Aggregations & Sliding Windows (Part 1)

Overview

This skill explains how aggregations over sliding windows work in the context of compliance and risk evaluation.

This is the core technical challenge of Part 1.

Key Concepts

• •Transactions are immutable events
• •Aggregations are computed at evaluation time
• •Windows are based on event timestamps, not system time

Examples of aggregations:

• •Count of transactions in the last 24h
• •Sum of amounts in the last 7d
• •Unique countries used in the last 30d

Sliding Windows

A sliding window:

• •Is defined by a duration (e.g. 24h, 7d)
• •Is anchored to the transaction being evaluated
• •Includes only past transactions

Never use wall-clock now() for evaluation logic.

Design Rules

• •Aggregations MUST be deterministic
• •Aggregations MUST be reproducible
• •Aggregations MUST NOT be stored as rolling state

Each evaluation recomputes what it needs.

Anti-Patterns

Avoid:

• •Precomputing counters on accounts
• •Background jobs to maintain aggregates
• •Storing derived metrics as source of truth

Expected Outcome

After applying this skill:

• •Aggregations are correct and explainable
• •Time-based logic is predictable
• •Historical evaluations remain valid