You moved your T4 turn-in five metres in a year — and the exit followed.
The same corner on four track days, read in the road frame: where the bike sat between the measured limits, not just how fast it went. 92 clean-pace laps (traffic laps more than 3 s off the day's median are excluded).
The line through T4, by day
What changed, in order
Entry. In July 2025 you arrived at the turn-in point at 79% of the width — hard against the outside edge, the textbook set-up. By August 2026 you turn in from 39%: mid-track, five metres further inside. The two lines are already 4 m apart 30 m before the corner, so this is a decision made on the approach, not a mid-corner correction.
Apex. The apex moved 6 m later (+10 → +16 m past the curvature peak) and 0.5 m closer to the kerb (1.1 → 0.6 m) — and became far more repeatable: the spread of the apex point across laps fell from ±15 m to ±2 m. The line the bike draws got tighter (own radius 38 → 35 m) while the apex speed didn't drop at all.
Exit. Exit speed onto the run to T5 went up 2.4 km/h (89 → 92 km/h), with 2026-08-01 the best day at 92.5. A later, tighter apex from a mid-track entry is the classic point-and-shoot line for a slow corner onto a straight — and it's what the data says you converged on, without anyone telling the analysis what a good line is.
The instructor day sits in between. 2026-05-26 (the 1:1 tow) kept the 2025 entry (71%) but already took the tight apex (0.5 m, 4 laps on the kerb). The entry change came after — the two August days are the first with the mid-track approach.
Day by day
| day | laps | turn-in | apex | exit | apex vs geometric | margin | line radius | v apex | v exit | best lap |
|---|---|---|---|---|---|---|---|---|---|---|
| 2025-07-28 | 26 | 79% | 9% | 35% | +10 m ± 15 | 1.1 m ± 0.6 | 38 m | 52 | 89 | 140.7 |
| 2026-05-26 | 34 | 71% | 5% | 35% | +10 m ± 5 | 0.5 m ± 0.4 | 37 m | 51 | 90 | 139.2 |
| 2026-08-01 | 16 | 45% | 5% | 28% | +17 m ± 5 | 0.6 m ± 0.2 | 35 m | 50 | 92 | 138.1 |
| 2026-08-02 | 16 | 39% | 4% | 33% | +16 m ± 2 | 0.6 m ± 0.5 | 35 m | 52 | 92 | 139.2 |
The same corner in the app
This is T4's report card on 2026-08-02 with 2025-07-28 as the comparison day. The dashed line is last year's best lap entering from the outside; the bold line is this year's, already inside at turn-in, crossing to the kerb later. The faint lines are the rest of the session — they sit on top of each other, which is the ±2 m repeatability in the table.
Below the strip, the two scatters are the within-day regression: on this day, laps that apexed even later exited slower (−1.5 mph per metre, r −0.60 over 17 laps). Read together with the year-over-year move: the later apex paid off up to about +16 m and costs beyond it. A regression can only see the spread you gave it on one day; the day comparison sees the move between days.
Two tools, two questions
Within a day: does my line variation predict my exit?
The regression over the day's clean laps. Same bike, same tyres, same air. It answers "if I apex 3 m later tomorrow, what happens?" — and for most Palmer corners the honest answer is |r| < 0.3: lap-to-lap line scatter is not what decides the exit there. T4 on 2026-08-02 is one of the few with a real slope.
Across days: did my line change, and did it stick?
Medians and σ per day, aligned for each day's GPS bias. It catches the things you can't see in a single session — the T4 entry, or T11 where the apex moved from 18 m early in 2025 to 18 m late on every day since. A regression is blind to this by construction: the whole day sits at one value.
What to take with a grain of salt
- One lap's metre is noise. GPS after alignment is good to ~1 m; the antenna correction assumes 0.8 m over the contact patch. Everything above is a median or a spread over laps, and the app's card says the same.
- The strongest correlation in the whole dataset was a traffic detector. T12 showed r −0.80 between apex margin and exit speed over 121 laps — because a third of the laps sat mid-track at 53 km/h, 11 s off the pace, behind someone. With those laps out it drops to −0.42. The card's regression should filter by pace before it draws a line; that is a fix I'd make next.
- "Geometric apex" is the tarmac's curvature peak, not a recommendation. A late apex at T4 is the right call; a late apex at T13 (+40 m on every day, half the laps over the inside line) is worth a look at the 1:1 detail view to see what you're actually aiming at.