Transparent methods. Meaningful analysis.

Insider Magazin analyses datasets made of time-stamped observations. Each observation has an identifier, a calendar date, a numeric value, a category (such as a region or channel) and a source label. Datasets also declare their unit, the precision values are recorded with, and whether values are continuous or discrete.

Customer data is connected through a server-side data adapter that reads from an HTTPS API configured during onboarding. Credentials remain on the server and are never sent to the browser. The public demo uses illustrative datasets: synthetic series generated from documented formulas with a fixed random seed, labelled as such wherever they appear.

Every dataset is validated against a schema before it reaches the workspace. A response containing a malformed record is rejected as a whole rather than partially loaded, so the analysis never silently runs on incomplete data.

• Dates must be ISO calendar dates (YYYY-MM-DD) and are interpreted as UTC calendar days. Weeks follow ISO 8601 (Monday start).
• Values must be finite numbers; they are stored at the dataset's declared precision.
• Categories are taken from the dataset metadata; filters apply to them exactly.
• Coverage is reported as the number of calendar days with at least one observation divided by the number of calendar days in the selected period.

All statistics are computed in the browser from the observations in the current selection. The definitions below are the ones used throughout the platform.

Worked example

For the values 2, 4, 4, 4, 5, 5, 7, 9 (n = 8): the mean is 40 / 8 = 5; the squared deviations sum to 32, so s² = 32 / 7 ≈ 4.571 and s ≈ 2.138. The median interpolates between the 4th and 5th ordered values, (4 + 5) / 2 = 4.5. Q1: h = 7 × 0.25 = 1.75, so Q1 = 4 + 0.75 × (4 − 4) = 4; Q3: h = 5.25, so Q3 = 5 + 0.25 × (7 − 5) = 5.5.

Values are grouped into contiguous classes [a, b). The last class is closed so the maximum is always counted. For each class the platform reports the absolute frequency, the relative frequency (count / n), and their cumulative forms.

When frequencies are compared across periods (for example by year or quarter), each period is expressed as relative frequencies that sum to 100%. Comparing raw counts would favour periods with more observations.

• Histograms draw each class with its true width on a continuous axis. The density view divides relative frequency by class width so the total area equals 1.
• Time series show the mean of all observations in each period (day, ISO week, month or quarter). Moving averages are trailing: the value at period t averages periods t − w + 1 … t, and is left empty until w periods are available.
• Heatmaps use a single-hue sequential scale from light (low) to dark (high), with a legend and exact values in tooltips. Empty cells mean no data, not zero.
• Box plots show Q1, median and Q3; whiskers extend to the most extreme values within 1.5 × IQR of the box (Tukey).
• Lag-1 scatter plots pair each day's mean with the previous calendar day's mean; non-consecutive days are not paired. The Pearson correlation r is reported with the number of pairs.

Series colours were checked for colour-vision deficiency, and charts never rely on colour alone: every multi-series chart has a legend, and the underlying numbers are available in tables.

Two subsets (two periods, or two categories within a period) are compared on identical terms: the same statistics, and relative frequencies on the same class boundaries computed from both groups together.

Descriptive statistics summarise what has been observed. They are only as good as the data behind them and the questions asked of it.

• Past patterns do not guarantee future outcomes. Frequencies and trends describe history. For processes whose outcomes are independent (for example draws of chance), past frequencies carry no information about the next outcome.
• Selection matters. Statistics refer to the filtered selection only. A different period or category can give a different answer.
• Small samples are unstable. Quantiles, skewness and class frequencies vary considerably when n is small; check the sample size shown with each result.
• Association is not causation. Correlation and differences between groups do not identify a cause.
• Aggregation hides detail. Weekly or monthly means smooth over variation within the period.
• Data quality limits conclusions. Gaps in coverage, changes in how data was recorded, or errors in the source carry through to every result.

Insider Magazin does not produce forecasts, predictions, probability estimates of future events, or financial, investment or gambling advice.