Type Alias Repeating<T, Max>

Declares a REPEATING PARAMETER GROUP: one declared parameter whose arguments arrive as a fixed-size cycle repeated N times. SUMIFS(sum, r1, c1, r2, c2, …) is one sumRange followed by a [criteriaRange, criteria] group repeated as many times as the caller supplied.

At the type level it is T[], so the body reads exactly what it is handed:

Period 1 is spelled without the tuple — a one-element tuple is a compiler error, because [IRange][] and IRange[] would otherwise be two ways to say the same thing. A period-1 group takes its slot name from the parameter itself, so array: Repeating<IRange> shows as array1, array2, … — the same display a rest parameter gets, and the reason Repeating can replace the first + ...rest split that reads as array, rest1, rest2.

Why this exists rather than a rest parameter

...rest: T[] has ONE type for every trailing argument, so a period-2 group has to declare its slots as their union — which makes the criteria slots of pairs 2+ invisible to every consumer that reads the declaration per position. The implicit-intersection site predicate is the one that matters: it saw SUMIFS's criteria as a range parameter and under-marked the formula. A group binds argument i to slot (i − prefix) % period, so every repetition types the same as the first.

A group also carries a real MINIMUM: at least one repetition is required. That is the whole difference from a rest parameter, which is 0-or-more and remains the spelling for it:

...x: T[]                    // 0 or more
x: T[]                       // 1 or more  (a period-1 group; an array IS a repeat of one)
x: Repeating<[a: A, b: B]>   // 1 or more, period 2+
Copy

So x?: T[] is refused — it would be a second way to say ...x: T[].

Arity is arithmetic, never type-directed

n = (argc − prefix − suffix) / period, and a non-zero remainder is an error.

The one short form

The LAST slot of a group may be declared optional, which means exactly one thing: it may be omitted when there is a single repetition. SORTBY's legal arities are 2 and 1+2n and nothing else — SORTBY(array, by1) is legal, SORTBY(array, by1, o1, by2) is not. It is not a general "the final element of the final group may be omitted" rule; that rule would accept a formula Excel refuses. An optional slot anywhere else in the group is a compiler error.

The maximum, Max

The second type argument is the largest number of arguments the WHOLE CALL accepts, which is how Excel states it: one number per NAME, compared against the call's total argument count and refused at entry (Excess_function_arguments), never a count of this parameter's own share. SWITCH's 254 therefore covers its expression and its trailing default as well as the pairs, and CHOOSEROWS' covers its array.

It defaults to 255, which is the ceiling for most names and the highest any name carries. Declare it only where a name draws the line lower — XOR, IFS, SWITCH, LAMBDA, LET, AGGREGATE, SORTBY, MAXIFS, MINIFS, CONCAT, TEXTJOIN, HSTACK, VSTACK, CHOOSEROWS, CHOOSECOLS and every DOTTED descriptive spelling (STDEV.S, VAR.P, MODE.MULT, …) all stop at 254 where their legacy twins take 255. It cannot be derived from the shape, so it belongs to the declaration.

A REST parameter is the one place Repeating may be written alongside ..., and only with an explicit Max: ...rest: Repeating<T, 254> keeps the rest parameter's own 0-or-more meaning and adds nothing but the ceiling. A bare Repeating<T> on a rest parameter stays an error — that is the spelling collision the marker exists to avoid.