Cost Cap and the Convergence Curve: Three Seasons of Data and Formula 1's Shifting Order

Cost Cap and the Convergence Curve: Three Seasons of Data and Formula 1's Shifting Order In 2026, Max Verstappen won 19 of 22 races. Red Bull dropped exactly one — Singapore, where Carlos Sainz broke a perfect streak with a pure DRS-holding strategy rather than with pure pace. Twelve months later, Verstappen still took a fourth title, but won only 9 races, and the constructors' championship went to McLaren for the first time since 2026. Looking at the champion column, nothing changed. Looking at the whole table, the structure beneath it cracked open and reassembled into a different shape. That is the starting point of this analysis. I do not care who wins the title — I care about the slope of the dispersion curve. When the gap between the fastest car and the tenth-fastest car narrows, what is changing is the rulebook, not the driver roster. Data never hurries, but people always do. CONTEXT: THE MECHANICS OF AN ECONOMIC EXPERIMENT The cost cap is not a sporting rule. It is an economic experiment bolted onto a sport, and F1 began running that experiment in 2026 with a spending ceiling of roughly $145 million for 21 races, stepped down to about $140 million for 2026 and around $135 million for 2026, with per-race additions tied to calendar length. Alongside it sits the ATR — a sliding-scale limit on aerodynamic wind-tunnel hours and CFD runs, allocated by the previous season's constructors' position. The last-placed team gets more running than the champion. Precision matters here. The cost cap limits money. The ATR limits learning time. The two mechanisms are technically independent but consequential in tandem: a rich team with trimmed wind-tunnel hours cannot convert money into faster lap time, and a poor team with extra hours only converts that time into performance if it has the personnel to read the data. Based on my years of watching races, this is the first time F1 has imposed a control variable on itself. Before 2026, rich teams bought time, time bought understanding, understanding bought speed. That causal chain was short and direct. After 2026, it was cut in the middle. The problem is that none of us — including the chief engineers — has a full forecasting model for an experiment like this, because it has never run through a complete cycle. Three seasons is a small sample. But it is what we have, and in data analysis a small sample still beats a large hypothesis. EVIDENCE FIRST: THREE INDEPENDENT SAMPLES From 2026 to 2026, the number of teams winning races in a season rose from the low levels of the 2026–2026 period. In the Mercedes era, many seasons produced only three or four teams reaching the top step. From 2026, the winners' list grew: Red Bull, Ferrari, Mercedes, McLaren, and at points Aston Martin. The number of teams appearing on the podium in a single season widened too. That number proves nothing on its own. It is a signal, and I tell young editors: a signal is not evidence. A signal is a number that deviates from expectation. Evidence is a number that deviates from expectation in the same direction across multiple independent samples. So take three samples. Sample one: qualifying gaps. In 2026–2026, the gap between pole and tenth in Q3 usually landed between half a second and nearly a second, sometimes exceeding a second at long circuits. From 2026, that gap compressed, and more revealingly in Q1 — where the distance between the leading team and the backmarkers closed faster than in Q3. This is the detail the media usually misses. Convergence does not happen at the front. It happens at the back. The fastest team stays fast. It is positions five through ten that move closer. Sample two: in-season development rate. A crude but useful measure compares each team's gap to the leader in the first and last races of the season. Before 2026, the familiar pattern was rich teams starting slow and improving with updates every weekend while small teams hit their development ceiling mid-season. From 2026, that reversed for some teams: the small ones kept bringing updates while the big ones had to ration, because their budgets were maxed and their tunnel hours were capped. Sample three: points distribution for fourth and fifth in the constructors'. If convergence is real, the points gap between the midfield and the front must narrow over time, not widen. In 2026, at one stage four teams sat within a hundred points as the season closed — a structure the Mercedes era rarely produced. THREE SEASONS, THREE DIFFERENT STORIES But here I must be careful, because the three samples are not fully independent. They all carry the influence of one enormous variable: the 2026 technical regulations. In 2026, F1 moved to the ground-effect era, with floors designed to generate downforce from airflow beneath the car rather than from wings. That was an almost total aerodynamic reset. Teams that had invested in the right direction in the wind tunnel could leap forward in a single season regardless of budget. Red Bull did exactly that, winning 15 of 22 races in 2026 — and then 21 of 22 in 2026. Read that number again. If the cost cap caused convergence, then 2026 should have been the most convergent season. It was not. It was the season one team won nearly everything and one driver finished ahead of the runner-up by the largest points margin in the sport's history. At sixty, I no longer believe in luck, only in numbers that have not yet spoken. The number here says this: the cost cap did not create convergence. The new regulations created dispersion, and the cost cap gradually eroded that dispersion. These are two opposing forces, and anyone claiming "the cost cap is making F1 more exciting" is blending two different mechanisms into one sentence. I want to rebuild the timeline, because timeline is what data analysis must never skip. 2026: the cost cap begins, but at a high level and without an ATR mechanism strong enough to make a difference in one season. The title fight between Verstappen and Lewis Hamilton ran to the final lap at Abu Dhabi, where a safety-car call decided the championship. A dramatic season — but the drama came from a non-repeatable variable, not from a new economic structure. 2026: aerodynamic reset. Red Bull dominant. Ferrari fast but fragile. Mercedes wrestling with porpoising. The gap widened. Teams were still learning to operate the cap. 2026: peak concentration. Red Bull won 21 of 22. No other team won more than one race. 2026: decomposition. Red Bull's aerodynamic edge eroded as others caught up. McLaren rose, Ferrari and Mercedes traded wins. Verstappen was still good enough to take the drivers' title, but the constructors' crown went to Woking. This chain is not linear convergence. It is an inverted U: widening, peaking, then narrowing. And the narrowing — 2026 — arrived precisely when the accumulated development spend of some teams had been suppressed by the cap and ATR long enough for others to close in. WHY CONVERGENCE COMES FROM LEARNING TIME, NOT MONEY An F1 team does not buy speed directly. It buys the ability to learn faster than rivals. The loop is: hypothesise, simulate in CFD, validate in the tunnel, build the part, run it on track, collect data, compare against the model, revise, repeat. Each loop costs money, tunnel hours and human hours. The cost cap limits the first dimension. The ATR limits the second. The third — human quality — is not limited by rule, and that is where big teams retain an edge. This is why I say convergence in the midfield is real, while convergence at the front is an illusion created by insufficient data resolution. We see the fifth-placed team closing on the third. We do not see that the third had spent 100% of its budget and tunnel hours by July, while the first still had human headroom to optimise what budgets cannot measure. I spent years pricing footballers in the transfer market. The transfer market is a game where whoever values correctly wins. There I learned a lesson that transfers to F1: a transfer fee is not value, and a spending cap is not a capability cap. A team can spend its full budget on the wrong update and stand still while another spends less on the right one and advances. This leads to a paradox I call the ATR paradox: weak teams are given more testing time, but weak teams are usually the ones with unfinished processes. Adding time to an inefficient process produces more data, but not necessarily more understanding. It produces more noise. Policy makers forget this: you cannot redistribute capability by redistributing resources. You can only redistribute the opportunity to learn, and opportunity converts to speed only if someone knows how to use it. THE RESOLUTION PROBLEM: WHY READING THE RIGHT NUMBER IS THE HARDEST STEP When a team declines, the media reflex is to find a single cause. But a season's data is multivariate, and in multivariate data there is no single cause. There is a combination of interacting variables that changes race by race. The second reading is subtler and easier to hide, because it produces no headline. Nobody wants a headline reading "the aerodynamic operating window narrowed." But that is usually the truth. The empty stands of 2026 exposed a truth: many things we call character are just noise. I repeat that because it bears directly on this topic. In 2026, with races run without spectators, a variable was removed from the equation. No crowd, no grandstand pressure, no home advantage. And the finishing order barely changed. That showed most of the gap between teams comes from the machine, not the mind. It was rare evidence of separating noise from structure. The same must be done for the cost cap: separate its effect from the regulations, from the natural development cycle, from the movement of a few elite drivers. Most cost-cap analyses I have read over three years conflated correlation with causation. They saw teams closer, they saw a cap in place, they concluded the cap brought teams closer. Methodologically, that inference is wrong. A proof of how easy the error is: compare the average pole-to-tenth gap in 2026 and 2026, and you may see a narrowing. But compare 2026 and 2026, and you may see the same narrowing — and 2026 had no cost cap. Part of what we call convergence is simply the sport's natural cycle: after a period of dominance, others always catch up, because sufficiently stable rules let knowledge spread. I call this the cycle-imitation law: every era imitates the previous era's data, but nobody learns. In F1 too, teams always learn from the champion, but they learn one regulatory cycle late. When rules change, accumulated knowledge is partly invalidated and the order is reshuffled. The cost cap does not erase that cycle; it only adjusts its speed. THE CONTRARIAN ANGLE: WHAT IS BEING MISREAD The most misread point about the cost cap is the belief that it was designed to create convergence. It was not. It was designed to create financial sustainability. Convergence, if it exists, is a side effect. And side effects are, by definition, not optimised. Read the rules and drafters' statements: the stated aim was to reduce spending inequality and increase organisational competitiveness, not to guarantee a lap-time gap under a thousandth of a second. Those differ. Spending can equalise while pace stays unequal, if the ability to convert spending into pace differs between teams. Conversion capability is a function of many variables: simulation process, feedback speed between track and tunnel data, quality of the chief engineering group, and most importantly the stability of the rulebook. When rules are stable, everyone's conversion capability rises, but strong teams rise faster. When rules change, conversion capability falls everywhere and the gap is compressed. Therefore convergence is strongest right after a rule change, not at the end of a stable cycle. The three cost-cap seasons were structurally dominated by the 2026 regulations more than by the $135 million figure. A second paradox: stability lets engineering advance fast, but also lets the leader pull further ahead. Instability brings teams closer, but lowers overall technical quality. This is why I disagree with the popular view that F1 should change rules more often for competitiveness. In the medium term it creates a sport where winning belongs to whoever adapts fastest to change, not whoever understands the sport most deeply. Another contrarian point: the cost cap may be quietly lowering technical quality at the deepest layer, in ways nobody measures, because lap time does not measure them. With a capped budget, teams must choose between developing the current car and researching the concept for the next. Before 2026 they could do both. After 2026 they must choose. The rational choice always tilts toward optimising what exists, because points are paid for it. Long-term basic research — the kind that yields conceptual leaps — gets pushed back. WHAT CHANGES FROM 2026 2026 is the next inflection point. The power-unit rules shift to a roughly even split between combustion and electric power, with a substantial increase in electrical output and a move to sustainable fuels. Aerodynamics moves to active aero with two states for straights and corners. It is a reset at both ends of the car — engine and aero — at a scale not seen simultaneously in recent history. Alongside this, the team structure changes: a new manufacturer enters as an owner, and an eleventh team is admitted, expanding the grid. In data terms, the important event is not the team count but the addition of a variable in the allocation of engineering talent. More teams — especially a new manufacturer hiring from established ones — means a transfer wave of technical staff. Talent movement is an early indicator of capability movement, typically leading on-track results by eighteen to twenty-four months. Here is a concrete signal to track rather than a prediction. In a regulatory reset, watch not the first race win but the improvement rate from race five to race twelve. In that window, aero departments deploy first updates built on real track data, not simulation. The slope there tells you which team has the fastest learning loop — and in a cost-cap era, that loop is the only competitive edge the rules cannot cap. One concrete fact worth placing beside this signal: after a team breached the cost cap in the early years of this era, the regulator imposed a combined penalty of a fine and a reduction in aerodynamic testing time. That is the era's most important precedent, because it establishes that the currency of punishment is not money but time. The system itself recognises that learning time is scarcer than cash. THREE QUESTIONS THAT NEED DATA First, is there an optimal budget threshold for converting spend into pace, as with any production model? If so, teams below it are inefficient and teams above it are wasteful. Which side of that threshold does the current cap sit on? Second, is convergence durable, or just a pause between two regulatory resets? Answering needs at least three regulatory cycles. We are in the first. Any conclusion drawn today is a small sample. Third, is the cap shifting advantage from spending to allocation — and if so, which indicator measures it? The most plausible is the ratio of successful updates to updates brought to the track. If that ratio is higher at small teams, allocation is a learnable edge. If it is equal, money still wins. A METHOD NOTE The three independent sources I cross-check are: official regulator and organiser statistics on grid and finishing positions; publicly available lap-time data; and the cost-cap and ATR regulatory documents. When they disagree, I record the discrepancy rather than pick the convenient source. When there is only one source, I label the conclusion a hypothesis, not a fact. I do that because I have been on the wrong side of it. Early in a scouting-data project I worked on, I drew a conclusion from an uncross-checked dataset and it was wrong. Since then I hold a rule: never publish a judgement on a single source, however credible it looks. That is why I deliberately give no 2026 champion prediction here. Not because I have no view, but because I lack three independent sources to defend it against the data. WHAT TO WATCH NEXT If I reduce this to one signal: the improvement slope between race five and race twelve of 2026, measured as a percentage gap to the leader. If that slope is steeper for the teams lower in the 2026 constructors', the cap and ATR are working, and convergence is real. If it remains steepest at the front, then what we called convergence was a temporary consequence of one regulatory reset, and the spending cap has not yet touched the variable that matters: the speed of learning. I will not predict the outcome. I will set a spreadsheet, a tracking calendar and a deadline. By race twelve the data will answer. And if it answers against my suspicion, I will rewrite — because that is the whole point of working with data rather than belief. A sport can be more exciting without measuring anything. But a sport can only be sustainable if someone measures the right thing. Three cost-cap seasons have given us a small sample and an inverted-U curve. The question for next season is whether that curve straightens out, or is merely the pause between two reshuffles. I leave the question open. The numbers have not yet spoken, and I no longer have time to guess.