Free sensor entropy audit

Drop a CSV. See which channels moved.

Every column in your file is profiled for entropy across the record, and the ones that changed most are flagged. Takes seconds, needs no account, and runs entirely in your browser — the file is never uploaded anywhere. Open your network tab and check.

Drop a CSV here, or click to choose
Any time-series, sensor, process or historian export. Stays on your machine.

Method

Each numeric column is converted to fractional changes between consecutive readings. Those changes are binned into five buckets at ±0.0005 and ±0.002, and Shannon entropy is computed over the bin distribution, normalised to 0–1. Low entropy means the channel is moving in a structured way; high entropy means it is behaving like noise. The record is split in thirds and the first third is compared against the last. Columns that are monotonic counters — timestamps, row indices, sample numbers — are detected and excluded, since their "entropy" describes the clock, not an instrument.

Those bin thresholds are not arbitrary — they were tested against 10× tighter, 100× tighter and 10× wider alternatives on public benchmark data, and performed best. Columns that are constant, non-numeric, or too short are skipped and reported as such rather than silently dropped. A channel marked "flatlined" is worth looking at — a sensor reporting one unchanging value across the whole record is either out of service or stuck, and neither shows up as an alarm.