Algorithms opens at once and Learning waits for a fair comparison

What changed for people using Bellwether

• Algorithms opens at once. The first visit after an update used to sit on gray placeholders for 16 to 19 seconds. The list now appears in about a second, with the same figures as before.
• Learning waits for a fair comparison. Repeated mistakes says Worsening or Improving only when the desk has about ten weeks of history and the role ran on enough of those days. Until then a row says Watching and shows the plain count.
• The counts did not move. How many times each slip was seen, and when it was first and last seen, are exactly what they were. Only the comparison against a usual rate changed, and the word drawn from it.

What you may notice

• On a desk younger than 74 days, every row under Repeated mistakes that has been seen says Watching, and its sentence says it is too early to call a trend.
• The usual rate is now measured over the days a role actually ran. A desk that paused for a while is compared fairly when it comes back.
• A role that did not run at all in the last two weeks is no longer told its mistakes are resolved. Resolved needs at least one day on which that role ran and the slip stayed away.
• Rows that were marked as worsening before this release are shown as Watching straight away. The evening after the update they are counted again under the new rule.
• Nothing in this release changes what the desk buys or sells.

Switched on

• Nothing is switched on or off by this release.

Known gaps and what’s next

• The sentence that explains why a row is still being watched was written for a young desk. Wording for a desk that paused comes with a later release.
• The small chart on each row still draws the days before the desk existed as empty days. A later release draws them as a gap.