GP ANALYSIS

These are the published times, taken from the timing feed's own classification rather than re-derived from the lap data. A lap deleted after the session shows up here exactly as it does on the timing screen.

Position and gap are absolute and never move with the driver filter. P1 is P1 whoever is on screen, and the gap is always to the pole time — filtering to six drivers must not promote the quickest survivor to pole.

The cut lines are read from the data, not assumed. The last position with a Q2 time and the last with a Q3 time mark the two eliminations, so a session where a car failed to set a time, or a short grid, draws the lines where they actually fell rather than at P15 and P10 by convention.

Best is the quickest of the three segments, which is not always the last one a driver ran — a driver can reach Q3 and go slower there than in Q2.

Each cell is a lap time minus the one before it, so a negative number is time found. Q1 → Q3 is the whole session's improvement, and it is only shown for drivers who reached Q3.

Most of this is the circuit, not the driver. A qualifying hour rubbers the track in and the surface cools, so the whole field improves through the session — typically most of a second between Q1 and Q3. What is worth reading is who improved more or less than the field around them, which is why the field's median improvement is drawn on the chart as a reference.

Runs are not like for like either. Q1 is often a single banker on a used set, Q3 two runs on new tyres, and a driver already through can afford to sit out a segment. A small improvement is as often a driver who did not need one as a driver who could not find one.

The colour scale is blue to orange through grey, not red to green: the notebook's red–green scale is the one pairing a colourblind reader cannot separate, and the sign of the number is the whole message here. Every cell carries its number anyway, so the colour is a second reading of it and never the only one.

Both numbers are the best qualifying lap of the session — the quickest of a team's two cars across all three segments, from the published classification of each year. A negative delta is quicker now.

Teams are matched by entry, not by car. Audi is the Sauber entry, Alpine the Renault one, Racing Bulls the Toro Rosso one and Aston Martin the Racing Point one, so those rows compare a constructor to what it used to be called. That is a fact about the paddock and not a claim that the cars are related. Cadillac has no predecessor and is listed as new rather than given an invented lineage.

A delta across a regulation change is not a measure of a team. 2026 changed the power unit, the aerodynamics and the minimum weight at once, so the whole grid moves together — the field median is drawn as the reference for exactly that reason. Read a team against the median, not against zero.

If the layout changed, the comparison is void and the page says so instead of quietly reporting a delta. Corner counts and measured lap lengths are compared year to year to detect it.

Sprint qualifying is not compared: it is a different session from grand prix qualifying, on a different tyre and fuel rule, and there is no sprint qualifying at all before 2021.

Pole lap and fastest lap are not the same lap, and the switch above picks which one is charted. The pole lap is the lap that took pole: the driver classified P1, on the time they set in their final segment. The fastest lap is simply the quickest lap anyone set in that qualifying, in any segment.

They agree in most years and come apart when qualifying was disrupted — a drying track, a red flag, a wind change. Of the 63 circuit-years measured here they differ in 7, and often by a different driver: at Canada 2023 pole was a 1:25.858 in a wet Q3 while the session's quickest lap was a 1:18.725 set earlier, seven seconds faster. Reading one as the other is the single easiest mistake to make with this data.

Both are taken from the measured samples and packed the same way. Where the two are the same lap, the telemetry is stored once and read twice rather than duplicated.

The traces are aligned on relative distance, each lap as a fraction of its own length — the same alignment the gap trace uses between two drivers, and the only one that means anything across years. Absolute metres would compare different corners: laps differ by tens of metres between drivers of the same session, and far more between resurfacings.

A layout change voids the comparison and is flagged, not smoothed over. Corner counts and lap lengths are compared against this year's, so Barcelona before the final chicane came out in 2023, or Yas Marina before the 2021 rebuild, is marked as a different circuit. The corner markers on the axis are this year's and only line up with the years that share the layout.

2026 is a regulation reset. A new power unit, active aerodynamics and a lighter car arrived together, so a gap to 2019 is a measure of the rules far more than of the drivers. That is the interesting part, but it is not a like-for-like lap.

Tyres, fuel and track temperature are not controlled for and cannot be. A pole lap is the quickest lap anyone managed on the day, and a wet or disrupted qualifying shows up here as a slow year rather than as a missing one.

A lapped car's gap is a real time, not a lap count. The timing feed gives the winner a total race time and everyone else a gap to it — including cars a lap down, where the gap is the wall-clock difference as they crossed the line at the flag. At Hungary LAW finished P8 on 69 laps, 28.033 s after NOR: both numbers are true, and printing the gap alone would report a lapped car as half a minute off the win. The lap deficit is measured separately, from the lap counts, and shown as its own badge.

So a bar can be shorter than the one above it. RUS finished on the lead lap 57.5 s behind; LAW finished a lap down 28.0 s behind and is classified below him. A lap down outranks any time — the same rule that puts a slower Q3 lap ahead of a quicker Q2 one in qualifying. The cut lines mark where each lap group starts.

Position and gap never move with the driver filter. P1 is P1 whoever is on screen. Filtering to six drivers must not promote the quickest survivor to the win.

Places gained is only computed for a classified finisher. A car that retired still carries a finishing position — it is placed by how far it got — but "gained three places" is not a thing that happened to it, so the figure is left blank rather than invented.

Grid slot comes from the feed, and the feed leaves it empty for a pit lane start. Where it is missing, the position at the end of lap 1 is used instead and the row is marked, because that is a lap of racing later and not the same thing.

A retirement shows how far it got, in laps completed — including 0 laps for a car that did not finish the first one. RUS started third at Belgium and retired during lap 1; a count of zero is a real answer and has to read as one rather than as a blank. Such a driver has no timed lap and therefore no chip in the filter above, so nothing could ever have deselected them and they are always listed.

Lap 0 is the grid slot, so the first segment of each line is the start itself — the one place on the chart where a line moves without a lap being run.

A line ends where the driver's race ended. A retirement is not carried along the bottom of the chart to the flag; it stops, and is marked. Drawing it to the end would show a car circulating that was in the garage.

Neutralisations are shaded behind the lines, on the leader's lap counter — the counter the race is run on. Positions barely move under a safety car and then move a great deal at the restart, which is worth being able to see rather than having to infer.

Position comes from the timing feed's own order at the line, so it already accounts for cars in the pit lane and for lapped traffic. It is not re-derived from lap times.

A car that completed no lap is still on the chart — a cross at its grid slot, which is the whole of what happened to it. Such a driver has no fastest lap and therefore no chip in the filter above, so a chart that only drew selected drivers dropped them silently: RUS started third at Belgium and retired during lap 1. Drivers with no chip cannot be filtered out, so they are always shown, exactly as they are in the result.

Both ends of every line are labelled — the driver at their grid slot on the left, and their result on the right. Twenty lines crossing each other forty times is not a chart you can follow from one end alone.

Start vs finish counts places, not time. Blue is places gained, orange places lost — the same ramp the qualifying improvement chart uses, where blue is always the better outcome. Only classified finishers appear: a retirement has a finishing position but no meaningful net movement.

A race lap is not a measurement until it is filtered. Safety car laps, in-laps, out-laps and laps stuck behind a slower car are all real laps and none of them measure pace. The controls above are the notebook's four filters, made live: green flag only, no pit laps, drop the slowest tail, and keep only laps within a percentage of the driver's own quickest. The window is relative to each driver's own best, so a slow car is not measured against a quick one.

The defaults reproduce the notebook exactly — same laps, same cut, same means, same order. That is deliberate: a number here that disagrees with the notebook is a bug until proven otherwise. The two switches beside the view buttons are where it can be taken further, and each says what it changes.

Mean is the notebook's statistic; median is the more robust one. One lap caught behind a backmarker drags a mean and barely moves a median. The dark rule, the ranking and the gap column all follow whichever is selected, so they can never come apart. The box is the middle half and every counted lap is drawn behind it, coloured by compound.

Dropping the retired drivers changes more than their own rows. The tail quantile is taken over whatever is left, so including them moves the cut and with it which laps survive for everyone. At Hungary that alone is the difference between LIN reading 86.706 and 86.715. The notebook drops them from its pace chart and keeps them in its by-finish one; the checkbox is that choice, made explicit.

The letters under each driver are the compounds they ran, in order — MHHS is a medium then two hards then a soft. They come from every lap the car did, not only the counted ones: the tyres it ran are a fact about the race, not about which laps survived a filter. A car on a one-stop is not comparable to a car on a three-stop, and this is how you tell which is which.

Trend is a rolling mean over the counted laps, which smooths the lap-to-lap noise and leaves the shape: where a car came alive, where it fell away, where a stop landed.

Sectors and Stints use the same filtered laps as the spread, so the three views cannot disagree about which laps were counted.

Tyre fit is the carousels' tyre-fit slide, live: one panel per compound, every counted lap a dot at the age of its tyre, and a least-squares line through each stint, with the fit written out underneath. Never one line per driver: two stints on one compound are forty laps of fuel apart, and a pooled line would fit the fuel. Teams draws one line per team and compound, its slope fitted inside each stint (each keeps its own intercept) so the step between stints is not read as wear. The trend is lap time — wear minus fuel burn — and the last column adds this circuit's fuel effect back.

True pace is a model, and it answers a different question. Raw mean lap time rewards a driver who stopped more often, because extra stops buy extra laps on fresh rubber. This fits LapTime ~ Driver + TyreLife × Compound + LapNumber and then scores every driver at the same tyre age, compound and fuel load. Read it as "who was quick for the tyre they were on", not as the on-track gap — the spread already shows that.

What the model cannot do: one degradation rate per compound is shared by the whole field, so a driver who manages tyres better than average has that folded into the residual rather than into their own number. There is no driver-by-compound interaction, which is exactly the assumption that makes scoring everyone at one reference compound legal. A driver with very few counted laps gets a confident-looking number from almost no data, so the lap count is printed beside every row. And if the field pitted in the same handful of laps, tyre age and fuel load stop being separable — the correlation is reported, and above 0.8 the two coefficients must not be read apart.

The panels are the pace card's laps; the table is every lap. The two views answer different questions and so read different sets. A panel laid beside a row of the spread above is the same laps, so the shape of a driver's race and their rank in it cannot disagree — the pace controls drive both. The table is the record: every lap the timing feed gave, in-laps, safety car laps and all, because a record with holes in it is not a record.

On the panels the gaps carry information. A missing dot is a lap the filters removed, and the two marks say which kind: a shaded band is a neutralisation, a dashed rule is a pit entry and takes its out-lap with it. Reading where a driver's dots stop is reading where their race stopped being racing.

The vertical rules are pit entries, drawn from the lap the car came in on. The lap after each rule is an out-lap and will sit high.

Fuel-adjusted removes the weight the car was still carrying: adjusted = raw − (final lap − this lap) × fuel effect, using this circuit's own figure from /retrieve_fuel. A race starts heavy and ends light, so raw lap times fall through the race for a reason that has nothing to do with the driver. Adjusting removes that slope and leaves tyre wear and traffic behind it.

Every panel shares one vertical scale, so a car that was a second off is a second off on the page. Panels are not individually zoomed to fill themselves.

Click a panel to enlarge it. It takes the full width of the card and the rest reflow around it, so the grid stays on screen and the next driver is one click away. Enlarged, it gains a lap axis and lap times on the vertical one — at thumbnail size there is no room for either, and a thumbnail is read as a shape rather than looked up.

Compare puts every selected driver on one set of axes. Who is on it is the driver filter at the top of the page, which already decides who every other card is about — a second, card-local idea of "on screen" would be one too many. There the dot is the driver and the ring is the compound: with several cars sharing a scale the first thing a mark must answer is whose it is. Teammates share a colour, so the second car of each team is drawn as a diamond — the same second channel the line charts get from a dashed stroke, and one that survives colour blindness.

Compare reads the recorded times and the fuel-adjusted view does not apply to it. The correction is one number for the circuit, taken off every car alike, so it tilts the whole picture and cannot move any driver relative to another.

A stint is a run on one set, bounded by the pit lane. The number in the bar is its length in laps. A hatched bar is a used set — a tyre that had already done a lap somewhere, usually in qualifying.

Neutralisations are shaded across the whole grid, on the leader's lap counter, because a safety car is the single largest thing that happens to a strategy: a stop under one costs roughly half what it costs under green.

Degradation is fitted per compound, over tyre age, on the race pace card's laps — driven by the same controls, so the two cards cannot disagree about which laps were counted. A fit over in-laps and safety car laps would be measuring the pit lane and the delta target, not the tyre. The slope is fuel-corrected: a car sheds 0.03–0.07 s a lap as fuel burns off, the same order as the wear it competes with, so an uncorrected slope reads far too flat and often comes out negative. This circuit's own fuel effect is added back.

Fuel correction moves the points, not just the number. Each counted lap is adjusted by the fuel still on board — adjusted = raw − (final lap − this lap) × fuel effect, the same correction the lap chart applies — and the curve is refitted to what is left. It used to be added to the finished slope instead, which gives roughly the same number and leaves the chart untouched, so the control looked broken.

The lift is close to the fuel effect but not equal to it, and the difference is real. Adding the effect to a slope assumes every tyre age draws on the same spread of lap numbers; it does not, because short stints drop out at high tyre age. So the correction bends the curve as well as tilting it, and at Hungary the three compounds lift by 0.043–0.057 s against a 0.051 s fuel effect.

A degradation rate is not a tyre's property. It is what this field did on this surface at this temperature, on the stints they chose to run. A compound only ever run in the last ten laps on a light car will look better than it is.

Compound usage counts laps, not stints, so a one-lap run to the flag does not weigh the same as a thirty-lap stint.

Built from session time at the line, not from summed lap times. 2.0% of this season's race laps carry no lap time — restarts, red flags, timing gaps — and a cumulative sum silently loses every one of them and drifts from there on. At Japan that is 460 s adrift by the flag; at Monaco, with its red flag, 2 430 s. The timing feed's own clock at the line has no such hole.

To the winner compares every car against the winner on the same lap number, so the line is the gap as it stood at that point in the race. To the leader compares against whoever led that lap, which is the gap the pit wall was looking at.

A car that pits drops a whole pit-lane time in one lap, so the sawtooth is the strategy, not noise. Under a safety car the whole field converges toward the leader for reasons that have nothing to do with pace.

A lapped car's line keeps climbing, and that is correct. The comparison is at the same lap number, so a car a lap down has taken roughly a whole extra lap of time to reach it — the gap grows by about a lap time each time it is lapped again. The moment a line steps up by a lap time is the moment the car was lapped, which is worth seeing.

It does mean the back of the field sets the scale. With the whole grid on screen the leaders are compressed into the bottom of the chart; the driver filter is the answer — pick the cars you are comparing and the axis follows them.

Every driver's lap 1 is stored at its measured resolution — the same packing as every other lap on this page — so the thresholds above are live. The notebook fixes them at 50, 100 and 200 km/h; here they are yours, and the whole chart re-measures.

The clock starts when the car first exceeds 2 km/h, not at lights out. A car's own release is the only start point that is measured rather than assumed, and it makes the comparison between cars fair without needing the light signal at all.

This is not reaction time. The gap between lights out and the car moving is not in this data. What is here is how long the car took to go from a standstill to speed, which is the launch and the first phase of the drag race to turn 1.

The places column is the change from the grid to the end of lap 1, which is the start plus the whole first lap — overtakes, a good line through turn 1, and anyone else's mistake.

Traces puts the lap-1 telemetry itself on the same panels the rest of the page uses, so the launch can be read channel by channel: where the wheelspin was, who lifted, who got the better exit onto the straight.

One measurement per lap, from the car's own speed channel. Each driver's car data is pulled once and sliced at the lap boundaries, so this is the true peak of each lap — not SpeedST, which is a reading at one fixed point that cars are often still accelerating past.

All three views read the race pace card's laps, and the controls up there drive this card too. That takes out the neutralised laps — a lap behind a safety car has a top speed that measures the delta target and not the car — and also the in-laps, the out-laps and the laps outside the pace window, which green-flag alone left in. Untick Green flag laps only up there and the neutralised laps come back here as well; that is the point of the controls being shared.

And one gate of this card's own. A lap's peak is the maximum of its car data, so when that channel drops out mid-lap the peak is of whatever arrived — COL's lap 17 at Hungary reads 153 km/h against a 313 km/h median, on a green lap in 86.011 s, and no filter that looks at lap time can catch it. A lap that earned a stored telemetry trace passed a gate on the same car data and is taken as measured; one that did not is admitted only if it clears nine tenths of its driver's median, which is the floor of the traced population across the season.

A race top speed is mostly a story about tow and fuel, not about power. A car that spent the race in traffic will read high; a car that led every lap in clean air will read low. The spread down a driver's column is the interesting part — a wide spread is a car that was sometimes towed and sometimes alone.

Every lap is the same laps again, laid out one column per lap of the race, so a high top speed can be read as one tow or as a characteristic — and placed in the race. A · is a lap with no counted peak. Neutralised lap numbers are coloured in the header, which is why those columns are empty.

Two cars have swapped when the sign of their position difference flips between one lap and the next. That is the whole detection — it needs no telemetry and works at every race in the season. The position channel is 99.7% complete: 1 448 of 1 452 laps even at Monaco, where the telemetry position channel never arrived at all.

This is a floor, not a total. Positions are known at the start/finish line, so a pass and a re-pass inside the same lap leave both cars crossing in the order they started it and cancel out. Nothing in lap-by-lap data can see them; the alternative is per-corner position, which is not in the cache and would be 30 MB a race.

Three kinds, because one total would lie. Pit cycle is a place taken while one of the two was in or out of the pit lane — an undercut, an overcut or a slow stop. Attrition is the car ahead retiring. On track is everything left, which is the racing. At Barcelona 172 of 196 swaps happen in the pit cycle, so a single figure would report a race of processional stops as a thriller.

Pit beats attrition beats on track: a car that pitted and then retired lost the place in the pit lane first.

Only the lap the swap happened on counts as a pit lap. The order is compared between the end of one lap and the end of the next, so a stop explains that change only if a car was in the pit lane during the second of them. Testing the lap before as well reaches back too far and swallows the first real racing lap after every stop — at Hungary Verstappen pitted on lap 14, rejoined eighth on his out-lap behind Hamilton, and passed him on track on lap 16 with neither car near the pit lane. That counted as a pit-cycle place until this was corrected.

Lapped traffic is thrown out. A car unlapping itself changes the classification without anybody overtaking anybody. Two cars more than half a lap apart at the line are not racing each other, and those swaps are excluded and counted separately under the table.

The map places a pass where the two cars actually changed order, which the detection above cannot say on its own — it knows the order at the line and nothing in between. Each lap carries its elapsed time at forty points around the circuit, so both cars’ clocks can be read at the same place on the road; the pass is where the difference between them crosses zero, interpolated between the two points either side. That puts it to a few tens of metres.

Only on-track passes are mapped. A place taken in the pit cycle did not happen at a point on the racing surface, and drawing it at one would be a lie. An attrition pass has no crossing either — the car ahead stopped. Both are counted in the table and left off the map, and the note says how many.

A hollow dot is an estimate. A lap only earns a time grid if its speed trace is the shape of a normal lap, and at some races almost none do — the Hungarian Grand Prix has one on 98 laps of 1431. Where a grid is missing, both cars are assumed to cover the lap at a constant rate; the gap is then linear in the lap fraction and crosses zero once, at a point built from two measured lap times and two measured line crossings. That places the pass in the right part of the circuit but cannot tell a move under braking from one down the straight, so it is drawn hollow and counted separately.

A pass with no crossing at all is not drawn. Where the gap never reaches zero inside the lap — which happens when the order changed during a lap one of them did not complete — the event keeps its row in the table and gains no dot.

This is an index, not radio. The feed carries {time, car, path} per capture and no transcript anywhere, so the page can say who spoke and when, and link to the audio. It cannot quote anybody.

The one thing on this page that needed the network. FastF1 knows the endpoint name, ships no parser and does not cache the stream, so the clip list is fetched from F1's livetiming host at build time and stored as a CSV. The audio itself is not downloaded: a race carries 20–40 clips at 50–150 KB, and embedding a season would add tens of megabytes to a page that is otherwise entirely self-contained and opens off the filesystem. The links need an internet connection, and they stop working whenever F1 stops hosting the file.

The driver and the clock come out of the file name — BEA_87_20260726_145310.mp3 — which is steadier than mapping a car number through a season of driver changes. There is no lap number: FastF1 leaves LapStartDate empty unless the car channel is loaded, which is minutes a session and gigabytes a season for one column. Time of day is what F1's own player shows beside a clip, so it is what this shows.

The FIA's own stream, verbatim. FastF1 caches it beside the laps, so it costs no network and exists for every session — practice and qualifying carry more of it than a race does, and most of the rest of this page has nothing to say there.

The kinds are this page's, not the FIA's. Race control publishes only three categories — Flag, SafetyCar and Other — and Other is half the stream: it holds the penalties, the deleted laps, the DRS calls and the weather alike. Each message is re-read against an ordered list of patterns and the first that matches wins, so a penalty for a track-limits breach files under penalty rather than under track limits.

Drivers are read out of the message text, not out of the feed's own field: FastF1 fills RacingNumber on 12 of Hungary's 80 messages, while the text carries CAR 43 (COL) on every one that is about a car. Selected drivers keeps the messages naming somebody on the filter; a message naming nobody — a flag, a weather note, the pit exit — is not about a driver and is kept either way.

The lap is cut into 25 equal minisectors by distance. Every selected driver's lap is resampled onto a common 1 m grid, and each minisector is won by whoever carried the highest average speed through it — the notebook's rule, unchanged.

At a race that lap is the driver's race average, not their fastest lap: their green, non-pit laps inside 107% of their own best, meaned channel by channel into one lap. A race's fastest lap is one lap on fresh tyres in clear air, and each car sets it forty laps from the next — a map built on those is comparing different races. In qualifying and practice there is no average to build, so the card reads the fastest lap and says so under the table. The traces card's source switch drives this card too; put it on a single lap and the map follows.

The Laps column is how many laps went into that average, against how many of that driver's race laps were eligible for it — green, no lap into or out of the pit lane, inside 107% of their own best. They match for most drivers; where they do not, the missing laps are ones whose distance axis failed the telemetry gate. Under half is marked, because a mean of five laps of seventy is a sample of the race, not the race.

The table is sorted, and its gap measured, on the same lap the map is drawn from — so the driver at the top is the quickest on average, who is often not the driver who set the fastest lap.

The label on the track is the answer, not the colour. Each unbroken run of minisectors is direct-labelled with the driver code that owns it. The 2026 team palette cannot carry identity on its own: Audi and Ferrari differ by an OKLab ΔE of 1.7 against a floor of 15, and there are four blues and two greys behind them. Colour tells you the team; the label tells you the driver.

Minisectors won is not a ranking of lap time — a driver can own the most minisectors and not be quickest, because a minisector is one twenty-fifth of the lap whatever it costs in seconds.

The Delta switch picks how the gap is worked out. Ours puts every lap on relative distance and reads the gap off each lap's own measured clock. F1 Insights Hub reproduces that site's comparison: the raw car feed, with the speed cleaned of dips and steep braking, each lap slid up to 45 m to line up with the reference, the time worked out from speed and pinned to the official sector times. It only runs on real single laps, so the race average and single race laps stay on ours, and every other card on the page always reads ours.

On a race there are two laps to choose from. Fastest is the real lap the driver set. Race average is one lap built by meaning every channel across their green, non-pit laps inside 107% of their own best — the pace card's default set — at each point of the circuit. It is packed like any other lap, so every panel, the gap, the speed delta and the modelled state of charge all work on it with no separate code.

The average is on relative distance, so laps are meaned at the same point of the road rather than at the same odometer reading, and a lap only joins it if its distance axis survived the same two gates the corner profiles use — length and shape against the session's own laps. A brake channel comes out fractional: 0.4 means on the brakes four laps in ten there, which is a truer thing to draw across sixty laps than a pedal that is on or off on a lap nobody drove. Gear is rounded, because half a gear is not a thing.

The average lap cannot follow the pace card's live controls. There is one average per driver, built when the race was cached, not one per setting — the corner and straight cards can follow those controls because they ship a number per lap, and a trace per lap would be 31 MB a season.

Stored at the resolution it was measured; drawn at 1 m. FastF1's car data is about 300 real samples a lap — roughly one every 14.5 m. This page stores those samples at their true distances and resamples onto the notebook's 1 m grid in the browser. Shipping the 1 m grid instead would cost 37 MB a season to carry 14× upsampling of the same 300 points.

Elapsed time is stored, not derived. Recovering it by integrating 1/v over distance lands 0.08–0.38 s out over a lap, which is most of the signal a gap trace carries, so the per-sample time is kept to the millisecond.

Four channels are derived and are labelled est. Longitudinal g is (dv/dd)·v/9.81. Lateral g is κ·v²/9.81, where the curvature κ comes from the reference lap's racing line — so it is the same geometry for everyone and differs only by each driver's own speed. Coasting is neither pedal (throttle < 5% and off the brake). Elapsed time is the stored clock.

DRS is absent because 2026 abolished it — the channel reads zero for every sample of every session, and an all-zero trace is not a measurement.

Teammates share a team colour and are told apart by line style — the first driver solid, the second dashed. Lightening the second car, as the notebook does, introduces 19 new collisions with other teams' colours, so it is not done here.

Identity is on demand, not printed on every line. The panels used to label each trace at the right-hand edge; with the whole field selected that is twenty-two stacked codes down the side of the panel carrying the most shape, and it answers a question nobody asks — a speed trace is not read at the point where it ends. Click a line and it is picked: drawn heavy, named on the chart and named in the bar above the panels, with the rest of the field dimmed to 16% rather than hidden, so there is still something to read it against. Click another to compare them, click clear air to drop the lot. The hover readout names every driver at the cursor whether anything is picked or not.

Out lap lays the last 1 000 m before the fastest lap began in front of it, on a negative ruler that ends at the line — how the car arrived: the final corners, the tow, where the throttle went down. It is cut from the raw car feed around the lap's own start time, not from the previous lap's timing slice, so it joins the lap at the same instant and the same speed. The gap there is time to the line against the reference, zero at the line by construction; lateral g reads the circuit's last kilometre, and the state of charge is not modelled before the line. It is usually the out lap in qualifying; where the lap before was one already running (a back-to-back push, a cool-down) the card says so. It exists for the fastest lap only, so the switch is shown only there.

Every lap is picked from a box per lap, not a slider. A lap you can pick is not a continuous quantity and not evenly available: at Hungary only 7% of race laps have a trace at all, so dragging a slider handle there landed on an empty lap nineteen times in twenty and the empty chart read as a bug. Each box says which of the three states its lap is in — every selected driver has a trace, some do, or none does — and an unavailable lap cannot be clicked. The arrow keys step to the next available lap rather than the next number.

These four cards read the page's one lap, through the telemetry card's Fastest · Race average · Every lap switch — so a race is measured on each driver's race average and not on one lap on fresh tyres in clear air. They have no source switch of their own any more, and the note under the stats table names what they are on.

Everything on these cards is on one ruler. A lap sits on the shared grid by relative distance, and the page draws that fraction as metres of the reference lap — which is where the corner markers, the panel rules and the x axis all live. The apex search used to place its window on that ruler and then report the answer in the measured lap's own metres, a different number for every lap. In qualifying the two agree, because the reference lap is the lap being measured, which is why the markers looked right there and drifted at a race: 10–50 m at most races, and 216 m at Hungary, where the fastest lap measures 4 097 m against the race's own 4 313.

An apex is the slowest sample belonging to that corner. The search window is the capture radius above, cut at the midpoint to each neighbouring marker — so a sample belongs to whichever corner it is nearest, and no two corners can compete for it. Lettered corners (1A, 12A) are dropped as segments of one complex, so Hungary resolves to its 14 official corners rather than 16.

That midpoint cut is what keeps the apexes in corner order. Hungary's turns 6 and 7 are 36 m apart against a 60 m radius: with plain overlapping windows, turn 6 was processed first and took the slower sample at 2373 m, leaving turn 7 pushed back to 2338 m — behind it. The straights built by walking that list came out T5–T7, T7–T6, T6–T8, a lap running backwards. Disjoint windows make the order correct by construction, at any radius.

Corner types are locked to the reference driver. Each corner is classified once, from the quickest selected lap, and every other driver is bucketed by that same classification. Typing each driver's own apex speed independently would move a corner between buckets from driver to driver, so the Avg Slow / Medium / High columns would not be averaging the same corners.

A corner the reference driver has no apex for keeps its own per-driver type, and the note above says so.

On the race-pace reading the merge happens lap by lap, then averages. Taking the peak of each averaged run instead would be the mean of a maximum against the maximum of a mean, and those are different numbers whenever a car is quicker down one run on one lap and down the next on another. So the cache stores the peak of every run between consecutive corners, and the page maxes across the runs a merged straight swallows within each lap before averaging over laps.

A straight is the run between two selected corners, and its value is the highest speed reached anywhere in that run. There is a button for every one of them, named by the corner it ends at: pressing T10 drops the straight that runs into turn 10, which is the one before it. The final run has no corner to end at, so it has its own Finish button.

The corners row does the merging. A corner that is switched off there does not bound a straight either, so dropping turns 12 and 13 joins the runs into a single T11–T14 straight and the peak is read across the whole of it. That is the control to use when a real straight has a kink in the middle of it — switching the straights off individually would throw away the stretch where the top speed actually is.

Nothing here is automatic and nothing is per-circuit in code. An earlier version decided this with a kink threshold — a corner losing less than 12% of its own entry speed did not split a straight. It was portable across circuits, but it could not tell a straight from a short link between two corners, and Hungary came out with ten "straights" nobody would name. Which runs are really straights is editorial, so it is chosen by hand and remembered per circuit.

Clipping is the car running out of electrical deployment before the end of a straight. The engine is still at full throttle but the battery has stopped pushing, so the car stops accelerating and, against the drag, starts to slow down — on full throttle. That is the one place a speed trace shows it, and it is what this card measures.

Top speed is the highest sample between the two apexes. End speed is the last sample of full throttle after it: the walk goes forward from the top speed until the brake comes on or the throttle drops below 95%, and reads the speed one sample before. Lost is the difference and over is how far the car travelled from its top speed to that point. A car that is still accelerating when it brakes reads zero.

A lift is not clipping. Every driver comes off the throttle a few metres before the brake, so that is still the end of the straight. Where the brake is still 50 m or more away when the throttle comes up, or never comes before the turn, the run ended on a lift and the row says so — at a race a driver lifting and coasting to save energy or fuel slows on the straight too, but off the throttle, so it is cut off at the lift and not counted as lost on full throttle. A tow and straight mode both move the top speed; neither moves the definition.

At a race it is the average over every clean lap — green, not into or out of the pit lane, inside 107% of the driver's own best — each lap measured on its own and then averaged, with the spread of the loss across those laps beside it. Everywhere else it reads the lap the telemetry card is set to, the fastest lap by default. The two turns are remembered per circuit in this browser.

S/F is the start/finish line, offered as a turn: from it the straight starts at the line, to it the straight is cut there, and a car still flat out at the line ends its run "on the line" rather than on a lift or the brake.

Telemetry draws the straight itself: speed and throttle against the distance from its start, every selected car on one set of axes. The top speed is the filled dot, the end of full throttle the open one, and the heavy stretch between them is the speed lost while still flat out. It reads the lap the telemetry card is set to - at a race the average lap - so where the chart averages every clean lap the two can differ by a few km/h.

The feed samples at about 4 Hz, so a speed is good to the sample and a short drop of 2–3 km/h is at the edge of what it resolves. Read the bigger numbers, and the pattern across the field, not a single kilometre an hour.

Only the road everyone is flat out on counts. Every car's lap is laid on the same distance ruler, and a point of the ruler is kept only where every car being compared is at 95% throttle or more with the brake off. Each car's time over those points is read off its own lap clock, stretch by stretch, and added up. The reference is the quickest of them, and every other figure is how much longer that car took over exactly the same metres.

Why the common stretch and not each car's own. A car that brakes 20 m later is flat out for 20 m more, and adding that time to its own total would charge it for being braver. Comparing only where all of them were flat leaves the corners out of it entirely, so the gap is straight-line speed - the power unit, how long the battery deploys before it clips, and the drag of the wing the team chose.

The set changes the stretch. Deselecting a car, or switching between teams and drivers, can add back metres the car you removed was lifting on, so the totals move a little with the comparison - the order rarely does. The metres used are stated under the chart. Every metre this card leaves out is counted by Time lost in corners below, and a lap that sits off the ruler is left out of both - see that card.

Teams take each team's quickest lap, the same rule as the corner-against-straight card. It reads the lap the telemetry card is set to - at a race the race average lap, where a tow or a defending line is averaged out.

The other half of the same ruler. The full-throttle card keeps the points of the distance ruler where every car compared is at 95% throttle or more with the brake off. This card takes every other point: wherever at least one of them is braking, lifting or on part throttle. Each car's time there is read off its own lap clock, and every figure is how much longer that car took than the quickest over exactly the same metres.

Together they are the lap. Every metre of the ruler is counted by exactly one of the two cards, both timed off the same lap clock, so each car's two figures add up to its lap time. Each card is measured against its own quickest car, so the two gaps add up to the lap gap only when the same car is quickest in both.

"Corners" is everything that is not common flat out. That is the braking zones, the corners themselves and the exits, and also a stretch where only one car lifted - a lift there counts as corner time for every car compared, the faster ones included. So the gap here is braking, mechanical and aerodynamic grip, traction and the driver, with a little straight-line speed from those shared stretches.

The set changes the stretch, as on the card above: removing a car that lifted somewhere hands those metres back to the full-throttle card. Teams take each team's quickest lap, and it reads the lap the telemetry card is set to - at a race the race average lap.

A lap that sits off the ruler is left out of both cards. Each lap is laid on the ruler by its own distance, and now and then a car's odometer is cut somewhere else - its brake points then sit tens of metres from everyone's, and its corner exits land on road the rest are flat out on. Each lap is slid against the field's braking, and one that is 25 m or more off is named under the chart and not counted. A lap that sits right reads 0-14 m. Every lap is already cut where the car crossed the line and its distance pinned to the road every 200 m, so this is now a check that rarely fires - once in the whole of 2026.

Downforce is grip in the fast corners, measured against the field at the same corner. Taken at the limit, a corner's lateral acceleration is the racing line's curvature times the speed squared - and the line is the same for every car - so one car's grip against the field's is (its apex speed / the field's)². Only the corners the type lock calls High Speed are read, where the grip is mostly downforce, and only where the car slowed by 4% or more for the corner: a fast corner taken flat is limited by power, not grip, and says nothing about downforce. Each car is the median over those corners. Where a circuit has fewer than two such corners (Monza) the medium-speed ones carry it, and the note says so.

Drag is fitted from the straight-line acceleration. On full throttle, brake off, on a straight (racing-line radius over 800 m), from 150 km/h, the car's acceleration is its power over mass and speed less its drag: a = α/v − k·v², with k = ρ·CdA / 2m. Fitted by least squares over the lap, the power term soaks up whatever the power unit is doing - deploying, harvesting or clipping - and what is left is the drag. CdA in m² assumes 798 kg and 1.18 kg/m³ of air, so it is a model; the index against the field is the measurement.

Efficiency is downforce index ÷ drag index × 100: grip bought per unit of drag. On the chart it is the diagonal - above it, more downforce than the field for the drag.

What moves it. A tow on the lap read lowers a car's drag; a lap with a mistake in a fast corner lowers its downforce. Practice and qualifying read the fastest lap, a race the race-average lap, like every card here. Tried and dropped: fitting lateral g = A + B·v² per car, which over six corners trades the two parameters off and put the quickest cars last on downforce.

Load ratio (the bars) is the quick read the card started from: the speed through the fast corners as a share of top speed. The corners are the downforce index's own; a car's fast-corner speed is the field's mean apex there times the car's median ratio to the field, so every car is read over the same corners; a team is the mean of its drivers, each ticked on its bar. Top speed carries power, deployment and any tow as well as drag, which is why the physics view separates them.

Full throttle % is a share of the lap by time, not by distance, and it reads 1–3 points below the notebook's figure. The notebook masks Time.diff() by the throttle at each interval's right endpoint, so an interval where the pedal climbs 50 → 100 counts in full — a right-endpoint Riemann sum, biased high. This solves for where the trace actually crosses 98% between two samples and counts only the time beyond it.

Top speed and minimum speed are this lap's, not the session's. The session-wide top speed is a separate chart.

Avg corner speed is the mean of every official apex. Avg straight speed is the mean of the per-straight peaks, so it depends on the controls above. Avg Slow / Medium / High average the apexes in each locked bucket, and read — where a circuit has none of that type — Monaco has no high-speed corner.

Δ S/E is the speed crossing the finish line minus the speed crossing the start line of the same lap. It reads negative for almost everyone, and that is the measurement working: a driver who has just set their fastest lap backs off on the way past the line, so they end the lap slower than they began it. A figure near zero is a driver who stayed on it into the next lap.

Both axes come from the turn selection above. Average corner speed is the mean of the apexes of the selected corners; average straight speed is the mean of the peaks of the selected straights. Switching a turn off in either row moves the matching axis, so the two views are always of the same circuit you have described.

The crosshair is the mean of whoever is shown, not a property of the circuit — deselect a driver and it moves. The quadrant it puts a car in is therefore a statement about this group of cars, on this day, at this track.

Team view takes each team's quickest lap, which is the notebook's version. Driver view is the default here so the driver filter has something to act on; the two read the same measurements.

This card follows the source switch too. On the race-pace reading a car's position is where it ran for two hours rather than where it was for ninety seconds, which is the more honest read of a car's balance — a wing chosen for the race is not the wing that set the fastest lap.

High corner speed with high straight speed is the efficient corner — more downforce without the drag that usually pays for it. The opposite corner is the car giving away both.

Neither figure is the speed trap. The lap table's SpeedST is a reading at one fixed point that cars are often still accelerating past. These are the maximum of the car's own speed channel — the first over the fastest lap, the second over every lap the driver ran.

The session figure is always at least the lap figure, and the gap between them is the interesting part: a large one means the driver's quickest lap was not their quickest down the straight — a tow, a different wing, or an earlier run on fresher tyres.

A driver whose speed channel did not survive is absent from both, because both read that channel.

The ideal lap is the sum of a driver's three best sectors, each taken from whichever lap of the session produced it. It is a lap nobody drove.

The gap is the interesting number. A small gap means the driver put their whole lap together in one go; a large one means the pace is there but has not been strung together yet. It is a measure of the session, not only of the driver — a red flag, traffic or a single hot lap on new tyres all show up here.

This chart does not need telemetry, so a driver whose speed channel died is still on it.

The two rows answer different questions. The fastest lap's sectors add up to a lap that was actually driven. The session bests do not — they are the same three numbers the ideal lap is built from, each from whichever lap produced it.

Reading them together shows where a driver's ideal-lap gap is: a sector that is much quicker on the bottom row than the top is the sector they have not yet put into a complete lap.

Neither needs telemetry. Both come from the timing feed.

Each stripe is one lap laid out flat, from the line to the line, coloured by the gear the car is in or by what the pedals are doing. Read down a column to compare drivers at the same point on the road: a driver a gear higher through a corner is carrying more speed through it, and a driver braking earlier gives the stripe away before anyone else's.

Throttle states are the notebook's four: braking whenever the brake is on, then full at 95% throttle or more, partial above 5%, and neutral — coasting — at or below it. Braking wins over throttle where both are registered, which is what happens on the way into a corner.

The 95% here is not the 98% in the Driver Stats table, and they are not measuring the same thing. This is a coarse four-state map of the lap on the notebook's own threshold; that one is an exact-crossing integral of time spent at or above 98%. Expect this to read a point or two higher. Neither is wrong — they answer different questions — but do not quote them as the same number.

The percentages are integrated, not counted. Time sums the elapsed time between samples, so the slow parts of the lap weigh what they should; distance sums metres. They differ by a lot and that difference is the point — third gear is a small share of the lap's distance and a large share of its time.

Gear is rounded off the 1 m grid. The trace is resampled before it is read, so a shift that lands between two grid points is placed to the nearer metre. It moves a boundary by under a metre and cannot move a gear.

Colours are validated, not the notebook's. Its throttle map is green / yellow / grey / red, and green against yellow measures an OKLab ΔE of 3.5 under deuteranopia — full throttle and partial throttle are the same colour to the most common form of colour blindness, which is the one distinction the chart exists to make. These four separate by at least 9.6 under every simulated deficiency. Gear runs one hue dark to light, so a higher gear is always a brighter stripe, with the number printed on every segment wide enough to hold it.

The pedal is a bit, not a pressure. The feed's brake channel says the car is on the brakes or it is not — there is no force in it — so nothing here measures how hard anyone brakes. What it does carry is where the pedal goes down and comes up, and the speed trace says what that achieved.

A race is measured on the race average, not on the fastest lap. Every figure on these cards is the median of that driver's clean green-flag laps at that corner — the same switch the corner and straight cards make, and for the same reason: a fastest race lap is one moment on the freshest tyre the stint had, in the clearest air the driver saw. Practice and qualifying keep the fastest lap, which is the only lap those sessions are about. The median is taken per corner and per field, so a driver's latest brake point and their longest trail do not have to come from the same lap. A corner with fewer than four clean laps behind it is dropped rather than averaged.

Brake point is metres from the start line to where the pedal went down, so a larger number is a later brake. It is shown as a gap to the latest braker of the selected set, because the absolute metres are a property of the circuit and the difference between two cars is the property of the cars.

The edge is a midpoint. The samples are about 14.5 m apart, and the true moment the pedal moved is somewhere between the last sample off it and the first on it — so it is reported as the middle of the two, which is the unbiased estimate. One corner on one lap therefore carries about ±7 m of noise. That is why nothing on this page is drawn from a single corner of a single lap without saying so.

A zone has to end at the corner it was for. Down a long straight the car may brush the brakes for traffic; a zone is only this corner's if the pedal comes up within 120 m of the apex, and where two applications both qualify the one that shed the most speed is taken.

Some sessions are missing. A lap whose brake channel reports the pedal down for more than 40% of the lap, or down and at full throttle at once for more than 2% of it, is not physically possible and is left out entirely rather than measured. That is the same test the season and circuit pages apply. Across 2026 it removes about one lap in seven — the Hungarian Grand Prix goes almost completely.

Deceleration needs three samples. Over two it is one speed difference divided by one time difference, which on a short zone reports rates no car can produce, so it is left blank there while the brake point — which the same two samples give perfectly well — is kept.

Trail is the share of the corner's speed loss achieved with the pedal still down: (von − voff) ÷ (von − vapex). 100% means the driver released at the apex having done all of the deceleration on the brakes. 70% means they came off early and rolled the last third of it off-brake. It is a ratio of speeds measured at one corner on one lap, so it compares between Monaco and Monza in a way that metres never can.

This is not a pressure trace. Trail braking properly means bleeding the pedal off as the steering goes on, and the feed carries no pedal force to measure that with. What this measures is the consequence — a driver who trails deep is still slowing the car when they let the brake go, and one who does not has already finished.

Release is the signed distance from where the pedal came up to the apex. Positive is off the brakes before the apex; negative is still braking past it. It draws the same story in metres, and unlike trail it is not comparable between corners — 20 m into a hairpin is not 20 m into a fast sweep.

The throttle gap is the time from the brake coming up to the throttle crossing 20%. Positive is a coast: the car is on neither pedal. Negative is a genuine overlap — throttle fed in while the brake is still on, which is left-foot braking or a driver settling the car against the rear brakes. 20% rather than full throttle, because the question is when the driver starts picking the car up, which out of a slow corner never reaches 95% until the exit is unwound.

Every figure is a median over the driver's corners, not a single corner. One zone carries ±7 m of edge noise; a median over five or six of them does not.

Two forces pull the braking point in opposite directions. A car sheds about a kilogram and a half of fuel a lap and gets easier to stop, which moves the brake point later. The tyre goes off and stops it less well, which moves it earlier. The slope here is the sum of the two, so a positive number is a stint where fuel burn won and a negative one is a stint where the tyre did.

Over the 2026 races the established fits run to a median of +0.8 m per lap — fuel burn wins more often than not, which is what should happen.

The fit is least squares over a trimmed sample, not Theil-Sen. Theil-Sen is the usual robust choice and it is the wrong one here: the brake points land on an 11.4 m grid, so most pairs of adjacent laps differ by exactly zero and the median of the pairwise slopes collapses onto zero with them — it returned a slope of exactly 0.00 for 561 of the 788 stints it claimed to have fitted. Least squares is unbiased under that noise; laps further than three median absolute deviations from the stint's own median are dropped first, so a lap spent in traffic or with a lock-up cannot drag the line.

Grey means the drift did not clear its own noise. A stint is drawn solid only where the slope exceeds twice its standard error. Across the 2026 races 587 of 3747 stint fits clear that — the rest are stints whose braking point did not measurably move, which is a finding rather than a failure.

Only clean laps count: green flag, no lap into or out of the pit lane, and a lap the timing feed gave a time to. A stint needs six of them at a corner before it is fitted at all.

The metres are a common ruler. Each race lap is stored against that driver's average lap on relative distance, so the same index is the same point of the circuit on every lap. That is what makes a drift measurable at all — the question is whether the car brakes earlier at the same corner, not whether today's speed channel integrated 1% long.

A long run lap is one inside a window around the driver's own quickest lap — the notebook's 102% to 108%. The lower bound throws out the qualifying simulation, the upper bound throws out in-laps, out-laps and anything behind traffic. Because the window is relative to each driver's own pace, a slow car is not measured against a quick one.

A stint has to be at least four laps to count, so a single lap on a new set is not read as race pace. All three bounds are controls above.

Pace ranks drivers by the median of their long run laps and draws the spread: the box is the middle half, the line the median, and every lap is plotted behind it, coloured by compound. The median is used rather than the mean because one lap caught behind a slower car drags a mean and barely moves a median.

Tyre life plots each stint against the age of the tyre, which is where degradation shows. The slope is fuel-corrected. A car sheds roughly 0.03–0.07 s a lap as fuel burns off — the same order as the wear it competes with — so a measured slope is wear minus fuel burn and reads far too flat, often negative. This circuit's own fuel effect, the one /retrieve_fuel fits, is added back. The correction can be switched off to see the raw slope.

Tyre fit is the same stints as the carousels' tyre-fit slide: one panel per compound, every long run lap a dot at its tyre age, a least-squares line through each stint and the fit as a table. Teams fits one line per team and compound with the slope taken inside each stint, because a practice car is refuelled between runs and two runs are two different cars as far as lap time goes.

Compound colours are F1's own and always carry the S/M/H letter, because they sit outside the lightness band this page's other colours are held to and cannot be told apart on colour alone.

None of this needs telemetry, so a driver whose speed channel died is here.