A conference talk arguing that line charts are a poor default for telemetry and observability dashboards. The speaker, an infrastructure engineer with years of on-call experience, walks through three core problems with line charts: misleading interpolation, noise from multiple instances, and the assumption that time is always the right axis. She proposes alternatives based on three principles: matching visualization to data shape (using smoothing, scatter plots, or local regression), matching visualization to telemetry type (step functions for counters, dots for gauges, heat maps for histograms), and matching visualization to the actual business question being asked (which often requires transforming time-series data into panel/tabular data with time as just another column). Practical examples include load-vs-latency scatter plots, version comparison distributions, and cloud SKU ROI tables. The talk closes with a call for observability tools to better surface data type hints, reduce friction for statistical exploration, and make it easier to export data from TSDBs into richer analysis environments.
Sort: