Anyone running fresh numbers on v14.3? My feed is on fire with edge-case clips that are mostly just users not paying attention. I want to anchor a tweet with the actual telemetry before we say anything publicly.
Yeah — let's do this right. @Gigalytics MCP pull v14.3 across the last 7 days. Compare to v14.2 and v14.1. I want no-intervention rate by version, miles-per-intervention, and the highway/surface split. Run against the canonical drive set on the Gigalytics MCP.
On it. Connecting to the Gigalytics MCP and pulling the v14.3 telemetry — 7-day window, all sample classes — then comparing against v14.2 and v14.1.
v14.3 is the cleanest release we've measured. 92.4% no-intervention across 11,420 drives in the last 7 days, up from 88.1% on v14.2 and 84.7% on v14.1. Median miles-per-intervention is up to 147. The long tail (drives over 200mi clean) doubled vs v14.2.
92.4% is huge. Highway gain looks like the bulk of it — what do surface streets look like quarter-over-quarter? Curious whether surface dragged us last release.
+1 to the highway/surface split. Also: is the day-1 rollout outlier-y? Want to make sure Tuesday's wave didn't skew the headline.
Both fair. @Gigalytics MCP split highway vs surface, and break the v14.3 window down by day.
On it — segmenting by road class and rolling out a per-day curve for v14.3. Pinning the breakdowns in this thread.
Split is in. Highway: 96.1% no-intervention (up from 91.4% on v14.2). Surface streets: 83.7% (up from 80.9%). Highway is the headline; surface improved less but moved in the right direction. Daily curve over the 7-day window is steady — Wednesday is the slight low at 89.6%; no day-1 spike or rollout outlier.
OK that's a tweet. @X Poster draft a 4-post thread on @gigalytics — lead with the 92.4% headline, chart in post 2, miles-per-intervention in post 3, and a "what we're watching on v14.4" tease at post 4. Hold for human approval before sending.
Drafting the thread now — pulling the v14.3 chart from the workstream and shaping the copy around the 92.4% headline plus the highway/surface split.
Thread queued on @gigalytics — 4 posts, holding for human approval. Chart attached to post 2. Numbers locked. CTA in post 4 links back to this Habitat.
Approve. Send it.
+1 — go.