After Salah and Trent: Why Liverpool’s System Flattened

Liverpool entered the 2025-26 season as favorites to win the Premier League. Not only did they win it the previous year, but Liverpool also spent 570 million dollars, the most money ever spent by a single club in one transfer window. Many pundits and football fans tipped Liverpool to break records because not only did they break the spending record in total, but they also broke the British transfer record on a single player twice, bringing in Alexander Isak and Florian Wirtz. To say Liverpool are not on title-winning form would be an understatement. As of writing, Liverpool are sitting in 5th, ten points behind league leaders Arsenal. The reasons behind this decline are numerous, and the simple narratives used to explain it fail to capture the full nuance.

Arne Slot’s system last season set Liverpool up to maximize Mohammed Salah’s output through Trent Alexander-Arnold’s incredible passing. Trent left Liverpool to Real Madrid in the summer, and Salah is 34, on the decline, and is going to be missing for two months in early 2026 due to the African Cup of Nations. So, when Liverpool lost its two primary creative focal points, did the system successfully redistribute responsibility, or did the attacking output collapse toward the mean? The data suggests Liverpool did not replace its creative hubs structurally. Instead, attacking and midfield roles compressed, output flattened, and responsibility was unevenly absorbed by a small number of players.

This project draws on a combination of match-level, player-level, and contextual data to examine how Liverpool’s system has changed across the 2024–25 and 2025–26 Premier League seasons. The primary source for match-to-match and player statistics is FBref, which provides detailed Premier League data on minutes played, expected goals (xG), expected assists (xA), touches, progressive carries, final-third entries, key passes, defensive actions, and shots. FBref’s per-90 metrics allow for fair comparison across players with differing minutes, which is especially important given squad rotation, injuries, and new signings still being integrated into the team.

Defensive stability is assessed using a combination of pressures and tackles, which together reflect both the intensity of Liverpool’s defensive approach and its effectiveness. Declines in these measures can signal structural breakdowns in pressing or midfield protection rather than isolated defensive mistakes. By pairing defensive actions with attacking and creative metrics, the dataset enables a system-level evaluation of how Liverpool’s balance has changed across phases of play.

Watching Liverpool this season, the biggest on-field change is how much less aggressive the press has become under Arne Slot. That shift shows up clearly in the data. At the start of 2025, Liverpool’s dip in form lines up almost perfectly with a drop in pressing and defensive actions. When the team briefly finds form again, pressing intensity rises with it. The relationship is direct enough that it’s impossible to ignore. When Liverpool press aggressively, they control and win games.

What makes this more concerning is what happens after Liverpool win the league in April 2025. At the time, the sharp drop in form felt understandable. The season was effectively over, rotations increased, and the focus shifted to celebration. But the pressing data tells a different story. Even after results normalize and competitive matches resume, Liverpool’s press never returns to last season’s level. The decline isn’t temporary, it’s the new baseline.

This reframes Liverpool’s strong start to the 2025–26 season. Rather than representing a stable system under a new manager, the data suggests the team was overperforming relative to its underlying defensive intensity. As pressing continued to trend downward, the margins that sustained early results disappeared. Without that aggression, Liverpool struggled to regain control of matches, recover possession high up the pitch, and generate the same attacking momentum.

Last Season, Trent Alexander-Arnold was the focal point of Liverpool’s build-up, and losing him in the summer was always an issue Liverpool needed to fix. Liverpool signed Frimpong, who has been uncharacteristically injured since he signed, and Connor Bradley has also been in and out of the starting XI with injuries. This forced Curtis Jones and Dominik Szoboszlai to play at right back, which has worked somewhat to show Curtis Jones as a possible answer to the buildup question.

Losing Trent, has also had a knock-on effect on Liverpool’s attack. Salah who was the focal point of the team’s attack and had far and away the most touches last season, now doesn’t see as much of the ball and has drastically less output because of this. Wirtz looks as or even more involved than Salah, is also struggling to have significant output. The system lacks a center of gravity and because of that, at times, Liverpool’s attack looks toothless.

Looking at Salah’s numbers, it would be easy to frame this as a simple decline. His xG drops, his touches fall, and his overall output flattens compared to last season. But the visualization makes it clear that this isn’t just about aging or form. Salah’s role within the system has changed. He’s no longer the clear focal point everything runs through, and the data reflects that shift.

Rather than being isolated as a high-usage outlier, Salah’s involvement now sits much closer to the rest of the attacking group. This matters because Liverpool’s previous system was built around maximizing what he does best. When that structure fades without a clear replacement, his regression toward the mean becomes less of a failure and more of a symptom. The problem isn’t that Salah can’t carry the attack anymore, it’s that the system no longer elevates him above everyone else.

If Salah’s influence recedes, the natural expectation is that the players around him step up. Mac Allister is a good example of how that transition has played out. The data shows him taking on more responsibility across phases of play, but without a corresponding increase in effectiveness. His creative and defensive contributions both decline compared to last season, suggesting role inflation rather than role clarity.

Instead of excelling in a defined function, Mac Allister appears stretched thin, asked to contribute everywhere without anchoring any one part of the system. This reinforces the broader pattern emerging across the squad, that responsibility isn’t being redistributed cleanly. The supporting cast isn’t failing individually, but the system no longer puts them in positions where their strengths are consistently amplified.

The midfield is where the consequences of Liverpool’s structural shift become most visible. Across multiple metrics, most midfielders show declines in touches, defensive actions, and progressive involvement compared to last season. Rather than responsibility being shared or redistributed evenly, the overall trend points toward contraction. The system is asking less of the midfield as a collective, even as control of matches becomes harder to maintain.

Curtis Jones stands out as the exception. He is the only midfielder consistently increasing his progressive carries and key passes, stepping into the space left by the loss of Liverpool’s former creative hubs. But this isn’t evidence of a fully formed solution. Jones’s rise reads more like compensation than design. With limited minutes relative to others, his increased responsibility highlights a vacuum rather than a new equilibrium. The midfield isn’t evolving around him; it’s leaning on him.

QS Project 2

Quantified Self Project: My YouTube Habits

For this project, I wanted to know what patterns show up in my YouTube viewing between 2022 and 2025, and how do they connect to what was happening in my life during those years?

I originally thought I’d look at my full watch history since 2014, but Google Takeout only lets me export up to 2022. That turned out to be fine, because 2022–2025 covers a really interesting stretch of my life, graduating from college, working, being unemployed, going back to work, and now returning to school.


Google Takeout gave me my watch history as an HTML file. Each entry had the video title, the channel name, and the timestamp of when I watched it. To make it usable in Tableau, I wrote a JavaScript script to convert the HTML into a CSV. I also added two extra columns: one for the day of the week and one for the hour of the day. That gave me enough structure to explore when and what I was watching.


var divElement = document.getElementById(‘viz1763412244307’); var vizElement = divElement.getElementsByTagName(‘object’)[0]; vizElement.style.width=’100%’;vizElement.style.height=(divElement.offsetWidth*0.75)+’px’; var scriptElement = document.createElement(‘script’); scriptElement.src = ‘https://public.tableau.com/javascripts/api/viz_v1.js’; vizElement.parentNode.insertBefore(scriptElement, vizElement);

The first bar chart shows my most-watched channels with at least 100 views. Two of my top three are ASMR channels, which makes sense since I’ve used ASMR to fall asleep since 2020. The other is a news channel I’ve followed since 2014–2015, though I thought it would rank higher. A lot of my top ten are Twitch streamers posting highlights, which tracks with me starting to watch Twitch around 2021–2022.

Breaking things out by year shows Twitch content rising fast. In 2023, Twitch barely shows up. By 2025, it dominates. My music watching also shifted—2023 was mostly one channel, but by 2025 I spread across several.

The heatmap shows which days I watched most, split by year. 2023 was the peak—graduated, more free time, lots of YouTube. In 2024, views dropped as I worked more, with spikes on Sundays and Wednesdays, my days off. The pattern continues in 2025, though less strongly, probably because my job schedule changed.

The bubble chart digs deeper: day of week plus hour. ASMR clusters late at night and early morning, which makes sense for sleep. News channels show dense bubbles on the days they post, reflecting their publishing schedule.


Looking at my YouTube history from 2022–2025, I can see habits shift alongside my life. ASMR dominates late nights, Twitch takes over in 2025, and my music taste diversifies. My overall viewing frequency mirrors my free time—more in 2023, less in 2024, when I was working more.

There are limitations. Google Takeout only gave me the video title, channel, and timestamp. No categories or video length. I couldn’t get data before 2022, so I missed the full decade I wanted. Adding day of week and hour myself means small errors are possible.

There are also biases. The dataset only reflects what Google recorded. Offline viewing or other platforms aren’t included. Heavy ASMR use skews the data toward certain channels and times.

If I did this again, I’d want to combine YouTube data with Twitch or Spotify exports to compare across platforms. I’d add video categories like news, entertainment, or music for richer analysis. And if Google ever allows longer exports, I’d expand the timeline to see how habits evolve over a decade.


Data Viz Project 1

The question I started with to explore the dataset was “Who is 311 really for?”. In a perfect world, response time would be quick and equitable around the city, but I wanted to see how complaints are distributed across New York City, how quickly they are closed, and whether patterns suggest differences by place or problem type.

The dataset comes from the NYC Open Data Portal’s 311 Service Requests and covers January–December 2024. Each request includes complaint type, borough, latitude/longitude, the responsible agency, status (open/closed), resolution description, and created/closed dates. From these fields, I focused on two core measures: where complaints are filed and how long they take to close. Resolution time is calculated as the difference between the created and closed dates. It’s important to name the limits. Some requests never show a closed date, which makes resolution time incomplete. Resolution descriptions can be vague or inconsistent across agencies, so “closed” doesn’t always mean “resolved to the resident’s satisfaction.” And there’s a deeper bias: 311 reflects who reports and how THEY report, not necessarily where conditions are objectively worst. Communities with less access to technology, language supports, or trust in city systems may be underrepresented. Part of my original question was to see whether my borough, the Bronx, was the dirtiest, and I can’t give a conclusive answer to that, but according to this dataset, it isn’t.

The first visualization displays a map of zip codes, shaded to indicate which ones have the most requests to the Sanitation department. Darker shades indicate the zip code has a higher amount of calls, and according to this map, Brooklyn and Queens have the highest volume of calls.

These visualizations look into resolution times in certain neighborhoods. The first looks at a street level, for example, the visualization is looking at the Grad Center, and the second looks at resolution times at a broader level, zooming out to Zip Codes. The Neighborhood map looks at resolution times in days, and the colors are set to Red, Green, and Yellow, where Green has short resolution times and Red has longer wait times, with Yellow in between them. This map, combined with the first, shows an interesting relationship between the number of calls and resolution times because Queens had the highest volume of cases, but as a borough, the resolution times are short in relation to the rest of the city. Also, Staten Island, in relation to the rest of the city, has relatively high resolution times similar to the Bronx.

The bar charts make these borough‑level differences even clearer.

The first chart shows the total number of sanitation‑related complaints by borough. Brooklyn and Queens stand out with the highest counts, while Staten Island has the fewest. The second chart looks at average resolution times by borough. Here, the Bronx and Staten Island show longer waits compared to Manhattan, Brooklyn, and Queens. Together, these visuals highlight an important contrast: the boroughs with the heaviest complaint volume aren’t always the ones with the slowest response. Queens, for example, generates a large share of complaints but still sees relatively quick resolution times. By contrast, Staten Island has fewer complaints overall but longer waits, suggesting that volume alone doesn’t explain service speed.

So, who is 311 really for? Looking at 2024, the answer is complicated. The system is used across the city, but not experienced equally. Some boroughs see quick closures, while others wait longer, regardless of how many complaints they file. That unevenness matters because it shapes how communities trust city services. I’d like to look at this data connected to population in the future, to see whether disparities reflect usage, need, or deeper inequities.

Workbook link: https://public.tableau.com/views/DSNY311Requests/Sheet3?:language=en-US&:sid=&:redirect=auth&:display_count=n&:origin=viz_share_link