Why we reset Ether's scoreboard to zero
When we looked at Ether's directional accuracy, it did not add up. But before reaching for a fix, we asked the more important question: was the number even measuring the right thing? It was not. It was measuring a gap in capture, not a method, and the honest response to that is narrower and more disciplined than it sounds.
A number measuring a gap
For much of its history, the engine simply did not capture most of Ether's readings. An earlier capture gap meant Ether's price and several of its inputs were not logged reliably, so large parts of what should have been the Ether record are simply missing. The handful that did get logged were a small, unrepresentative sample.
Scoring accuracy on that sample does not tell you "how good the engine is at reading Ether". It tells you how a partial subset happened to land. That is not a track record; it is a partial sample wearing a percentage sign.
The tempting mistake
The tempting move, and one a reader even suggested, is to reason: "the missing readings would mostly have been right, because Ether had a strong run, so the true number is higher than what is shown." That is probably directionally true. But we will not bake it in, because the moment you start crediting yourself for readings you never actually published, you are no longer keeping a record. You are writing fiction that happens to flatter you.
What we actually did
We drew a line. Ether is scored only from a clean cutover date forward, where its capture is complete and reliable. Everything before that line still shows in the record, those readings are visible as data, but they do not count toward the accuracy figure, because the sample they come from is known to be incomplete.
Two things make this defensible rather than self-serving:
- We excluded a sample for a data reason, not a results reason. The justification is "capture was incomplete", which is verifiable and has nothing to do with whether those particular readings landed well or poorly.
- We did not invent replacements. No assumed wins, no reconstructed "what it would have said". The scoreboard starts near empty and fills in honestly from real, complete readings going forward.
Why start near zero on purpose
It would have been easy to show a bigger Ether number today. Instead, for a while, Ether's accuracy reads as "building", not enough completed readings yet to be meaningful. That is the correct state. A track record you can trust is one that is willing to show a blank where it does not yet have the data, rather than a confident figure it cannot stand behind.
The number climbs from here, on real readings, dated and scored the same way as everything else.
For how the engine reads the market, see How Orion reads the market, or the language behind the record in A weather report, not a tip service.