Roll-ups & Derivations

Real-time aggregation and computed streams. Data arrives, statistics compute, formulas evaluate — automatically.

Roll-ups compute statistics — time-weighted average, sum, count, min, max, gap count, and more — automatically as data arrives. Derived streams take it further: Excel-style formulas that reference other streams, roll-up statistics, and cross-component values to produce new computed streams in real time.

Real-Time Roll-ups

A roll-up calendar defines the cycles — 1 minute, 5 minutes, hours, days, months, years — that the engine maintains for every stream that references it. Each cycle is independently configurable (interval size, time zone, reference date). Statistics compute and store automatically as data arrives; queries read pre-computed values instead of scanning raw samples.

Roll-up calendar editor showing the standard cycle hierarchy: 1 Minute, 5 Minutes, Hours, Days, Months, Years — A calendar lists the cycles available to streams that reference it.

Cycle definition editor showing fixed-size-interval configuration for the 'Days' cycle — interval size, unit dropdown, time zone, reference date — Each cycle is configurable — interval size, unit, time zone, reference date.

Pick a Cycle. Pick a Statistic. Done.

Anywhere a stream is referenced — charts, dashboards, derivations, GS SQL — you can request a specific roll-up cycle and statistic. The engine returns the pre-computed value: time-weighted average, sum, count, min, max, gap count, and more.

Widget configuration showing roll-up cycle and statistic dropdowns — cycles like 1 Minute, 5 Minutes, Hours, Days, and statistics like TWA, sum, min, max, count — Selecting a roll-up cycle and statistic in a dashboard widget — same selectors appear in chart configs, derivation inputs, and GS SQL `CycleId`/`Stat` parameters.

One Stream. Every Cycle. Live.

The same stream rendered four ways — hourly, daily, monthly, yearly — each widget pulling a different cycle and statistic from the same source. The roll-ups are pre-computed and refresh as new samples arrive; the dashboard just reads them.

Solar Farm Demo dashboard with four Big Number widgets all sourcing 'Site KPIs → Total Generation (kW)' — each widget configured with a different roll-up cycle (hourly, daily, monthly, yearly) and showing live values — One source stream — `Site KPIs → Total Generation (kW)` — rendered as four Big Number widgets at different cycles. The widget config (top) maps directly to the live values (bottom).

Time Filters — Time of Day, Day of Week, Season

Time filters narrow a roll-up to specific slices of the calendar — time of day (business hours, on-peak, off-peak), days of week (weekdays vs weekends), and even seasons (e.g. Winter On-Peak for utility rate analysis). Configure them once and reuse them anywhere a temporal window appears, including as parameters inside GS SQL Range() clauses.

Time filter editor showing time-of-day, day-of-week, and seasonal filters — including a 'Winter On-Peak' filter for utility rate periods — A time filter narrows roll-ups to time-of-day, day-of-week, and seasonal slices. The same filter is available as a temporal parameter in GS SQL.

Temporal Alignment

Each stream has its own independent time axis. Roll-ups project mismatched time axes onto a common temporal grid, enabling cross-stream correlation with simple equi-joins on entity identity and aligned timestamps.

Chart overlaying two solar streams — generation and consumption — both rolled up to hourly time-weighted average, aligning naturally on a common time axis — Two streams with different native sample rates — both rolled up to hourly time-weighted average, plotted on a single common time axis. No interpolation, no NULL-filling.

Derived Streams

Write formulas like spreadsheet cells — they evaluate as data arrives and produce new streams that can be charted, queried, and used as inputs to further derivations.

Derivation editor for a Site Meter revenue stream — formula references the on-peak rate via the rate_fk link stream and the meter's export, recomputing as either source changes — **Cross-component:** a Site Meter revenue formula that follows the meter's `on_peak_rate_fk` link to read the current on-peak rate and multiplies it by the metered export.

Derivation editor for an Inverter Efficiency stream — formula computes the ratio of AC output to DC input from streams on the same component — **Same-component:** an Inverter Efficiency formula that divides AC output by DC input — updating in real time as new samples arrive.

Both editors expose the same primitives: stream references, roll-up cycle & statistic selectors, and the full Excel-style function library. Pick a raw value, a TWA, an hourly sum, a daily max — the engine routes the request to the pre-computed roll-up.

Chained Derivations

Derivations can chain across the schema. Inputs are GS SQL queries that follow link streams from one component to the next — a tank reads its current batch, the batch reads its recipe, the recipe supplies the high and low temperature targets. When the tank is reassigned to a new batch, the derivation automatically picks up the new recipe's targets — no code change, no pipeline rebuild.

Derivation editor for a brewery Tank temp_in_range stream — hi and lo input variables are GS SQL queries that traverse from tank through batch_fk to batch and then through recipe_fk to read temp_max and temp_min from the recipe — A brewery tank's `temp_in_range` derivation follows FK links from `tank` → `batch` → `recipe` to read the recipe's allowed temperature range. The link streams are themselves temporal, so historical evaluations use historically-correct recipes.

hi input variable

SELECT recipe_fk, 'temp_max'
FROM batch
WHERE _component_uid = (
  SELECT batch_fk
  FROM tank
  WHERE _component_uid = @_component_uid
)

lo input variable

SELECT recipe_fk, 'temp_min'
FROM batch
WHERE _component_uid = (
  SELECT batch_fk
  FROM tank
  WHERE _component_uid = @_component_uid
)

Each input is a two-hop traversal: tank.batch_fk → batch._component_uid, then batch.recipe_fk → the recipe component's temp_max or temp_min stream. @_component_uid binds to the tank that owns the derivation. The result column tells the engine which stream to read from the resolved component.

Inspect the Graph & Diagnose Live

Right-click any derived stream and choose Derived Stream Diagnostics to see its dependency graph and step through the live computation. Trace which inputs resolved to what values, where roll-ups were sourced, and how each stage produced its output — in real time, against real data.

Derived Stream Diagnostics view showing the derivation's dependency graph and live evaluation trace — inputs, roll-up source values, intermediate calculations, and the resolved output — The dependency graph and live diagnostic trace for a derived stream — available from the right-click menu on any derivation.

Compute in Real Time

From raw data to rolled-up statistics to derived insights, automatically.