Signal CSV export and resampling

outputs.signal_csv exports updates of selected original signals. You can additionally request regularly spaced values estimated by linear interpolation. Export settings do not change the analysis integral. See Outputs for common scopes, publication, and optional device labels.

Complete minimal and grid examples

export-signal-minimal.yaml reads power from the supplied quickstart.asc and teaching.dbc. Its integral remains 38 J; it additionally exports the four original updates at 0, 1, 3, and 4 seconds within [0,6). Sources and signals are defined. The device label is optional; omitting it does not prevent export.

export-signal-grid.yaml adds a 500 ms grid and max_gap=3s, producing 12 points from 0 to 5.5 seconds. The estimate at 0.5 seconds is 3 W. The 4.5, 5, and 5.5 second points have no right-hand update inside the range, so remain missing instead of being extrapolated. The integral is still 38 J. Run chronaxis run FILE.yaml --output RESULT_DIRECTORY, read the manifest returned by the CLI, then use its exports.index_file and exports.directory to locate this export's index and original.csv or resampled.csv.

Export original signal updates

The following fragment assumes signals already binds the numeric signal voltage and the export label requirement is met. Start from dual-grid.yaml, find root outputs, and add or replace only its signal_csv child. Keep the other output settings and unshown root fields; do not paste a duplicate outputs key.

yaml

outputs:                # Additional result files to publish
  signal_csv:           # Export selected signals as CSV
    signals: [voltage]  # Use the signal name from the analysis definitions.

Add a regular output grid

Replace that signal_csv child with the following. The two fragments are alternatives, not two blocks to append together.

yaml

outputs:                     # Additional result files to publish
  signal_csv:                # Export selected signals as CSV
    signals: [voltage]       # Names of the analysis signals to export
    resampling:              # Optional output grid and value selection
      continuous: [voltage]  # Only signals that can meaningfully be interpolated.
      period: 250ms          # Spacing of exported values.
      max_gap: 2s            # Largest permitted gap between actual observations.

These fields are relative to outputs.signal_csv. Its signals field selects names; it is not the root list that defines signals.

Field	Type	Required	Default	Meaning and constraints
`signals`	String list	Yes	None	1–4096 distinct original signal aliases from `signals[].name`, e.g. `["voltage","current"]`. Derived signals cannot be selected directly.
`scope`	String or object	No	`session`	Whole session, or `{window: context}` for a named window. See Export scope.
`resampling`	Object or `null`	No	`null`	Omitted/`null` keeps original updates; an object enables the regular grid described below.

Without resampling, original.csv contains one timestamp and one signal update per row. It does not force signals onto the same sampling frequency. Invalid values leave value empty, with a reason recorded in quality; different updates at the same time remain separate. Its public header is time_ns,signal,value,unit,quality, with no valid, status, or source-reference columns.

With resampling, signals listed in continuous go to resampled.csv; other selected signals retain their original updates in original.csv. continuous must be a numeric subset of signals. Boolean, enum, state, and counter values should not be treated as continuous physical quantities.

The following fields are relative to outputs.signal_csv.resampling:

Field	Type	Required	Default	Meaning and constraints
`continuous`	String list	Yes	None	Nonempty, distinct subset of selected original signals that can physically be linearly estimated, e.g. `["voltage"]`. A numeric representation alone is not sufficient justification.
`period`	Positive duration string or mapping	Yes	None	Grid spacing, e.g. `250ms`, or `{mode: dbc}` to obtain a DBC period. The mapping may specify a `driver` from `continuous`; it must identify a DBC message with a valid period.
`start`	Time string	No	`"0s"`	Reference grid origin. `start=50ms`, `period=100ms` gives 50, 150, 250 ms, etc.; it does not restart at file or batch boundaries.
`max_gap`	Duration string	Yes	None	Positive maximum actual interval between the two interpolation records. With `2s`, a 2-second gap is allowed but a 3-second gap leaves the value missing.

A mapped period accepts these fields. This output grid is independent of root intervals.

Full path	Purpose	Choices and omission behavior
`outputs.signal_csv.resampling.period.mode`	Select where the grid period comes from.	Only `dbc`. For a fixed period, write `period: 250ms`, not `mode: fixed`.
`outputs.signal_csv.resampling.period.driver`	Select the message period through one original signal in `continuous`.	All continuous signals must originate from raw CAN. Omit only when there is one physical-message candidate; multiple candidates require a driver. Explicit `null`, derived signals, or names outside `continuous` are invalid.

The points arranged from start and period form the time grid. Each point has one of three statuses:

`status`	How its value is obtained
`observed`	A valid actual update exists exactly at that time.
`estimated`	No update exists at that time, but valid records on both sides are no farther apart than `max_gap`; the value is interpolated on their straight line.
`missing`	A left/right record is absent or invalid, same-time records conflict, or the gap is too large. `value` is empty and `reason` explains why.

For a 100 ms grid with valid updates only at 10 and 120 ms, 100 ms can be estimated; 0 ms cannot, because it has no left-hand record.

With conflicting records for one signal at one time, error marks the point and affected interpolation missing with reason=same_time_conflict if that signal is used only by the export grid. If it also participates in ordinary calculations or rules, error rejects the conflict and fails the analysis. dual-grid.yaml also integrates voltage, so it follows the latter behavior.

To choose the last record by stable source sequence, explicitly set signals[].same_time_policy to last_by_sequence. This does not choose by numerical value or thread completion order. The resampled file's public header is time_ns,signal,value,unit,status,reason,quality; it does not include endpoint timestamps or source references.

Complete example using DBC periods

Download analysis.yaml, grid.asc, and grid.dbc into one directory and run chronaxis run analysis.yaml --output grid-result.

The example exports power and voltage. The DBC declares a 1000 ms period for the Power message and 500 ms for Voltage; input updates occur at 0, 1, and 2 seconds. Signal CSV uses one grid, so period.mode=dbc and driver=power explicitly select 1000 ms. Removing the driver is an error because the continuous signals belong to two physical messages; the program does not choose arbitrarily.

CAN reconstruction uses each message's own DBC period, so that export uses period: {mode: dbc} without a driver: Power every 1 second and Voltage every 0.5 seconds. Selecting a signal-grid driver does not alter CAN message periods. The observation range is [0,2); inspect both data and manifest generated/skipped counts for endpoint behavior.

The signal CSV has four rows: at 0 and 1 seconds, power is 10/20 W and voltage is 20/30 V. The reconstructed ASC has five frames: Power at 0 and 1 seconds, Voltage at 0, 0.5, and 1 seconds with 20, 25, and 30 V. Voltage's 1.5-second point needs the update at 2 seconds, which is outside the half-open request range; that point is skipped. Different row counts follow from the two independent grid rules.