Role:
Sole designer & engineer
Runs:
Entirely in the browser
Status:
Live at das.dixon.cx
DAS: Data Analysis System
A private workbench for turning spreadsheets into explainable incidents. Load CSV or Excel data, let DAS find the right detector, and get a graph that shows exactly where the data changed: with a plain-English reason for every highlighted point.
- Data analysis
- Anomaly detection
- Browser-based
How it works
The file never leaves your browser, which makes it usable for production extracts, customer exports and sensitive operational data that shouldn't be uploaded anywhere. Press Identify and DAS studies the selected columns for the shape of the problem: timestamp gaps, frozen runs, flapping, volatility bursts, seasonal rhythm, level shifts, sudden shocks, one-sided excursions and peer outliers, then scores the likely detector families and tunes the winning configuration. The recommendation isn't a black box: DAS shows its confidence, the signals it found and the nearest alternatives.
Every breach gets a 0-100 score built from separate weighted components: why it fired, how long it lasted, the worst-sample magnitude and how many columns were affected, with a short plain-English narrative alongside the evidence table. Where a dataset includes a 0/1 incident label column, DAS can also estimate the probability that each row is an incident from the features you select, with readiness, model-quality scores and per-row contributions kept auditable.
The detector library
Static threshold
Flags values above, below, inside or outside a fixed boundary: SLA breach, temperature ceiling, disk usage.
Rolling baseline
Compares a recent window with a longer local baseline, for values that have shifted from what was normal a moment ago.
Sustained degradation
Ignores a lone blip, fires only after a run of bad samples: slow regressions, persistent latency.
Lag comparison
Compares the current window with an earlier same-length window, for period-over-period drift.
Seasonal comparison
Judges a point against the same offset in prior cycles, for daily and weekly rhythms.
Rate of change
Finds sudden point-to-point shocks, where the speed of change is itself the warning.
Volatility burst
Detects a metric becoming erratic by comparing recent dispersion with normal dispersion.
Flapping
Counts repeated crossings around a boundary: chattering health checks, unstable services.
Stuck metric
Catches long runs of identical or near-identical values: a sensor silently freezing in place.
Missing data
Looks at timestamp cadence and flags gaps: sometimes the incident is an absence, not a bad value.
Cross-series outlier
Compares one column with its same-time peers, for a fleet where one instance drifts from the pack.
Binomial regression baseline
Analyses success/total proportions with their underlying counts, so scale is never lost.