Mental model

Hold this one picture and the rest of core follows.

A series is a schema-typed, ordered collection of events. Events carry a key plus a typed payload. Operators transform series into series; reducers compress events into values.

The schema S carries through every transform — the type system narrows column shapes correctly, without as casts. That single invariant is what makes the algebra closed: because every operator returns a series of a known schema, you keep chaining, and a bucket is just a smaller series (same API, no second primitive to learn).

The three series types

The key's kind picks the type, and the type decides which operators exist:

Type	Keyed by	Carries
TimeSeries	a temporal key (batch)	the full algebra incl. calendar/clock operators
LiveSeries	a temporal key (streaming)	the same operator surface, incremental cost model
ValueSeries	a monotonic non-time axis	the ordering-based operators only — no calendar (by type)

TimeSeries and LiveSeries are peers, not layers: same schema type, same operators, different cost models (whole-collection vs. incremental). ValueSeries drops in when the natural x of your data is a quantity (distance, work) rather than the wall clock.

The vocabulary each operator assumes — temporal keys, sequences, windowing, triggers, partitioning, late data, the value axis — is defined page by page in Concepts. Read those reference-style once the picture above is clear.

Coming from pandas?

pandas's .resample() covers both directions; pond-ts splits them:

• Downsample (fewer rows out than in) — aggregate(seq, mapping) ≈ .resample().agg()
• Upsample / regrid (one row per grid point, hold or interpolate) — align(seq, { method }) ≈ .asfreq() + .ffill() / .interpolate('time')
• rolling(window, mapping) ≈ .rolling().agg()
• within(range) / overlapping(range) / trim(range) ≈ .loc[] slicing with three explicit semantics
• reduce(mapping) ≈ .agg() on the whole frame
• groupBy(col) ≈ .groupby(col)

See Aggregation for the deep dive on bucketing.

Coming from pondjs?

pondjs	pond-ts
TimeSeries (sorted)	TimeSeries (1:1 conceptually; methods differ)
Collection (unordered)	TimeSeries — pond-ts only exposes the sorted variant
IndexedEvent	Event<Interval>
TimeRangeEvent	Event<TimeRange>
Pipeline + processors	method chain on TimeSeries
Pond / Aggregator	batch aggregate() or LiveAggregation

pond-ts deliberately drops the Collection / TimeSeries distinction: a bucket is a smaller TimeSeries, so anything that works on a series works on a bucket — same API, no second primitive.