Time-Indexed Data

Promoting the timestamp makes it what rows are named and ordered by, so a range can be asked for by naming two moments and time operations find the ordering unasked. Only one of the record's times can hold that slot.

A stamp with no zone is a clock reading, not a point on the world timeline: it says what a clock showed, not when. Until the recording zone is attached, it cannot be safely compared with, or converted for, another source.

Match each quote to the most recent price record whose stamp is at or before the quote's stamp - an inexact ordered match rather than an equality match. The driving side sets the row count: at most one matched record each.

A moving window of fixed length keeps its edges a fixed distance apart, so old records drop out as new ones enter. A window anchored at the start never moves its left edge, so each answer covers everything so far.

Two independent defaults decide it: which edge closes each bucket, so an edge value joins the bucket before it or the one after, and which edge names the row, so the same fold may be stamped 09:00 or 10:00.

Three behaviours are live across designs: refuse the mixed operation, adopt the process's own configured zone for the zoneless side, or assume UTC for it. Two of the three return a plausible number that is wrong by whole hours.

A centred window is stamped in the middle of the records it covers, so roughly half of them come from after that stamp. The value therefore encodes observations that had not happened at the moment the row describes.

A calendar month is 28 to 31 days, so a per-bucket sum scales with the bucket's width and February sits below its neighbours even at a flat rate. Divide each total by its own bucket's width before reading a decline.

Asking for a range by two moments, and anything that walks the ordering, assume the labels are non-decreasing. Out of order you may get a refusal, a slow full comparison, or a plausible wrong subset. Verify the order yourself.

Asking for a repeated label returns every row carrying it, so the result is a sub-table rather than a single row. The shape now depends on the data rather than on the code, and in several designs nothing raises.

Calendar arithmetic targets the same clock time on the next calendar day; duration arithmetic adds exactly twenty-four hours. Across a clock shift they land on different moments, because the calendar day is twenty-three or twenty-five hours long.

A staleness tolerance caps how old a matched value may be. Beyond that age the match is abandoned and an absent value is produced instead, so a reading from hours ago is never silently presented as the value in effect now.

A minimum observation count — how many real observations a window must cover before it answers. At the full length the leading positions hold no value; below it they answer early, from fewer observations than the length implies.

A count-defined window always covers exactly seven records and therefore a variable span of time. A span-defined window always covers exactly seven days and therefore a variable number of records. On irregular data the two select different sets.

Every grid position nothing was measured at now carries a manufactured value, so the rows count grid positions rather than observations. Averages become time-weighted, variability shrinks, and nothing about it raises an error.

Because a step alone does not fix a grid. The first bucket edge is pinned somewhere, and if it is pinned to the earliest stamp in the input, a different input span produces different edges and therefore different groupings.

The moment your questions are asked against, which for analysis is normally when the event happened, because that is the ordering the world had. Label by arrival instead and the same source re-run produces different numbers. Keep all three as columns.

At a clock shift one local hour occurs twice and one never occurs. A repeated reading names two possible instants; a skipped one names none. Attaching a zone must therefore raise, choose a side, or move the value.

An offset is one answer from a zone's dated rulebook. Storing it pins the moment but throws the rule away, so the captured moment and its clock reading survive while local time at any other moment cannot be worked out.

Nearest takes whichever reading is closer, earlier or later. Every row where the later reading was closer carries a value that did not exist at the label's stamp, so the table contains information from the future. Only the latest-at-or-before direction cannot.

The window walked straight across the boundary between devices, so each device's first answers mix in the tail of the device before it. The window has to be restricted so it never reaches past an entity boundary.

That the match stayed inside one sensor, and that both inputs were in stamp order. Neither is inferred: without an entity restriction one sensor's reading attaches to another's record, and an out-of-order input is walked as if ordered, producing a plausible table.