Dataset and analysis range

For ordinary analysis, name the inputs with dataset.name. entity, session and a time range can be omitted: the two identifiers inherit name, and the range comes from actual observations. Files are declared in sources, not selected by the dataset name. One analysis describes one entity in one common session; use batch tasks for independent tests. These identifiers appear in results.

Configuration location and example

This root-level mapping sits alongside sources and signals. Start with analysis.yaml and replace its existing dataset: section; do not add a second dataset, or put a list marker before it.

yaml

dataset:              # Name this analysis; other dataset fields are optional
  name: bench_run_01  # Identifies the dataset in results

Set entity and session only when equipment and acquisition runs need separate labels. They do not select files or prove that sources are synchronized. Add domain only to restrict the time range. The reference table describes choices for those needs, not a checklist of required fields.

Analyze only 2 to 5 seconds

Replace the dataset section of analysis.yaml:

yaml

dataset:        # Name the analysis and restrict its requested time range
  name: energy  # Dataset name included in results
  domain:       # Use the same time basis as the sources
    start: 2s   # Include this starting instant
    end: 5s     # Exclude this endpoint from continuous duration

The reference calls these positions dataset.domain.start and dataset.domain.end: find dataset, then domain, then start or end. The actual keys do not contain dots. domain: {start: 2s, end: 5s} is the equivalent inline mapping. See configuration syntax.

The synthetic ASC records power 2, 4, 8 and 10 W at 0, 1, 3 and 4 s, with an unrelated message at 6 s establishing observation through that time. Selecting 2–5 s gives 4 W during 2–3 s, 8 W during 3–4 s and 10 W during 4–5 s: 4×1 + 8×1 + 10×1 = 22 J. The real samples within the range are 8 W at 3 s and 10 W at 4 s, giving sample mean 9 W and time mean 22/3 W. The held value at 2 s contributes to integration but is not an additional sample. These numbers apply to the supplied synthetic ASC.

Field reference

Full path	Type, required status and default	Purpose and constraints
`dataset.name`	Required string	Name these inputs/results, for example `run_01` and `run_02`. The program retains your choice rather than deriving it from filenames. Identifier syntax is below.
`dataset.entity`	Optional string; defaults to name	Equipment or specimen label, such as `motor_01`, taken from project records. It does not select a DBC or convert units.
`dataset.session`	Optional string; defaults to name	Label the common acquisition session, such as `warmup`. Joint analysis has one session but still needs source clock alignment.
`dataset.domain`	Optional mapping	Requested range containing both `start` and `end`. Omission uses actual observations. Neither `null` nor `{}` means the whole file.
`dataset.domain.start`	Required time string when domain is supplied	Inclusive start, in calibrated source time: for example `0s` or `2026-01-01T08:00:00+08:00`. Apply source offsets before comparing clocks.
`dataset.domain.end`	Required time string when domain is supplied	End not earlier than start, excluded from continuous calculation. 0–6 s means six seconds; equal endpoints mean zero duration. Real samples at the endpoint may inform boundary decisions but do not extend duration.
`dataset.metadata`	Optional mapping; defaults to `{}`	Project-defined notes, such as `{operator: tester_01}`. Omission records no notes and changes no input, time or calculation.

The first three identifiers start with an English letter or underscore, followed by letters, digits or underscores. Python keywords and reserved expression names such as abs and isclose are disallowed. Empty strings, nulls and numbers are not identifiers.

0s is a time string with a unit, 0 is a number, and null is an empty value. CSV time columns may use bare numbers interpreted by sources[].time.unit; analysis endpoints require an explicit time representation. Source time.offset is applied before comparing input time to the requested range.

Requested versus observed range

A request for [0,12) with observations ending at 8 s does not make 8–12 s valid. The program retains the requested range and missing-data information; valid_for cannot extend observation. Execution and calculation settings determine whether partial data is accepted.

Omit the analysis range

Both ordinary analysis and per-message accumulation allow domain to be omitted. The program obtains first/last observation times during the same read, without a preliminary file scan or storing complete history just to determine the range. Rules, sequences, windows, calculations and requested exports use it.

If source A ends at 8 s and B at 5 s, the session extends to 8 s while B's missing data is handled separately. Empty inputs produce no events and null session statistics with no_observation; an always-true rule does not invent an event. A single record with no later observation has equal start/end times: no sample lies inside the half-open range and no duration can accumulate. The range is not extended to the signal's expiry time.

Supply domain when analyzing a subset; null and an empty object are not omission. Source time declarations and actual synchronization evidence still determine clock alignment.