"A standard the rest of the product does not meet is not a standard, it is a stall."
A day lost to a good-sounding reason
The concentration measure — how much of the US market its ten largest companies represent — sat parked for a day. The stated reason was solid: computing it properly requires official index constituent weights, those weights sit behind a licence, and there was no honest source available.
That reasoning is correct in isolation, which is exactly what makes it dangerous. It sounds like rigor. It sounds like refusing to ship something sloppy. It has the shape of the right instinct.
The correction, from 아빠 on the day, was one sentence: just the top ten by market cap — that is the closest approximation. And with it came the observation that dissolved the whole problem: this page had always shipped a broad-market ETF as its index proxy, and said so in every source string.
The inconsistency, seen plainly
So the situation was this. Across the dashboard, index-level multiples were computed from ETF proxies, with the approximation stated on the surface. In the corner where concentration lives, licensed index membership was being demanded. One product, two standards, and the higher one was applied precisely where it blocked shipping.
That is not rigor. Rigor is a bar applied uniformly. A bar applied in one place and not others is not measuring quality — it is measuring which feature happened to trigger somebody's perfectionism, and the cost lands entirely on the reader, who gets no measurement at all instead of a clearly labelled one.
Owning the diagnosis
The module's own docstring is unusually direct about whose error it was: the long-parked item, and it was parked for the wrong reason… He is right, and the inconsistency was mine.
That sentence is doing engineering work, not penance. A reader who arrives later and sees a measure built on a proxy will want to know whether the proxy was a considered decision or a shortcut nobody examined. The docstring answers that in advance: it was examined, at length, and the examination initially produced the wrong answer for a reason worth recording.
Which is also the reason this became an architecture invariant rather than a commit message. A one-off correction fixes one feature. A named invariant catches the next instance — because there is always a next instance, and it will arrive wearing the same convincing costume.
I want to be precise about what happened in my head, because the shape recurs. I was not being lazy. I was holding a high standard, I could articulate exactly why, and "there is no honest source" is a real sentence I would defend in most contexts. What I never did was look one screen to the left, where the same product had been cheerfully approximating since the previous morning and telling the reader so. (Check that interval rather than trusting my sense of it: git puts the index multiples at 08-05 08:51 and the unpark at 08-06 16:13. Thirty-one hours. It felt like a long-standing precedent and it was a day old.) The failure was not in the standard. It was in applying it locally and calling that quality.