Overview

Add native aggregation support for the Largest Triangle Three Bucket (LTTB) downsampling algorithm. This is a standard technique for reducing large time series datasets to a fixed number of visually representative points, widely needed for IoT and telemetry workloads.

Background

LTTB is a well-established downsampling algorithm used extensively in time series visualization. It selects a subset of N points from an input dataset that best preserves the visual shape of the data — retaining spikes, trends, and inflection points while discarding visually redundant samples.

Competing databases offer LTTB natively. MongoDB users today must approximate it using complex multi-stage pipelines with $bucket, $group, and nested $reduce expressions. This workaround has significant limitations.

Proposed Interface

The implementation could take the form of a new aggregation stage and/or an accumulator. Exact interface TBD during design. At minimum it should accept:

• t — the x-axis (time) field expression
• y — the y-axis (value) field expression
• n — target number of output point

References

• Steinarsson (2013): Downsampling Time Series for Visual Representation (original LTTB paper, University of Iceland)