Do Mixed Fixed & Random Time‑Slice Schedules Shorten Experiment Recovery? Simulation Insights

Introduction

Current time‑slice experiments are increasing, and without constraints they interfere with each other. Huolala uses fixed‑order carousel nesting to keep interference controllable.

Theoretical Basis

Time‑slice nesting is essentially an implementation of full‑factorial experimental design, where each factor represents a module strategy (e.g., broadcast radius and PK duration). To observe one factor while cancelling others, the experiment must rotate through all factor combinations with equal time.

The minimal experiment cycle equals the time needed to rotate through all possibilities with equal group duration, ignoring natural weekly fluctuations.

Impact of Time‑Slice Length and Order

Two design variables exist: time‑slice length (e.g., 5 min, 30 min, 1 h) and experiment order (random or fixed). For illustration we consider two factors (A/B and X/Y) each with two levels, requiring AX, AY, BX, BY to appear with equal duration.

When both factors have equal time‑slice length, a reversal of one factor’s order is needed to traverse all combinations. If lengths differ (m and n), equal‑duration combinations require 2·max(m,n) or more complex rules: if m=2n, length 2m; if m=3n, length 4m; otherwise 2·LCM(m,n) without order change.

Simulation Scenarios

We compare fixed‑order and random‑order carousel cross‑usage to see if it shortens the recovery cycle. Experiments include:

Fixed‑order carousel (baseline) with equal‑length slices.

Random‑order carousel combined with fixed‑order.

Scenarios with equal and unequal slice lengths, over 16‑day and 60‑day periods.

Results show that overall order differences produce 1‑4 percentage‑point variations in order volume, which are acceptable. Fixed‑order alone yields better homogeneity and shorter cycles than mixed approaches when slice lengths are equal. When lengths differ, mixed approaches achieve comparable homogeneity, but fixed‑order alone can also reach similar results.

Conclusion

Mixed fixed and random time‑slice scheduling does not noticeably shorten the observation period and may degrade homogeneity. For multi‑layer time‑slice nesting, fixed‑order carousel is preferable despite the need for manual orthogonal handling.