fee Version Gap

[ FVG = \mathbb{1}[v_{router} \neq v_{canonical}] \times version_age_seconds ]

Volume-weighted FVG captures how long meaningful flow traded under stale assumptions.

5.2 NCE — Net-Cost Error

[ NCE = realized_net_cost_bps - expected_net_cost_bps ]

Track by venue/flag/session. Persistent positive NCE indicates model/config drift.

5.3 TMS — Tier Margin Stress

[ TMS = \frac{|projected_tier_distance - realized_tier_distance|}{|projected_tier_distance|+\epsilon} ]

Measures instability of tier assumptions embedded in routing.

5.4 FDT — Fee Drift Tax

[ FDT = \frac{C_{fee_mispricing} + C_{reroute_distortion}}{executed_notional} ]

Primary KPI for this regime; monitor p95 and worst-episode values.

6) State machine and controls

ALIGNED

• fee versions synchronized,
• tier model within tolerance,
• full route optimization enabled.

DRIFT_WATCH

Triggered when FVG or NCE exceeds watch thresholds.

• increase uncertainty penalty on fee-sensitive routes,
• reduce fee alpha weight in route objective,
• trigger immediate config consistency check.

MISPRICED_ROUTING

Triggered on confirmed version mismatch or sustained FDT breach.

• freeze fee-optimized routing branches,
• shift to impact/liquidity-prioritized conservative map,
• cap venue concentration,
• enforce short refresh interval + checksum validation.

SAFE_RECONCILE

• apply canonical fee version globally,
• recompute tier state from authoritative ledger,
• gradual return with hysteresis (require stable FVG/NCE window).

7) Practical modeling workflow

1. Build a canonical fee ledger
   • versioned, timestamped, venue+flag+tier resolved.
2. Join execution decisions with fee version used at decision time
   • not just settlement-time fee truth.
3. Label drift episodes
   • aligned / watch / mispriced / reconcile.
4. Run counterfactual routing replay
   • same market path, corrected fee table.
5. Estimate incremental drift cost
   • by branch, venue, and urgency bucket.
6. Calibrate policy thresholds
   • FVG/NCE/TMS triggers that minimize p95 FDT while preserving completion.

Key principle: evaluate policy on tail outcomes, not average fee delta.

8) 30-day rollout plan

Week 1 — Fee truth contract

• define canonical fee schema (venue, flag, tier, effective time, timezone),
• add version checksum to every routing decision log,
• enforce explicit “unknown fee state” value (never silent default).

Week 2 — Shadow diagnostics

• compute FVG/NCE/TMS/FDT without control changes,
• identify top venues and symbols with highest drift tax,
• validate fee-version propagation latency.

Week 3 — Guardrail activation

• activate DRIFT_WATCH penalties in small canary slice,
• auto-trigger consistency reconciliation on sustained mismatch,
• compare p95 FDT and schedule-completion against control slice.

Week 4 — Full deployment + runbook

• roll out mispriced-routing containment globally,
• add paging for stale-version exposure beyond SLA,
• codify incident playbook: detect → contain → reconcile → postmortem.

9) Common anti-patterns

• Treating fees as static constants in routing objective.
• Logging only settlement fees, not decision-time fee assumptions.
• Assuming one global tier state for all strategy flows.
• Mixing timezone cutovers between exchange calendar and config deployment.
• Optimizing maker rebates while ignoring drift-induced completion catch-up cost.

10) Bottom line

Fee economics are part of market microstructure, not back-office trivia.

If routing uses stale fee truth, “best execution” logic can be directionally wrong even with perfect impact models. A version-aware fee ledger, drift metrics, and containment states convert invisible fee leakage into a controllable risk process.

When fee truth is uncertain, route for robustness first, optimization second.