"The composite states its own as_of, names its oldest input, and reports the narrowest window under it."
What aggregation destroys
The composite takes eight indicators, scores each, averages within three categories, weights the categories, and returns one number. Aggregation is compression, and compression discards. The question is always what it discards, and here it is two things that matter enormously.
It discards the spread of dates. One leg comes from a quarterly statistical release and can legitimately be months old. Another is a daily series updated last night. Averaged together, the output has no date of its own — and whichever date you print beside it will be wrong for most of its inputs.
It discards the spread of windows. A composite whose members were measured against three years, forty-one years, and a century and a half is only as measured as its shortest ruler. But the composite is a single number between zero and one hundred, and single numbers look equally solid regardless of what went into them.
Three fields that put it back
Rather than hiding the spread, the return value names it. The as-of is the newest contributing reading, so the reader knows the freshest thing in the mix. The oldest input is named with its date, so they can see the laggard by name — the quarterly leg, months behind, called out rather than averaged away. The narrowest window is the shortest ruler any member was measured against, because the composite's claim to be historically grounded is capped by its least-grounded member.
The docstring puts it in one clause worth memorizing: a composite is only as measured as that.
The decision to keep two markets out
There is a second honesty decision in the same module, and it is the more interesting one. The Korean and Japanese readings are scored on the same published scales as everything else, and they are deliberately outside the composite, presented as context cards instead.
The reason given is precise: the composite characterizes the US market against itself, and folding Korea into that average would quietly change the question it answers. The number would still compute. It would still look like a composite. But it would silently become a different measurement — some blend of three markets with no name, no reference history of its own, and no way for a reader to notice the substitution.
That is the subtlest failure mode in this whole track. A number that changes meaning while keeping its name is worse than a number that is wrong, because a wrong number can be caught by checking it, and a redefined one checks out perfectly.
More inputs feels like more rigor, and "the composite covers all three markets" is a better sentence than "the composite covers one and the others sit beside it." It took reading the reasoning to see that the better sentence describes a worse instrument. The composite answers where does this market sit against its own history — and there is no version of that question with three markets in it. Keeping them out was not a limitation to apologize for; it was the only way the question stayed askable.