Driver debrief

LisboaKart, Pontinha · layout Lisboa V3 · kart 19 · 8 September 2026, 17:45–18:23 · 47 laps of the circuit · second session with GPS on the kart

Four days after the last one, and this is the first session that can be scored rather than merely described: there is a previous recording to measure against, and the last debrief left a falsifiable experiment on the table. Both answers are narrower than they first look. You had the circuit to yourself throughout, so the honest comparison is against 4 September's solo laps rather than its whole session — and on that basis the front of your pace improved clearly (43.940s to 43.731s best, fastest five 0.285s better) while the typical lap barely moved (-0.072s). The experiment, meanwhile, did not do what it predicted. Underneath both is one clear finding: your speed is essentially unchanged and your line is not, which is where every section difference on this page comes from.

Best lap

43.731

sheet lap 26; 0.209s faster than 4 Sep solo

Fastest 5 avg

43.787

0.285s better; the five span 0.114s

Median lap

44.264

only 0.072s better than 4 Sep solo

Laps under 45s

42 of 47

4 Sep solo: 20 of 21; under 44s 9 against 2

1 · What this debrief can and cannot say

Measured The printed sheet is transcribed correctly. The 49 lap times reproduce the sheet's own totals exactly: they sum to 2266.058s against a printed total of 2266.058s, average 46.246s against 46.246s, best 43.731s and last lap 44.224s. Those four checks run every time this page is built.

Measured The logger put its start/finish somewhere else this time. On 4 September the device's line sat within about a metre of the circuit's; this session it is at 132 m in that session's distance frame — a third of the way round a 641 m lap. In space the two points are only about 9 m apart, because the circuit runs back alongside itself there, but along the racing line they are a long way apart. Each logger lap is still one complete circuit, so lap times remain valid; they are simply cut at a different place, which is why logger and sheet lap times differ by up to a couple of tenths this session where they agreed to 25 ms last time. Nothing on this page uses the logger's lap numbering.

Method Everything cross-session here is measured through the same physical gates. Section boundaries are derived from a session's own fastest lap, so they are not portable — 4 September divides into 13 sections, this recording into 12, at different distances. Instead both recordings are walked through 13 fixed gates taken from 4 September's section starts: a point on the ground plus the direction of travel through it. That produced 46 complete circuits on 4 September and 47 here, each timed on identical stretches of tarmac. Section numbers on this page are 4 September's throughout.

Measured The recording is clean, and cleaner than last time. 54,847 GPS fixes at nominal 25 Hz over 36.6 minutes and 31.3 km. Satellites used ranged 7–16, most often 13; the receiver's own accuracy estimate had a median of 71 cm (59–110 cm). Unlike 4 September, where the stock parser's guard discarded 31,130 fixes, this recording loses none: every GPS record arrives in increasing time order, so patched and unpatched readers would agree on it. That is a property of this recording, not a fix to the reader.

Not available Still no pedal or engine data. RPM again recorded zero for the whole session, and this logger has no brake or throttle channels. Everything below describes speed and position only. Where braking or throttle would be the natural explanation, it is named as a question. The RPM pickup remains the one cheap fix that would change what a debrief can say.

2 · The circuit, and what the section numbers mean

Every section number on this page refers to this map. Blue badges are corners, grey are straights, and the yellow ticks are the gates both sessions are timed through. Numbering runs the way you drive it, from the start/finish marker.

Aerial imagery: Esri, Maxar, Earthstar Geographics and the GIS User Community. User visual calibration — provisional best visual fit: Esri imagery moved 3.356 m west, 1.854 m north. GPS coordinates and timing are unchanged; this is a subjective display alignment.

3 · The session as it ran

Kart 19 again, so the kart is held constant against 4 September. You had the circuit to yourself for the whole session — your own account, and what the sheet's single driver row and empty gap-to-second already showed. That is the single most important difference from 4 September, which ran in traffic until its lap 28 and alone only for its last 21. Every comparison here is therefore against those solo laps; the whole-session column is shown next to it so the difference between the two questions stays visible. The first two sheet laps (103.2s and 73.9s) are the timed opening from the pit lane rather than laps of the circuit, and are excluded on both sides.

Measure	4 Sep, all laps	4 Sep, solo	8 Sep	Change vs solo
Best lap	43.873s	43.940s	43.731s	-0.209s
Fastest five	43.980s	44.072s	43.787s	-0.285s
Median lap	44.428s	44.336s	44.264s	-0.072s
Mean lap	44.755s	44.399s	44.446s	+0.048s
Spread, all laps	0.816s	0.300s	0.785s	+0.485s
Spread, laps under 45.5s	0.816s	0.300s	0.328s	+0.028s

Measured The gain is at the front of the distribution, not in the middle of it. Against 4 September's solo laps the best improved 0.209s and the fastest five 0.285s, but the median improved only 0.072s and the mean is +0.048s. Against the whole of 4 September the median looks 0.164s better — that larger figure is mostly its traffic, and is the number to distrust.

Measured Consistency is unchanged once the two incident laps are set aside. Raw spread reads 0.785s against 0.300s, which looks like a clear regression. But 4 September's solo phase contains nothing above 45.141s, while this session has two laps above 45.5s. Excluding those, the spread is 0.328s over 44 laps against 0.300s — the same repeatability, over more than twice as many laps.

Confound It was cooler and much windier than 4 September. Across this stint (17:45–18:23) the air was 27.2 °C at 17:00 falling to 25.7 at 18:00, humidity 50–57%, cloud 100%, wind 15 km/h, gusting 44 km/h. On 4 September (18:22–18:59) it was 29.0–30.1 °C with wind 4–5 km/h, gusting 25 km/h. So roughly three to four degrees cooler, in wind gusting nearly twice as hard, on a surface now twelve days without rain rather than eight days. Cooler, denser air and a cooler surface both plausibly favour lap time, so an unknown part of the improvement below is the weather rather than the driving. This is a proxy from a model grid point about 3 km away, and air temperature is not track temperature.

Measured The two offs are in the data, and they are locatable. You report leaving the circuit twice and rejoining. Ranking every lap by how far it strayed from your own fastest lap's line puts exactly two laps clear of the field, and they are the two slow ones: the worst reached 6.4 m off line, at 601 m into the lap — S12 — carrying 38 km/h there and dropping to 25 km/h; the second reached 4.2 m at 196 m — S5, the exit of the corner before it — at 52 km/h. The median lap this session peaked 2.1 m from that line, so the worst is 3.0 times the normal wandering. That cost 1.4s on the S12 lap alone.

Limit The reference is your own racing line, not the edge of the asphalt, so this locates the events and ranks their size; it does not measure how far off the track you actually were. With two offs reported and exactly two laps standing out, the identification is as good as this data gets.

4 · The experiment from last time

The 4 September debrief committed to one change and one measurement: carry more speed out of the kink at 463–497 m onto the long straight, and judge it by the median time through S11, the 75 m straight from 497 to 572 m, on solo laps only — the baseline used here, matching how the target was written. It was set in advance: a solo median of 4.249s against 4.309s — an improvement of 0.060s — over at least eight laps, with the falsification rule that if the S11 median did not move, the hypothesis was wrong.

Verdict: the change happened, the predicted consequence did not. Measured on identical gates, this session's 47 circuits against 4 September's 21 solo ones:

The kink: S10 median 2.021s → 1.902s (-0.119s). Median speed at the gate onto the straight rose +0.6 km/h, to 60.5 km/h. You did drive the corner differently, and it is 0.119s quicker.
The straight: S11 median 4.352s → 4.342s (-0.010s) against a target of −0.060s. Target missed by 0.050s; the straight is essentially where it was.
Together: The two sections combined are 0.129s quicker, and 92% of that is the corner itself rather than the straight after it.

Supported interpretation Part of S10's gain is not speed at all: the route through it is 1.1 m shorter than on 4 September. The hypothesis was that S11 was slow because you arrived on it slowly. You now arrive 0.6 km/h faster and S11 is 0.010s quicker — a change roughly a fifth of the predicted size. Entry speed is therefore real but small at this scale, and the straight is not where this lap is being decided. By its own rule, the experiment is not supported, and the honest reading is that the time it was chasing was never there.

Caveat One change was requested; a whole session's worth of things moved. Track state, tyres, an empty circuit and four days more practice all differ, and no section here isolates driving from any of them. A cleanly-run experiment says the gain did not appear; it cannot say the change was not made.

5 · Where the time actually went, section by section

Each of 4 September's 13 sections, timed on both recordings through the same gates — this session's 47 circuits against 4 September's 21 solo ones. Down and blue is quicker.

Sec	Type	Distance	4 Sep median	8 Sep median	Change	8 Sep best	Spread	Median−best	Entry Δkm/h
S1	straight	0–44 m	2.536s	2.596s	+0.060s	2.508s	0.071s	0.088s	-0.1
S2	corner	44–104 m	5.184s	4.915s	-0.269s	4.598s	0.135s	0.317s	-1.6
S3	straight	104–146 m	2.836s	2.866s	+0.030s	2.803s	0.045s	0.063s	+0.0
S4	corner	146–192 m	2.872s	3.067s	+0.194s	2.882s	0.093s	0.184s	+0.5
S5	straight	192–229 m	2.299s	2.336s	+0.037s	2.232s	0.128s	0.104s	-0.3
S6	corner	229–333 m	8.780s	8.769s	-0.011s	8.537s	0.195s	0.232s	-1.4
S7	straight	333–364 m	2.284s	2.277s	-0.007s	2.222s	0.040s	0.056s	+0.6
S8	corner	364–415 m	3.276s	3.165s	-0.112s	3.005s	0.070s	0.159s	-0.2
S9	straight	415–463 m	2.929s	2.852s	-0.078s	2.802s	0.042s	0.050s	-0.5
S10	corner	463–497 m	2.021s	1.902s	-0.119s	1.799s	0.040s	0.103s	+1.1
S11	straight	497–572 m	4.352s	4.342s	-0.010s	4.208s	0.435s	0.134s	+0.6
S12	corner	572–610 m	2.821s	3.081s	+0.260s	2.849s	0.230s	0.232s	+1.6
S13	straight	610–641 m	2.141s	2.114s	-0.027s	2.058s	0.083s	0.057s	+0.1

Measured The gain is concentrated in S2, and it is bigger than the whole experiment. S2, the corner from 44 to 104 m, is 0.269s quicker — more than S10 and S11 together. Next are S10 (-0.119s) and S8 (-0.112s). Summed across all 13 sections the medians are only 0.049s quicker, which agrees with the 0.072s the median lap improved on the sheet — two independent measurements of the same modest change. Against 4 September's whole session the same sum reads 0.186s, which is the traffic again.

Measured Two sections went backwards. S12 (572–610 m) is +0.260s and S4 (146–192 m) is +0.194s. Together they gave back 0.455s of the 0.631s found elsewhere.

Measured Both offs happened where the lap got slower — and neither regression is caused by the off. The worse excursion is in S12, the other at the exit of S4 into S5. Removing the two worst circuits entirely, S12's median is still 3.071s against 2.821s on 4 September, and S4's slowest lap all session is 3.284s — no spike at all. Those two sections were slower on every lap, not on the two you went off.

Supported interpretation A corner you ran wide at, and which is also slower on all 47 laps, is more consistent with a corner you were not comfortable in all session than with two isolated mistakes. Whether that is the corner, the kart on the day, the wind, or having gone off there once already, this data cannot separate. It does say the time is not lost in the two incidents — it is lost in the other forty-five laps.

6 · What you actually did differently

A section time can change two ways: a shorter route, or more speed along it. Section times alone cannot tell them apart, but path length can, and the answer here is emphatic.

Measured This session is a change of line, not of speed. Median path length through S2 fell from 59.5 m to 55.3 m — 4.1 m less tarmac — at 0.8 km/h less average speed. The two sections that lost time did the opposite: S12 is +3.2 m and S4 +2.9 m, both at within half a km/h of their old average speed. Average speed barely moved anywhere on the lap. What moved is where you put the kart.

Why this measure and not the drawn lines The two sessions' fastest laps sit a median of 1.6 m apart and never more than 3.3 m; part of that is a real line change and part is the receiver's session-to-session bias, which nothing here separates. Path length survives that: a constant positional offset slides a path sideways without changing its length. Sampling noise would inflate every section instead, and it does not — summed over the whole lap the two sessions differ by -0.47 m (633.9 m against 633.5 m). So the lap is the same length; it is distributed differently. Read the drawings below as illustration and the metres as the evidence.

The four sections that moved most, with each session's typical lap drawn on the circuit and its speed alongside. Orange is 4 September, blue is 8 September.

S2 · 44–104 m · corner 5.184s → 4.915s (-0.269s)

Across every lap, the route through here is 4.1 m shorter (59.5 m → 55.3 m), at 0.8 km/h less average speed. The whole gain is the tighter line: you are carrying slightly less speed through the hairpin and still arriving sooner, because you drive four metres less to get there. The two drawn laps are typical rather than fastest — the 5.184s lap on 4 September and the 4.915s lap here, each closest to its own session's median for this section — so read them as examples of the pattern, not as the measurement.

S10 · 463–497 m · corner 2.021s → 1.902s (-0.119s)

Across every lap, the route through here is 1.1 m shorter (33.4 m → 32.3 m), at 1.6 km/h more average speed. The kink the last debrief asked you to change. Both things moved the right way here — a shorter path and more speed — and it is the clearest single corner improvement on the lap. The two drawn laps are typical rather than fastest — the 2.021s lap on 4 September and the 1.902s lap here, each closest to its own session's median for this section — so read them as examples of the pattern, not as the measurement.

S11 · 497–572 m · straight 4.352s → 4.342s (-0.010s)

Across every lap, the route through here is 0.5 m longer (74.7 m → 75.2 m), at 0.6 km/h more average speed. The straight that was supposed to benefit. Marginally more speed, marginally more distance, and the two cancel: this is what a hypothesis failing looks like when nothing dramatic happens. The two drawn laps are typical rather than fastest — the 4.352s lap on 4 September and the 4.342s lap here, each closest to its own session's median for this section — so read them as examples of the pattern, not as the measurement.

S12 · 572–610 m · corner 2.821s → 3.081s (+0.260s)

Across every lap, the route through here is 3.2 m longer (37.1 m → 40.3 m), at the same average speed. The section that cost the most, and where the worse off happened. Running wide is exactly what the extra metres mean, and it is on every lap rather than only the one you went off. The two drawn laps are typical rather than fastest — the 2.821s lap on 4 September and the 3.081s lap here, each closest to its own session's median for this section — so read them as examples of the pattern, not as the measurement.

7 · Where the remaining time is

Within this session alone: the median through each section against the fastest you actually recorded there. The gap compares different laps entering at different speeds, so it is a benchmark, not a lap you have driven.

Measured The median-to-best gaps sum to 1.78s. The top three are S2 (0.317s), S12 (0.232s) and S6 (0.232s), holding 44% of the total. Adding the fastest section times together gives a composite lap of 42.501s, against 43.128s on 4 September and an actual best of 43.686s in this frame. That composite is a diagnostic, not a target: its pieces come from different laps that entered each section at different speeds, and the number moves with how finely the lap is cut. The theoretical-best page cuts this session on its own 12 sections rather than 4 September's 13, and reaches a different figure for that reason; neither is more correct, and neither is a lap you have driven.

Supported interpretation The largest gap, S2, is not the same kind of gap as the rest: it is also the section that improved most (-0.269s), so its spread is what an actively changing corner looks like rather than time sitting untouched. The one that has sat still is S6 — 0.232s of gap here, 0.219s on 4 September, and a median that moved -0.011s between the two. It is the longest section on the lap (104 m, 8.77s), it was the fallback experiment named last time, and two sessions have now left it alone.

8 · Findings, in priority order

The pace step is real and it is distributional. Best 43.731s, median 44.264s, 42 of 47 laps under 45 s. Every summary statistic moved the same way, which is what a genuine change in pace looks like as opposed to one good lap. How much of it is driving is not established: the air was three to four degrees cooler than 4 September, and an empty circuit replaced a session with four other karts on it.
The prescribed experiment failed its own test. S11 moved -0.010s against a 0.060s target, even though entry speed rose and the kink before it got 0.119s quicker. Entry speed onto that straight is not where the time is. Retiring the hypothesis is the finding.
You changed your line, not your speed. Average speed is within about a km/h of 4 September everywhere on the lap; path length is not. S2 is 4.1 m shorter, S12 and S4 are 3.2 m and 2.9 m longer, and the lap as a whole is the same length to within 0.5 m. Every section time on this page is a route change.
S2 is where the session was actually won: 0.269s, unprompted, in a corner nobody was working on — and it is bought entirely with 4.1 m less distance.
S12 gave back 0.260s and is the 2nd-largest remaining gap (0.232s). You were quicker there four days ago, so this is the one piece of time on the list that is known to be available.
S6 is the standing opportunity: 0.232s of median-to-best gap, -0.011s since 4 September, and the longest section on the lap at 104 m. It led the gap table on 4 September too and nothing has touched it since.

9 · One thing to test next session

Take S12, the corner from 572 to 610 m, and get back the 0.260s it lost. Not the biggest gap on the lap, but the only target on it whose time you have already been seen to set.

Why this: It is also where the worse of your two offs happened, and it was slower on every lap rather than only that one — so this is not a target chosen from an incident. Every other candidate is a median-to-best gap, which compares laps that entered the section at different speeds and may not be repeatable. S12 is different: you ran it at 2.821s four days ago and 3.081s this session, same corner, same kart, measured through the same gates. It also holds 0.232s of median-to-best gap within this session, so the two measures agree.
What to change: Nothing else. Same kart (19), and leave S10 and S11 alone now — they have been tested, and they are finished as an experiment.
Measure: Median time through S12 and speed at the 610 m gate. This session: median 3.081s, best 2.849s, spread 0.230s, entry speed 55.4 km/h against 53.8 on 4 September.
Target: Median through S12 back to 2.821s or better over at least ten laps. That is a return to known pace, not a projection.
Falsifiable: If the median does not return, the difference was not driving — something about the kart, the tyres or the track changed between the two sessions, and the section-by-section table above becomes the place to look for what else moved with it.
Still standing: S6 remains the largest untouched block: 0.232s of gap, -0.011s across two sessions, 104 m long. It was the fallback named last time and it is still waiting. One experiment at a time, though.

10 · How this was put together

Jorge Segers, Analysis Techniques for Racecar Data Acquisition, 2nd ed. (SAE International), chapter 14, “Analyzing the Driver” — contents.
Driver61, Prioritising Circuit Corners — the reasoning behind the experiment that has just been retired.
How optimal lap time is calculated — micro-sector summing, and why teams constrain it before believing it.
Your “theoretical best” lap is a lie — but still useful — the case for treating the composite as a diagnostic.