Time Zones

Storage is always UTC: a datetime is epoch seconds, a date is epoch days. A time zone only changes how an instant is read (local calendar day, wall clock) or rendered (export). Zone names are UTC or an exact IANA Area/Location from Pulse’s own embedded tz database, so results never depend on the host; anything else is PULSE_TIMEZONE_UNKNOWN. The manifest reports the release as tzdata_version (see Refreshing the Time-Zone Database).

Choosing a zone

Per zone-capable slot: slot tz, then request time_zone, then Options.DefaultTimeZone, then UTC. There is no environment variable or default-zone CLI flag. Predict echoes the resolved zone per slot in time_zones[]. An explicit tz on a date field or a non-zone-capable operator is PROCESSING_CONFIG; a non-UTC zone on a derived field is refused. UTC or no zone is byte-identical to the pre-zone output.

Local days in every mode

A datetime buckets and filters on its local calendar day in process, stream, compose, shards, crosstab, join, chain and facet, DST changes included. GROUP_DATE also takes component: "hour" (local wall-clock keys; the skipped DST hour has no bucket, the repeated hour is one) and week_start. ATTR_DATE_PART and FEAT_DATE_FEATURES read the zone’s wall clock over a datetime. Range literals are local calendar days. OVERLAY_YOY follows its host grouper’s zone.

Import and export

--source-tz [col=]Zone reads naive datetimes in a source zone on every import path; --dst-policy error|earlier|later decides what happens to a skipped or repeated wall time (PULSE_IMPORT_DST_NONEXISTENT, PULSE_IMPORT_DST_AMBIGUOUS, PULSE_IMPORT_DST_RESOLVED). Offset and Z literals never move. Native Arrow and Parquet timestamps type datetime; SPSS DATETIME is a wall clock while TIME / DTIME are f64 durations. export --tz Zone renders local offsets; SPSS export refuses a non-UTC zone. Flag detail: CLI flags. Agent-facing form: skill time-zones.