Allison Kelly and the Data Gap Inside UVA's Class of 2028 Recruiting File

I reopened the University of Virginia women's swimming recruiting board on a midweek evening, right after closing out a GPS analysis sheet from a V.League match. UVA's class of 2028 board carries five names side by side: Karina Plaza, Shelby Hutchinson, Skylar Zulegar, Zayda Miehl and Allison Kelly. The first four come with national ranking markers printed next to them — Plaza at No. 5, Hutchinson at No. 7. The fifth name comes with something else entirely: 2 minutes 03.24 seconds in the long course 200 freestyle.

The women's 200 freestyle world record sits around 1:52. A gap of more than eleven seconds says nothing about the future of a 17-year-old. The way that gap is presented does. Across the entire recruiting file published through SwimSwam's college recruiting channel, there is not a single line about stroke rate, distance per stroke, reaction time off the blocks, the 15-metre underwater segment, or turn splits. There are final results. There is no process that produced those results.

What made me linger on this name was not the time itself. It was the empty cells sitting beside it.

Context: one file, two coaching systems

Allison Kelly comes from Jupiter, Florida. She is attached to Bolles School and Bolles School Sharks, and previously to Jupiter Dragons. In her thank-you note after announcing the commitment to Virginia, she names Coach Peter, Jake, Alexis and Claire at Bolles, Coach K at Jupiter Dragons, and Coach Todd at UVA. Six names spread across two training systems, at two different stages of a career that has not yet reached 18.

Bolles School, based in Jacksonville, is known within American high school swimming as one of the stronger technical development programmes. That is context, not evidence. I need to state this up front, because it is exactly where sports data analysis tends to slip: a well-regarded programme does not automatically transfer its reputation onto every individual who walks out of it.

On the destination side, Kelly joins UVA's women's swimming and diving programme. Two time labels tend to get read together and should not be. The class of 2028 label is her high school graduating and recruiting class. The class of 2032 label is her expected college graduation year. Some reports merge the two and create a sense of contradiction. There is no contradiction — only two different timestamps recorded for two different events.

The competitions in the file also sit on two different tiers. The school tier holds the Florida High School Class 1A State Championships. The federation tier holds the Florida Swimming LSC Senior Short Course Championships. Ahead of her sits the 2026 LCM season — long course, measured in metres.

These three tiers cannot be compared directly. A high school result describes a position inside one state. A federation result describes a position inside a wider and deeper age group. A long course result describes the ability to convert between two measurement systems — and that is the tier I care about most, because it is where most recruiting projections fail.

One more layer concerns the ecosystem. This file arrives through SwimSwam's college recruiting channel, a media outlet specialising in swimming recruitment. Kelly has also attended Fitter and Faster swim camps. There is a clearly commercial structure here: the media channel generates attention, the camp generates experience, and the commitment announcement generates a media moment for all three parties. That structure does not make the data wrong. It simply means the data was selected to tell a particular story.

Core: the evidence chain and the measurement gaps

Start with the raw data. Kelly holds personal bests in both course types: short course yards (SCY) and long course metres (LCM). Her event range is broad — freestyle across sprint and mid-distance, backstroke, butterfly and individual medley. For a high school athlete, that breadth is a meaningful data point, but I need to place it on a tier.

Event breadth can be read two ways, and both are technically valid. The first reading: a strong general technical foundation, high adaptability, developmental potential in several directions. The second reading: no single event has yet reached the threshold required for specialisation. The same data, two opposing conclusions. Which reading you choose depends entirely on what the source article does not provide: the mechanics of each event.

I built a check table for this file the way I still build one for player GPS data. Five technical indicator groups are needed to assess a swimmer: start and underwater phase, turns and finish, stroke efficiency, stroke rate across the race, and venue adaptability. In Kelly's file, the first four groups are entirely empty. No technique video. No reaction data. No 15-metre split. No stroke count per length. No turn time.

The fifth group — venue adaptability — is the only one that permits inference, and that inference is very weak. The existence of personal bests in both SCY and LCM shows she does not depend on a single course type. But a list of results does not tell us whether her conversion between courses is better or worse than the age-group norm. I marked that inference at the lowest confidence level in the entire file.

Here I have to be direct about the projection tier. There is a working hypothesis, not a conclusion: her strongest long-term event family is most likely freestyle, particularly the 200, with backstroke as a secondary option. That hypothesis rests on the structure of her event range and the placement of the 200 free inside the file, not on stroke analysis. I rate it low to medium confidence and will have to revisit it the moment split data appears.

Another detail worth noting sits in the comparison structure. Within the same UVA class of 2028, Karina Plaza and Shelby Hutchinson are identified by national ranking — No. 5 and No. 7. Kelly carries no ranking marker in the available information. That does not mean she is weaker; it means the public data system had not assigned her a national ranking position at the time of publication. For a data analyst, the absence of a ranking label is a data point, not a meaningless silence.

On performance positioning, I placed the 2:03.24 long course 200 free against three reference rings. Ring one, the world record: a very large gap, more than eleven seconds. Ring two, the international tier: not reached. Ring three, the high school and college recruiting tier: this is the tier the time belongs to. The conclusion is not to undervalue or overvalue it, but to place it correctly. A recruiting file is a recruiting file.

A small GPS deviation taught me everything about verification. In 2026, while working as a data consultant for Sanna Khanh Hoa BVN, I miscalculated a striker's sprint distance — logging 1.2 km when the real figure was 0.8 km. The error came from the synchronisation software, not from the measurement itself. But it exposed four other systemic errors across three months of data, and from that point a two-source cross-check became the club's internal standard. The lesson applies here very concretely: a single data point, however good or bad it looks, is never enough to conclude anything.

In Kelly's file, that single data point is 2:03.24. It is not bad. It is also not proof of an immediate championship-final berth. It is a data point with no vertical axis to compare against, because the interpolation curve from splits is missing.

Why does the absence of splits matter so much in swimming? Because swimming is a sport where the final result is the sum of four segments that can be separated and assessed independently. Two swimmers can both touch in 2:03.24 with completely different profiles: one goes out fast and collapses over the final 50, one goes out slow and surges home. In recruiting terms, the same number, two opposite projections. The fast opener has a different developmental ceiling than the strong closer. With only the final time, we cannot tell them apart.

Turn time works the same way. In short course yards, turns occur roughly twice as often as in long course metres over the same race distance. A swimmer with strong turn mechanics benefits disproportionately in short course, and their gap between SCY and LCM results will be wider than usual. When the short course to long course conversion data is not published, every projection about college-level performance has to drop one confidence step.

I believe in numbers, but only after a number has cleared three rounds of verification. For Kelly, round one — the number exists and is consistent across sources — is cleared. Round two — the number broken down into its component splits — is not cleared, because the data has not been published. Round three — the number placed against age-group norms and specific competition conditions — does not yet have a sufficient sample to run. Two of three rounds are still pending. The correct conclusion right now is an open one, and I have no intention of closing it early just to make the article read more decisively.

There is one more methodological point. When a file is missing all four technical indicator groups, an analyst has two options: stop and state the limits clearly, or fill the gap with inference drawn from programme reputation. The second option reads better and produces a more shareable article. It also produces most of the error in recruiting projections. I choose the first, even when it makes my ending less certain.

Contrarian angle: programme reputation is not individual ability

There is a very common reasoning error when reading recruiting files, and it appears in almost every recruiting commentary I have read in eighteen years on this beat. The error is converting the reputation of a training programme into a forecast about an individual's ability.

Bolles School has a strong reputation for technical development. That is true. But the relationship between a good programme and an athlete who matures inside it is correlation, not causation. A good programme can hold swimmers who never break through. A talented swimmer can mature in a low-profile programme. In 2026, when I declined to recommend signing a foreign striker from the Thai League for a V.League club, the club leadership told me data could not replace the eye for a player. That striker had scored 18 goals from just 11.2 xG in 19 matches, with 70 percent of his goals coming from set pieces dependent on a system. After the move, he scored 4 goals in 20 matches and suffered two hamstring injuries. That story does not say the eye was wrong. It says the reputation of the place you came from is not the ability of the person who arrived.

The second error is subtler. Event breadth is often read as a sign of all-round talent. Sometimes it is. Often at the high school level it reflects that the athlete has not yet found a signature event, and not finding a signature event at 17 can mean specialisation will begin later than for peers. In American college swimming, where the schedule is dense and every championship berth must be earned, late specialisation is a risk variable, not a bonus.

The third error is reading recruiting class size as a quality index. UVA has five athletes in the class of 2028. A large class shows the programme is recruiting hard, not that each individual in it will contribute proportionally. Size is a programme metric. Contribution quality is an individual metric. Two different tiers, and I see them merged almost every recruiting cycle.

Finally, there is a technical risk I have to flag clearly in this file: every technical judgement lacks split data to support it. That is not the athlete's fault. It is the limit of the public record. And in my line of work, recognising the limits of data matters as much as reading the data correctly.

What to watch in the next cycle

The 2026 LCM season is the next data marker I will wait for. Not to see what Kelly's final time turns out to be, but to see whether split data appears for the first time.

Specifically, I need three things. First, the first 100 and second 100 splits in the 200 freestyle: if the differential between the two halves narrows, the profile changes character. Second, the gap between SCY and LCM results in the same event: if that gap is smaller than the age-group norm, turn mechanics and the underwater phase are better than current data suggests. Third, reaction time off the blocks: this is the least published indicator but also the one that can improve fastest between ages 17 and 20.

None of those three sits on a results sheet. All of them sit in the data layer that recruiting reports rarely touch. Data does not tell stories; it records everything so that I can tell them myself. And most of the real story of a young athlete lives in the columns a results sheet has never had.

Allison Kelly is seventeen, committed to Virginia, has passed through two training systems, and carries a broader event range than most of her cohort. That is what the available data says. What the available data does not say is far longer — and that is the part I will keep reading when the 2026 long course season opens the board.

A culture of support does not live in the volume of the cheer. It lives in the frequency of patience.