SENTINEL: Time Series Quality Analysis

SENTINEL is the starting point of the Kronts data pipeline. Upload a time series and SENTINEL automatically produces a quality dashboard — a visual, data-first report that tells you what your data looks like, where its problems are, and what to do about them before any processing begins.

What SENTINEL Does

When you upload a file, SENTINEL runs four independent quality analyses simultaneously:

Dimension	What it measures
Completeness	Whether all expected readings are present — gaps, outages, and sampling consistency
Accuracy	Whether values are plausible — statistical outliers, noise level, and transient spikes
Consistency	Whether the signal behaves predictably over time — trends, stationarity, and dominant cycles
Validity	Whether the distribution of values makes sense — shape, skewness, and parametric fit

Each dimension produces an independent quality rating (Excellent / Good / Fair / Poor) and a set of recommendations. These feed directly into FORGE, which uses them to suggest and apply appropriate data-cleaning operations.

Supported File Formats

SENTINEL accepts time series files in the following formats:

CSV — Two columns: timestamp and value. The timestamp column is auto-detected by name.
Excel (.xlsx) — Same two-column layout, first sheet used.
JSON — Array of {timestamp, value} objects, or a {timestamps: [...], values: [...]} dict.

Timestamps can be in ISO 8601 format (2024-01-15T09:32:00Z), Unix epoch (seconds), or most common human-readable date/time strings. Mixed timezone representations are normalised automatically.

Upload Limits

User type	Maximum points
Standard	10,000 points
Superuser	10,000,000 points

The Quality Dashboard

After upload, the dashboard shows:

Main chart — An interactive time series plot with progressive zoom-to-load for large datasets. Supports date/elapsed time axis, markers/lines/both display modes, and solid/dash/dot line styles.
Four quality panels (right sidebar on desktop, below chart on mobile) — One per quality dimension, each showing a summary badge, key metrics, and a "View Full Report" button that opens a detailed modal.
Summary statistics — Point count, min, max, and mean value displayed in the footer.

Each panel has a Re-run Analysis button. Use this if you want to recalculate quality metrics after the initial upload — for example, after understanding your data better and wanting to confirm a suspicion.

Quality Ratings

Each dimension is rated on a four-level scale:

Rating	Badge	Meaning
Excellent	Green	No issues found. Data is ready for analysis.
Good	Blue	Minor issues present but the data is suitable for most analyses.
Fair	Amber	Noticeable issues that may affect some analyses. Preprocessing recommended.
Poor	Red	Significant issues. Preprocessing required before reliable analysis.

The Pipeline

SENTINEL sits between data ingestion and processing. Its outputs are designed to be consumed downstream:

Upload (CSV / Excel / JSON)
        │
        ▼
   SENTINEL
   ├── Completeness analysis  ──┐
   ├── Accuracy analysis        │  Quality ratings +
   ├── Consistency analysis     │  Recommendations
   └── Validity analysis     ──┘
        │
        ▼
     FORGE  (apply cleaning operations)
        │
        ▼
    CEREBRO (label events)

FORGE reads the recommendations from SENTINEL and presents the most relevant cleaning operations for your specific data quality issues.