Metrics

Metrics lets your organization define its own key figures. You describe, in one short expression, what should be measured in a single 15-minute interval. The platform does the rest: it applies your expression to each plant, over the period a report covers, and folds the 15-minute values into the figure you want to see.

A metric is defined once for the whole organization and is then available on every plant — you never write a metric "for a plant".

Where to Find It

Open the Reports page of any plant and switch to the third tab, Metrics. The list you see is your organization's; the plant is simply the place where you look at it and test against.

  • Organization administrators create, edit and delete metrics.
  • Every member can see the metrics and use them in reports.

Defining a Metric

A metric consists of a few descriptive fields and one to three expressions.

New metric creates the metric straight away, and from then on every change is saved as you type — a status chip shows Saving and then Saved, and there is no separate step to confirm. Until its expression validates, the metric stays a Draft: the table marks it with a Draft badge, and a draft is never offered as a data source in a report. The moment the expression validates, the badge disappears and the metric can be used everywhere.

  • Name, description, unit — how the metric appears in the report data picker and in the finished report.
  • Expression — what is measured in one 15-minute interval. This is written in MetricsQL, the query language of the platform's time-series store. You write only the per-interval part: the platform adds the plant, the time window and the folding. Never name a plant or an organization in the expression — doing so is rejected, because the same metric has to work on every plant.
  • Fold — how the 15-minute values become one number:
    • mean — the average of the interval values
    • sum — the total of the interval values
    • ratio — the share of expected intervals in which the expression was 1
    • last — the value at the end of the bucket
  • Expected (optional, for the ratio fold) — a second expression that decides which intervals count at all, for example "it is daylight". An interval with no expectation is ignored; an interval that was expected but has no measurement counts as not available; a period with no expected interval at all is shown as a dash.
  • Weight (optional) — a static per-component value, such as a component's peak power, used to combine the component values into one plant value. Without a weight, components count equally.
  • Report as — the shape the metric answers in. Plant value gives one number per plant; where the expression works per component, that number is the weighted roll-up described above. Per component gives one column per inverter, combiner box, string or logger, plus the plant value. Duplicate a metric to have the same figure in both shapes.
  • Finest resolution — the finest step a report may print the metric at: 15 minutes, hourly or daily. Coarser steps are always available; choose Day for a figure that only makes sense per day.

You also choose the resolution the metric is reported at: 15 minutes, hourly, daily, monthly, or a single value for the whole period.

While you type, the editor validates live — syntax, unknown series (with suggestions for what you probably meant), the self-scoping rule, whether the expressions fit the chosen fold, and size. A test run then evaluates the metric on the current plant, and a preview shows a table and a graph with every inverter or component lane and the resulting plant value for a month, quarter or year you pick — a metric reported as a plant value shows that one value alone. Validation and test messages name the plant they were checked against.

The Building Blocks

The editor's side panel lists everything you can build from, so you rarely have to remember a name:

  • Series — the same data as the report's data selection: plant feed-in and production, expected production and reference series, compensation and curtailment, inverters, combiner boxes, strings, battery, weather and infrastructure, each with the labels and steps it carries. Search by name, filter by level, and a mark shows which series the current plant has not delivered yet — the metric still works organization-wide. What a report computes rather than reads (performance ratio, event counts, settlement figures) is listed at the bottom as not expressible, so you know before you write it.
  • Configured facts — values from the plant configuration, not measurements: inverter, combiner-box and string DC peak power, plant peak power, the plant inverter capacity, and component counts. They are resolved into the query for each plant when it runs, so an expression uses the name like a constant series — powerplant_inverter_peak_power to weight or threshold per inverter, for example. A plant where no component has the value configured shows an empty preview; the metric itself works on every plant that has them. Two of them are valid per period rather than constant: the plant's effective EEG applicable value — the capacity-weighted anzulegender Wert of the bands in force at that moment — and the Marktwert the plant is settled against, monthly or annual according to its market configuration, both in EUR/MWh. They change over time, and the platform inserts the value in force at each 15-minute slot, so they need no window of their own. A market premium built from the two — clamp_min(powerplant_park_eeg_value - market_marktwert, 0) — is a blended approximation for reading trends: the settlement prorates every band separately, so it is never the settled figure.

Shipped Examples

The Metrics table already holds five metrics that Mirox ships to every organization: Connectivity, Logger availability, Inverter availability in daylight, Sun hours and Feed-in data availability. They carry a lock and an Example badge — they belong to no organization, so they cannot be edited or deleted.

What you can do with them:

  • Open one and read its expressions. They are worked examples of everything described above: the daylight gate, both shapes of the ratio fold, the per-component identity, a unit scale.
  • Run the preview on the plant you are on, to see the individual lanes and the plant value they add up to.
  • Duplicate one. The copy belongs to your organization, is fully editable, and is the usual starting point for a metric of your own.

The examples are deliberately the simple, transparent version of each question, and every description names what it does not account for. The report's own availability figures — which use the plant's configured logger roster and its settlement methods — stay where they were, under Availability in the report data picker. That is why the examples themselves are not offered in the picker; a duplicate of one is.

Limits

  • An expression may be up to 4 000 characters; after the platform adds the plant, the store rejects anything longer than 16 KB.
  • A metric may answer with at most 500 series on a single plant.
  • The test run over one day must finish within 5 seconds.
  • Nothing can be measured finer than 15 minutes.

If a metric hits one of these, the editor says so as you type, and the metric stays a draft until you resolve it.

A Worked Example

Inverter availability (2 % of DC peak) — the share of daylight intervals in which an inverter produced at least 2 % of its DC peak power, weighted by peak power, reported as a percentage with the ratio fold.

Expression:

max by (inverter_id) (max_over_time(powerplant_inverter_power_ac[900s])) > bool on(inverter_id) 0.02 * powerplant_inverter_peak_power

Expected (which intervals count — there is usable light):

(avg(avg_over_time(powerplant_pyranometer_radiation_power{segment_id!~".+"}[900s])) or avg(avg_over_time(powerplant_weather_radiation_power{segment_id!~".+"}[900s]))) > bool 100

Weight (so large inverters count more in the plant value):

powerplant_inverter_peak_power

Creating a Metric with the Assistant

You do not have to write MetricsQL yourself. The assistant sits inside the metric editor: open New metric — or any metric you are allowed to edit — and it is the chat panel at the top right. It stays quiet until you write to it; nothing is sent before that.

Describe the figure you want in your own words. The assistant already knows the plant you are on, what you have filled in so far (name, expression, fold, …), the result of the last validation and which series this plant actually writes, so it rarely has to ask the basics. It proposes an expression, a name and a description, validates them and previews them on the plant, and tells you when the platform does not hold the data you asked for or when the query would be too heavy.

A proposal goes into the fields the moment it arrives — the assistant edits with you, and each change is saved as you make it. If you would rather keep what you had, Revert on the proposal in the chat puts the fields back the way they were before it; the proposal stays in the chat and can be applied again. The assistant finishes on its own: a proposal is stored on the metric the moment it is made, so you can send your request, close the page and find the finished metric in the list when you come back. After that it is ordinary editing: change any field, run the preview again, or ask for something different.

The conversation belongs to the metric, not to the browser tab. Close the editor and open the metric again, or reload the page, and the same chat is there beside the fields it wrote.

The chat in the editor takes text only. Photos and documents — the formula page out of a contract, for example — remain a feature of the full chat page; Open in chat carries the conversation over there.

Using a Metric in a Report

In the report editor, open the data picker of a block and choose the level Organization metrics. Your metrics appear there next to the platform's own items and behave exactly like them — the same resolutions, the same tables and charts. Because a metric is organization-wide, the same block works unchanged in a report for any plant.

What Cannot Be Defined

Some figures are computed by the platform itself and must be taken from the existing items rather than rebuilt:

  • Performance ratio and expected production — use the platform-computed items, which are also available as building blocks.
  • Curtailment settlement — derived from your market and tariff configuration.
  • Per-inverter AC nominal power — not stored per inverter, so it cannot be referenced.
© 2026 Mirox Verwaltungs GmbH. All rights reserved. | Privacy Policy