Window

The Window operation keeps a rolling collection of recent numbers and reports one figure over it — the minimum, maximum, mean, sum, or how many samples it is holding. The collection is bounded either by time (everything from the last N seconds) or by count (the last N samples).

A count window of one sample aggregates only the latest sample, and the Editor marks Samples with an info mark when it is set to 1.

Inputs

Port	Type	Required	Description
>	Any	No	Optional gate. When closed, the path stops and no sample is recorded.
In	Integer or Float	Yes	The sample to add. A non-numeric or Nil value fails the Node; its On Error verb resolves the failure.

Outputs

Port	Type	Default	Description
Out	Float	Nil	The chosen aggregate over the collection as it stands after this sample was added.

Properties

Name	Type	Default	Description
Window Mode	StackedDataSet	Time	How the collection is bounded.
Time — Seconds	Integer	60	Keep samples taken within the last N seconds, 1–2147483647.
Count — Samples	Integer	10	Keep the last N samples. The upper bound is the broker’s configured sample cap.
Aggregation	Combo	Mean	Min, Max, Mean, Sum or Count.
On Error	Combo	Stop this branch	Per-op On Error verb (Stop this branch / Continue with the last value / Continue with the fallback). See the Error Handling guide.

State on Reload

Out is preserved across restarts. The collection is not: after a restart the window refills from the next sample onward.

A window only moves when a sample arrives

Window does nothing on its own. In Time mode, samples that have aged out are discarded when the next sample arrives, not the moment they expire.

So a producer that goes quiet leaves Out reporting the last aggregate it computed, however old the samples behind it are. Where a stale figure would mislead, pair the Window with a Scheduler so it keeps running, or check freshness with TimeStamp and Delta.

---

How It Works

1. Gate check — if the Gate is connected and closed, the path stops and nothing is recorded.

2. Read the input — In must carry a number. Anything else fails the Node, and its On Error verb resolves the failure.

3. Add — the value is appended to the collection with the moment it arrived.

4. Retire — in Count mode the oldest samples beyond Samples are dropped. In Time mode every sample older than Seconds is dropped. A broker-wide cap then trims whatever remains above it.

5. Aggregate — the chosen figure is computed over what is left, Out is written as a Float, and the path continues.

The current sample is always included

The sample that triggered the run is added before the aggregate is computed, so the answer always reflects the value you just fed in. A window holding one sample reports that sample for Min, Max, Mean and Sum, and 1 for Count.

---

Examples

Count window, Mean

Window Mode: Count = 3   Aggregation: Mean

Run	In	Held	Out
1	10	10	10.0
2	20	10, 20	15.0
3	30	10, 20, 30	20.0
4	40	20, 30, 40	30.0

Count window, Max

Window Mode: Count = 3   Aggregation: Max

Run	In	Held	Out
1	10	10	10.0
2	50	10, 50	50.0
3	20	10, 50, 20	50.0
4	30	50, 20, 30	50.0
5	30	20, 30, 30	30.0  (the peak has aged out)

Time window, Count

Window Mode: Time = 60   Aggregation: Count

Out is how many samples arrived in the last minute — a messages-per-minute figure that costs one Node.

---

Use Cases

• Smoothing a noisy sensor — a Mean over the last 60 seconds gives Compare or Hysteresis something steadier to act on than the raw reading.
• Peak detection — a Max over a rolling window holds the highest value seen recently without keeping a running maximum by hand.
• Measuring arrival rate — Count over a time window answers “how busy has this been” directly.
• Totalling over a period — Sum over a time window accumulates without a separate counter to reset.

---

Error Handling

Window fails when In is unconnected, Nil, or not a number; the Node’s On Error verb resolves the failure.

Verb	What it does
Stop this branch (default)	This Node and its dependents stop; Out keeps its value and reads as a failed update.
Continue with the last value	Dependents run on the value Out already holds, not stamped as new; it reads as a failed update.
Continue with the fallback	Dependents run on Out’s per-output Fallback, stamped as this run’s value; it reads as a failed update. The Editor refuses a Task where the Fallback is unset.

A rejected sample is not added to the collection.

---

Limitations

• The window bound and the aggregate are fixed at configuration time; neither can be driven from an input port. To report several aggregates over one signal, use one Window per aggregate.
• Only numbers can be collected.
• Out is always a Float, including for Count. Feed it through Convert when a downstream port requires an Integer.
• The collection does not survive a restart, so the first aggregates after one are computed over a partly refilled window.

---

Configuration

One host setting bounds Window. It is read at start-up from the host’s configuration file; see the Core configuration reference.

Key	Type	Default	Effect
core.windowMaxSamples	Integer	1000	The largest count window a Window node may request.

The value is carried in the operation’s schema as the upper bound of the sample count, so the editor constrains the setting to it and states the accepted range.