I noticed that UVA lists @UVASafety and @UVAPolice on X as its only social-media channel for emergency notifications. I understand that social media is only one part of UVA’s multi-layered, redundant system, but I would like to ask why that layer is exclusive to X.

Access#

My first concern is access. Without an X account, a visitor to @UVASafety’s account page can see only a short list of about five recent posts. Visitors must log in to open an individual post, view its replies, or see more recent posts. By contrast, most other social platforms provide public web access; see Meta, Bluesky, and Mastodon, for instance. Why does UVA use such a closed platform as its only social-media option?

Reach#

Maybe the office’s rationale for choosing X is reach. However, X is no longer dominant: Meta reported 500 million monthly users for Threads in June 2026, while Similarweb estimated 141.5 million daily mobile users for Threads and 125 million for X that January.

X is also not clearly dominant in academic community, which is especially relevant to UVA. Despite Bluesky’s much smaller overall user base, Altmetric found that it hosted more original posts linked to new research than X on most days in March 2025, suggesting that headline user counts understate Bluesky’s reach within academic communities. A large-scale study of more than 2.6 million Bluesky posts later found substantially higher interaction than previously reported for X, while Techdirt reported that Bluesky sent it far more referral traffic than X in 2025.

Neither global user counts nor these narrower comparisons establish local usage and reach. However, visible engagement with UVA’s emergency-alert posts on X appears limited. For instance, three @UVASafety updates posted later during UVA’s August 14 tornado warning (5:52 p.m., 6:02 p.m., and 6:37 p.m.) received little engagement (3–10 likes, 0–1 replies, and 1–2 reposts), comparable to or lower than that of two Bluesky posts about the same storm by local meteorologist Travis Koshko (6:36 p.m. and 7:37 p.m.). Four recent @UVAPolice emergency-alert posts—the July 31 all-clear and the August 7 initial alert, gas-leak update, and all-clear—showed similarly little visible engagement: 0–1 likes, no replies, and 0–3 reposts. Of course, this is not a rigorous comparison, but these examples suggest that the official accounts’ reach and visible engagement on X is quite limited and not substantially greater than engagement with similar posts on much smaller platforms.

X’s suppression of outbound links#

Another access concern is that X has a documented history of reducing exposure to outbound links, as well as selectively delaying access to links from rival platforms and news organizations. An alert on X that links to further instructions or news articles may therefore receive less visibility and fail to reach some users. This suppression of external links is not universal: for instance, Bluesky says external links are not penalized.

Misleading identity signals#

There is also an information-integrity and trust problem. On X, a blue check means that an account has an active Premium subscription or is affiliated with a paying organization; it does not mean that the account’s identity has been verified. In 2025, the European Commission found the design deceptive because X did not meaningfully verify who controlled an account, making it difficult to judge the authenticity of accounts and content. The risk is not hypothetical: paid blue-check accounts have impersonated Eli Lilly, LeBron James and Nintendo, and customer-service agents in refund scams. An emergency-alert account requires strong, immediate trust in the sender’s identity, making X’s ambiguous blue check particularly ill-suited to the task.

By contrast, Meta Verified requires a government-issued ID for Instagram creators, and Threads carries over Instagram verification. Bluesky and Mastodon offer identity signals tied to trusted organizations or control of a website.

Grok and information integrity#

X’s deep integration of Grok into the platform raises several concerns. First, in 2025, xAI acknowledged that an “unauthorized modification” caused Grok to inject a specific political response into unrelated public conversations. Second, X has placed a Grok button directly on posts and made @grok available in public replies as an instant fact-checker, inviting users to treat it as a source of “facts.” During the 2025 Los Angeles protests, Grok misidentified real photographs as recycled from Afghanistan, driving hours of debate before the bot corrected itself.

These incidents were not isolated. In February 2025, xAI acknowledged a hidden instruction telling Grok to ignore sources that criticize certain people. In July 2025, xAI removed a public system instruction telling Grok not to shy away from “politically incorrect” claims after it posted antisemitic tropes and praised Hitler. Together, these incidents show that hidden system instructions have altered Grok’s public answers. An AI system so deeply integrated into X may itself spread misinformation, making the platform a problematic choice for emergency alerts.

X’s CSAM and safety problems#

In 2026, Canada’s Privacy Commissioner reported that Grok had been used to create abusive imagery, including child sexual abuse material (CSAM) and non-consensual intimate imagery. X had removed 126 reported posts, but the Commissioner found that X and xAI had not demonstrated safeguards. More recently, xAI has been accused of training Grok on CSAM.

X’s safety problems extend beyond Grok. The company reduced its safety-engineering staff by 80% after the acquisition. Harassment is not unique to X, but it remains widespread there. In a 2024 Pew survey, 73% of U.S. X users said harassment was a problem on the platform, including 32% who called it a major problem. In a Nature survey, 65% of the 201 researchers who used Twitter to discuss COVID-19 said that they had experienced trolling or personal attacks there at least sometimes.

The consequences can extend beyond the platform. George Washington University researcher Rebekah Tromble received death and rape threats after participating in research on Twitter discourse; she moved her office and had police patrols outside her home. In 2024, Australia’s eSafety Commissioner Julie Inman Grant said that Musk’s attacks on X resulted in death threats against her and doxxing of her family. In a 2026 first-person account, Eugene Vinitsky described receiving antisemitic slurs and death threats that referenced his workplace after his posts went viral on X. These cases show how harassment on X can follow researchers and public officials into their workplaces and homes.

Although this may not be unique to X, it is generally understood that other platforms enforce safety more rigorously and the problem is most severe on X.

Cost#

UVA’s use of X also raises a separate question about cost. X says a blue check can come from Premium or Premium+, whose published U.S. web prices are $8 and $40 per month, respectively. Its separate Premium Organizations plan costs $1,000 per month plus $50 per affiliated account. X also charges for API access under pay-per-use pricing. By contrast, Bluesky’s AT Protocol and Mastodon’s ActivityPub protocol are open protocols that third parties can implement without paying a platform-level API fee. Their authenticated APIs let authorized applications publish Bluesky posts and publish Mastodon statuses.

Question for UVA#

Given these concerns, why does UVA single out X as its only social-media alert channel rather than other free, more accessible, and safer platforms that may offer comparable or greater reach?

Sincerely,
YY Ahn
School of Data Science, University of Virginia

The geometry does not scale#

NACTO’s peak-capacity estimates show how many people different modes can move through the same width of urban street under normal peak conditions:

Note that this is about capacity; a disconnected bike lane will not attract thousands of riders, and a bus stuck in car traffic will not reach transitway capacity. But the order-of-magnitude difference is the point: private cars require way more street space per person. Whenever we turn a car lane into a connected transit, bike lane, or sidewalk that can be used well, we double, triple, quadruple the capacity of the street without widening it.

More lanes create more driving#

Road expansion can reduce delay temporarily, but the easier trip changes people’s behavior. They change routes, departure times, destinations, modes, and how often they travel. Moreover, wide, multi-lane roads make them drastically more hostile to pedestrians and cyclists, further suppressing alternative and way more space efficient modes. Over time, households and businesses also change locations. This is Induced demand.

Gilles Duranton and Matthew Turner found that vehicle-kilometers traveled rose roughly in proportion to interstate lane-kilometers across US cities (2011). The added traffic came from more driving by current residents, more commercial traffic, and migration. The authors also found no evidence that merely providing public transit reduced total driving. Alternatives must be actually useful, and cities must stop continually making driving cheaper and easier at the same time.

Fewer car trips, not worse access#

The scalable goal is not to prevent people from reaching jobs, schools, shops, or friends. It is to provide the same access with fewer car trips. Walking, biking, and transit must be safe, connected, reliable, and competitive enough that people can choose them without accepting a major penalty.

Congestion pricing, parking reform, compact land use, telework, and better non-driving infrastructure use different mechanisms, but they share the same basic logic: fewer cars compete for limited road space at the busiest times. If no practical alternative exists, reducing car capacity merely makes travel harder. If good alternatives exist, some trips shift modes, times, destinations, or disappear from the road network altogether.

Not everyone needs to stop driving. Every person who can make a trip without a car frees a disproportionate amount of space for people who still need to drive. Good walking, biking, and transit can therefore make driving better too.

Traffic can disappear#

Road-space reallocation does not mean that every displaced car moves to the next street. A review of more than seventy cases across eleven countries found that traffic problems were usually less severe than predicted. The median share of traffic formerly using the treated road or area that could not be found in the surrounding network was 10.6%; the mean was 21.9%. People adapted in many ways rather than all sitting in the same queue somewhere else. This is Traffic evaporation.

Traffic demand is not a fixed quantity waiting to be accommodated. As Traffic projection explains, street design helps create the traffic that models later treat as inevitable.

Bottom line#

You cannot solve urban traffic by fitting more cars into a city. The durable solution is to make fewer trips require one by making other modes genuinely viable.

Short version#

tl;dr: Don’t try to generate the whole poster. Find or generate a background image first, and then create a poster using HTML and CSS. You can use AI agent in each step and don’t have to know HTML and CSS—just specify what you want. However, also remember that this workflow can very easily create a low-quality AI slop.

Image generation#

Supply the image-capable model the following information:

1. Intended dimensions
2. Theme
3. A (rough) color palette
4. The information the design must contain and what must not appear (e.g., The exact words and existing logos that should be included)
5. A background image (either a copyright free image or a generated image) that fits the theme and color scheme of the poster (rather than a finished poster).

Example prompt:

Create a text-free background for an 8.5 × 11-inch portrait poster.

Theme: [subject and visual metaphor]
Rough palette: [dominant colors and accent colors]

Leave quiet space for:
- [headline]
- [dates, venue, and other necessary information]
- [logos and QR code]

Composition: [where the visual focus and quiet space should sit]
Mood: [a few adjectives]

Do not include text, letters, numbers, logos, QR codes, or watermarks.
Avoid: [visual clichés or unwanted elements]

Keep the information out of the background#

Use a text-free background image. Put the headline, schedule, logos, and QR code in HTML so that factual corrections do not require regenerating artwork. A sample structure:

poster/
├── index.html
├── build.sh
├── assets/
│   ├── background.png
│   ├── organization-logo.png
│   └── qr-info.svg
├── poster.png
└── poster.pdf

Record the image prompt, asset sources, and logo-use notes in a README.

Fix the canvas#

.poster {
  width: 816px;
  height: 1056px;
}

@page {
  size: letter portrait;
  margin: 0;
}

@media print {
  html, body {
    width: 8.5in;
    height: 11in;
  }
}

Set the physical format first. Otherwise every spacing and alignment decision is tuned to the wrong shape.

Build both outputs#

Put the rendering commands in a build script (a similar script can be generated by AI agent):

#!/usr/bin/env bash
set -euo pipefail

poster_dir=$(CDPATH= cd -- "$(dirname -- "$0")" && pwd)
chrome_bin="/Applications/Google Chrome.app/Contents/MacOS/Google Chrome"
poster_url="file://$poster_dir/index.html"

"$chrome_bin" \
  --headless=new \
  --hide-scrollbars \
  --window-size=816,1056 \
  --screenshot="$poster_dir/poster.png" \
  "$poster_url"

"$chrome_bin" \
  --headless=new \
  --no-pdf-header-footer \
  --print-to-pdf="$poster_dir/poster.pdf" \
  "$poster_url"

Save it as poster/build.sh, then run:

chmod +x poster/build.sh
./poster/build.sh

This minimal version assumes Google Chrome on macOS. The @page rule in the HTML controls the PDF’s physical page size; --window-size controls the PNG dimensions.

A useful build script asks headless Chrome for a PNG and PDF, checks the PDF page size, renders the PDF back to an image, and decodes the QR code from both artifacts. It can also normalize PDF metadata with qpdf so an unchanged source produces an unchanged PDF.

To replace the QR destination and rebuild everything:

./poster/build.sh --qr 'https://FINAL-URL'

Do not stop after checking the source SVG. A QR code can be valid before layout and unreadable after scaling or PDF rendering. Decode what people will actually receive.

Set the layout system, then iterate#

Use CSS Grid or Flexbox for the overall structure. Define the main rows and columns, then group related elements such as sponsor logos and a QR code. Shared rules should control heights, gaps, and baselines; reserve per-element offsets for small optical corrections.

Once that system is in place, the rest is iteration. Render the poster and adjust alignment. Then tune element sizes and colors. Finish with small spacing and decorative tweaks, rebuilding after each pass.

It may be the leaf blower! In a 2011 laboratory test, the two-stroke blower emitted nearly 300 times as much non-methane hydrocarbon per minute as an F-150 truck.

Small two-stroke leaf-blower engines can emit large amounts of unburned hydrocarbons and carbon monoxide, while resuspending dust into the air.

This rabbit hole began with welcome local news. On August 24, Charlottesville announced that the Downtown Mall custodial crew now uses electric leaf blowers exclusively. Staff report comparable power and performance, much less noise, and almost no remaining blower-noise complaints.

The comparison#

In that test, Edmunds compared a 50.8 cc two-stroke Echo backpack blower and a four-stroke Ryobi blower with a 2011 Ford F-150 Raptor. The blowers followed an adapted version of the same 31-minute federal driving cycle. Emissions are weighted grams per minute:

The two-stroke blower emitted nearly 300 times as much non-methane hydrocarbon and 23 times as much carbon monoxide per minute as the truck. Edmunds calculated that about 30 minutes with that blower produced as much hydrocarbon pollution as driving the Raptor 3,887 miles.

This is one test of specific machines, not a universal conversion between leaf blowing and driving. It compares particular pollutants, not total environmental harm: the Raptor emitted more carbon dioxide, while the blowers were dramatically worse for local hydrocarbon and carbon-monoxide pollution.

Why the two-stroke engine is so dirty#

A California Air Resources Board report explains:

Typical two-stroke designs feed more of the fuel/oil mixture than is necessary into the combustion chamber. Through a process known as scavenging, the incoming fuel enters the combustion chamber as the exhaust is leaving. This timing overlap of intake and exhaust port opening can result in as much as 30% of the fuel/oil mixture being exhausted unburned. Thus, exhaust emissions consist of both unburned fuel and products of incomplete combustion.

The blower adds a separate problem: high-velocity air lifts dust from the ground. CARB found the amount difficult to measure reliably, so estimates of this resuspended particulate pollution should be treated cautiously.

It may not just be the leaf blowers#

Leaf blowers belong to the broader category of small off-road engines, along with lawn mowers, chainsaws, pressure washers, air compressors, and portable generators. In its 2021 rulemaking baseline, CARB estimated that this category emitted more nitrogen oxides and reactive organic gases in California than light-duty passenger cars. Without further regulation, it projected the category would emit 1.8 times as much of those pollutants as passenger cars in summer 2031. That comparison covers the entire category, not leaf blowers alone.

Towards lower pollution in general, especially for children#

Children are not small adults. A Paris-region health assessment notes that they breathe more air relative to their body weight and that their lungs are still developing, making avoidable air pollution especially consequential.

The policy evidence makes that risk concrete. A London’s natural experiment followed 3,414 primary-school children during the first four years of the Ultra Low Emission Zone (ULEZ). As residential nitrogen-dioxide exposure fell faster in London than in Luton, children’s lung-function growth was also faster; the share with impaired lung function fell from 14% to 9% in London, compared with 9% to 7% in Luton. The study cannot isolate the ULEZ from every concurrent change, but it connects cleaner neighborhood air with children’s developing lungs.

If we remove dirty vehicle exhaust around schools, people, especially children, become much healthier. Why do we still run these highly polluting engines for purely cosmetic yard cleanup, especially when electric alternatives exist?

What the city can do#

The Downtown Mall transition shows that electric blowers can handle demanding daily municipal work. Now Charlottesville can make that success the default:

• publish an inventory and phase-out schedule for the city’s remaining gas-powered lawn equipment;
• make electric equipment the default for new city purchases and landscaping contracts, with narrow documented exceptions;
• prioritize schools, parks, sidewalks, and dense residential areas, where people are close to the exhaust;
• help small landscaping businesses switch through rebates, grants, or shared charging infrastructure.

What we can do#

• Leave leaves where practical; otherwise, rake or use electric.
• Ask hired landscapers to use electric equipment.
• Thank the city for making the Downtown Mall quieter and cleaner.
• Ask the Office of Sustainability and City Council for a timeline to extend the transition across city operations and contracts.

Sources#

• Electric Leaf Blowers Bring Cleaner, Quieter Maintenance to Downtown Charlottesville, City of Charlottesville
• Emissions Test: Car vs. Truck vs. Leaf Blower, Edmunds
• 2021 Small Off-Road Engine Rulemaking Staff Report, California Air Resources Board
• A Report to the California Legislature on the Potential Health and Environmental Impacts of Leaf Blowers, California Air Resources Board
• London’s Ultra Low Emission Zone and children’s lung-function growth, CHILL study
• Air quality around Parisian schools, Airparif
• Expected health benefits for children from a Paris-region low-emission zone, Santé publique France

Rue Ferdinand-Flocon in Paris after its conversion into a planted “school street.” Photo: Guallendra, CC0.

Suppose the following question:

“On this school route, would you prefer the street without any traffic calming, or a calmer design that slows cars and makes walking and biking safer, with roughly 50% fewer children killed or seriously injured and generally modest effects on travel time?”

I doubt the answer would look like a minority preference.

(See Traffic calming works for detailed evidence)

Status quo bias—mode share is not a referendum#

Let’s start with the number that makes driving look like an overwhelming preference. In the 2024 American Community Survey, 69.2% of US workers drove alone and another 9.2% carpooled. 13.3% worked from home, so cars, trucks, and vans carried about 90% of workers who physically commuted. New York City is one conspicuous exception (2024 Citywide Mobility Survey) where sustainable mode outnumbers car. Does this mean Americans vastly prefer driving?

Not necessarily. The ACS records what workers did under the transportation and housing options available to them, not what they would choose from equally safe and convenient alternatives. Nor does the evidence suggest that people simply love commuting by car. An analysis of more than 22,000 US travel episodes found a modest negative association between trip duration and mood, especially for drivers; the authors conclude that “Long commutes significantly degrade the mood” of drivers and bus riders. A survey of 828 workers in Portland concluded that “people who bike and walk to work are happier with their commutes.” And in a 2026 survey of 2,000 adults in the fifty largest US metropolitan areas, nearly two-thirds said they would pay more to live in a walkable community, and a majority preferred a smaller lot in a walkable neighborhood over a larger homesite requiring more driving.

Two US surveys found that 56% of adults in the Portland region and 51% across the 50 largest metropolitan areas were “interested but concerned” about biking (Dill and McNeil 2013; Dill and McNeil 2016). Only 31% in Portland and 37% nationally were in the “No Way, No How” group. The rest, a clear majority, was at least open to biking. Given the status-quo bias, this is probably an underestimation of how likely people may choose sustainable modes when they are viable and safe. FHWA’s Bikeway Selection Guide therefore treats this group as the typical urban design user that the design should target.

Not Just Bikes’ “I am not a Cyclist” makes a strong, general point: people tends to choose the mode that is most convenient and efficient. There are very few “cyclists cyclists”, people ride bikes when it’s a safe and efficient mode of transportation.

That is also how I think about my own trips. Although I usually ride an e-bike, I am not married to it. I drive, bike, walk, or ride the bus depending on which option works best. If a city has excellent transit, people tend to use it. If a city is an extremely car-dependent hell, people have to drive. Treating people as fixed “drivers” or “cyclists” mistakes a practical choice for an identity.

The dominance of driving also creates an empathy problem. Because most people experience most streets from inside a car, it is difficult—even with good intentions—to imagine what the same street feels like to a child crossing it, a parent pushing a stroller, or someone biking while a truck passes inches away. A driver can easily picture the inconvenience of a speed hump, a lane reduction, or a slower turn; the danger and stress experienced outside the car are largely invisible. Public opinion can therefore overweight small, familiar costs to driving while underweighting risks that most respondents rarely experience.

These studies do not prove that driving is a minority preference. But they do suggest that a silent majority might choose not to drive if reliable, safe alternatives existed. Because it is difficult to imagine a completely different built environment, the surveys may also underestimate potential mode change.

The current mode share is not a neutral measure of preference. A person choosing between driving where every infrastructure is built for driving and biking where infrastructure is hostile to them is not choosing between equally competitive options. When one mode is safe and convenient while the alternatives are dangerous, the resulting behavior is not one’s voluntary choice but an engineered one.

We need to raise the bar for safe streets for non-drivers#

The double standard is strong here. We don’t poll residents on whether pavement structure, drainage, grading, or guardrails should meet minimum engineering standards for motor vehicles. Those protections for cars are treated as minimal necessities. But when the subject is a safer crossing, a protected bike lane, or traffic calming, cities tend to poll residents, commission studies, and then use opposition as a reason to delay or weaken it. Safety for driving is treated as a must-do baseline; safety for walking and biking is treated as an optional amenity that must win a popularity contest.

Public input is important. It should help set priorities and resolve real tradeoffs. But it, especially when the vast majority of people cannot “see” the issues, should not determine the minimum safety bar for sustainable modes.

The federal Safe System approach considers death and serious injury unacceptable and calls for proactive design that anticipates human mistakes. FHWA’s Traffic Calming ePrimer documents 22 measures, and its Bikeway Selection Guide helps practitioners select facilities for different roadway conditions. NACTO’s design guidance provides more complete designs for safe urban streets and bikeways. Yet incorporating even simple treatments into streets in many US cities remains a constant struggle.

Bike infrastructure is also not a niche protection that benefits only people biking. A 13-year study of 12 large US cities, covering more than 17,000 fatalities and 77,000 severe injuries, found that better safety outcomes for all road users were associated with more bike facilities—especially protected and separated facilities—not simply with having more people biking. This result directly challenges the idea that bike infrastructure is solely for “cyclists” who constitute a very small minority of people. Well-designed, safe streets reduce the risk for everyone.

The benefits are too great to delay#

Safer streets create massive societal benefits. A connected, low-stress walking and biking network can:

• prevent deaths and serious injuries among people walking, biking, and driving (Traffic calming works);
• reduce the economic losses from dangerous streets—medical care, emergency response, lost productivity, property damage, and congestion (Traffic calming is cheap);
• give people practical alternatives to driving, because cars are space-inefficient and shifting trips to walking, biking, and transit reduces traffic (Traffic evaporation) and transportation emissions (EPA);
• make physical activity part of daily life and improve health;
• reduce the household cost of forced car ownership (Cost of car ownership);
• support the local economy: a review of 23 US and Canadian studies found a consensus of positive or neutral benefits of safe street infrastructure on nearby retail and food-service businesses, and I wrote why those findings can still apply to small cities; and
• help children breathe cleaner air, travel independently, and build physical activity into the trip to school.

The case of Charlottesville#

These are also Charlottesville’s adopted goals. The 2021 Comprehensive Plan calls for a connected network of safe and convenient walking, biking, and transit accommodations, safe walking and biking routes to every public school, and walkable, bikeable, transit-accessible neighborhoods. It explicitly ties walking and biking infrastructure to the city’s climate goals. The Climate Action Plan commits the city to cutting greenhouse gas emissions 45% below 2011 levels by 2030 and reaching carbon neutrality by 2050; its first transportation strategy is to increase travel by walking, biking, and transit through connected networks. Safe streets are therefore not a competing amenity. They are necessary to implement the city’s own transportation, land use, health, and climate commitments.

Traffic policy also shapes air pollution and physical activity. A prospective parallel cohort study of 3,414 children in London and Luton found that, during the four years after London’s Ultra Low Emission Zone took effect, residential nitrogen dioxide exposure fell faster in London and children’s lung function grew faster. London children had started with lower average lung function; after four years, they had caught up with the Luton comparison group. This natural experiment does not isolate the ULEZ from every concurrent difference, but it provides unusually direct evidence connecting a traffic emissions policy, cleaner air, and children’s lung development.

Walking or biking to school can also build physical activity into daily life. A systematic review of 68 studies found moderate-quality evidence that active school travelers were more physically active and that biking to school was associated with better cardiovascular fitness. One cross-sectional study of 1,700 Spanish adolescents found higher cognitive performance scores among girls who actively commuted, but not among boys. The broader evidence for learning is less clear: a systematic review and meta-analysis found no statistically significant overall association with cognitive performance or academic achievement. The health case is stronger than the learning claim.

The cost of the status quo: why only prioritizing maintenance can be bad#

Public Works and Traffic Engineering have both explained that limited staff capacity requires them to prioritize maintenance. But maintenance is not neutral. Repaving a dangerous street and restoring the same geometry preserves the danger, often until the next maintenance cycle. If all available capacity goes toward reproducing the existing network, a safer network, and its massive benefits, will remain perpetually next in line and never happen.

Also note that maintenance and improvement is not mutually exclusive. Maintenance projects are often the cheapest and fastest opportunity to improve safety. FHWA explicitly calls installing bicycle facilities during resurfacing an efficient and cost-effective way to build connected networks, using tools such as lane narrowing, road diets, roadway reconfiguration, parking changes, and new striping. FHWA reported that a four-to-three-lane road diet with bike lanes cost about $36,000 per mile when bundled with resurfacing, compared with $100,000 as a standalone project. These opportunities are time-sensitive: once a street is repaved and the old markings are restored, the low-cost window closes.

So why do so few resurfaced streets come back calmer and safer? The 2021 Comprehensive Plan says the city should continue including bicycle, pedestrian, accessibility, and transit improvements in all major road projects, including paving; Rose Hill Drive shows that the city can use scheduled repaving to reconsider striping and safety. Yet the city’s current public process appears to separate pavement maintenance from new traffic-infrastructure changes, and I could not find a published rule requiring every resurfacing project to undergo a Complete Streets or traffic-calming redesign. Does every street in the paving plan undergo such a redesign before its old geometry is restored? If not, why not? Safer striping, daylighting, crossings, and speed management should be the default scope of routine road work, not separate projects residents must repeatedly request.

Some improvements do require dedicated staff, engineering, and capital. That is not a reason to defer them indefinitely; it is evidence that current capacity is inadequate. If maintaining unsafe designs consumes all available capacity, the city must add capacity or change how maintenance is scoped. Otherwise, “prioritizing maintenance” becomes a policy of making today’s danger permanent.

Charlottesville recorded 16 traffic deaths and 2,409 injuries from 2019–2025, according to the annual Virginia Traffic Crash Facts reports. Crashes in 2024 alone produced an estimated $23–$30 million in economic losses.

Meanwhile, Hoboken has gone nine consecutive years without a traffic death. That comparison does not identify the effect of any single treatment, but it shows that Charlottesville’s outcomes are not inevitable.

Name the constraint, then fix it#

The central question is straightforward: what specifically prevents Charlottesville from applying these well-established solutions? Is the constraint staff capacity, funding, law, engineering, or something else? What would be required to overcome it?

If the constraint is staff capacity or funding, the city should make the case for the resources it needs. I raised the same problem when the bicycle-infrastructure budget was slated to fall to $100,000 per year: inability to execute is a capacity problem to fix, not a justification for inaction. Is the department asking Council for the necessary resources? If so, residents can help advocate for them. If not, why not?

Workflow#

• Workflow

Tools#

• Copilot

GitHub Pages#

• Creating and Hosting a Personal Site on GitHub
• jekyll: github pages

GitHub data#

• Blythe2019massive
• https://worldofcode.org
• The Promises and Perils of Mining GitHub
• Network Structure of Social Coding in GitHub
• Understanding the group dynamics and success of teams

@article{rodriguez2019optimal,
    author = {Nathaniel Rodriguez and Eduardo Izquierdo and Yong-Yeol Ahn},
    title = {Optimal modularity and memory capacity of neural reservoirs},
    journal = {Network Neuroscience},
    volume = {3},
    number = {2},
    pages = {551--566},
    doi = {10.1162/netn_a_00082},
    archivePrefix = {arXiv},
    eprint = {1706.06511},
    primaryClass = {cs.NE},
    year = {2019},
}

Shows that modular structure controls the dynamics and memory performance of threshold-neuron reservoirs. Memory peaks at an intermediate modularity that balances local cohesion with global connectivity.

Reservoir computing, Optimal modularity, Neural graph architecture and function

@article{lu2020cocontributorship,
    author = {Chao Lu and Yingyi Zhang and Yong-Yeol Ahn and Ying Ding and Chenwei Zhang and Dandan Ma},
    title = {Co-contributorship network and division of labor in individual scientific collaborations},
    journal = {Journal of the Association for Information Science and Technology},
    volume = {71},
    number = {10},
    pages = {1162--1178},
    doi = {10.1002/asi.24321},
    archivePrefix = {arXiv},
    eprint = {1911.10745},
    primaryClass = {cs.SI},
    year = {2020},
}

Builds author–task and co-contributorship networks from 138,787 PLOS articles to study division of labor within papers. It distinguishes specialists, team players, and versatile contributors and relates those patterns to common research tasks and author roles.

Scientific collaboration, Team science, Division of labor

@article{perry2022new,
    author = {Brea L. Perry and Meltem Odabaş and Kai-Cheng Yang and Byungkyu Lee and Patrick Kaminski and Brian Aronson and Yong-Yeol Ahn and Carrie B. Oser and Patricia R. Freeman and Jeffrey C. Talbert},
    title = {New means, new measures: assessing prescription drug-seeking indicators over 10 years of the opioid epidemic},
    journal = {Addiction},
    volume = {117},
    number = {1},
    pages = {195--204},
    doi = {10.1111/add.15635},
    year = {2022},
}

Compares traditional and network-based indicators of prescription drug-seeking using US claims from 2009–2018. Traditional binary indicators became less predictive over time, while continuous measures and PageRank-based measures retained more predictive value.

Opioid epidemic, Drug seeking behavior, Network centrality

@book{lehmann2018complex,
    editor = {Sune Lehmann and Yong-Yeol Ahn},
    title = {Complex Spreading Phenomena in Social Systems: Influence and Contagion in Real-World Social Networks},
    series = {Computational Social Sciences},
    publisher = {Springer International Publishing},
    doi = {10.1007/978-3-319-77332-2},
    isbn = {9783319773315},
    year = {2018},
}

An edited volume on spreading dynamics in social networks, collecting 19 chapters across five parts: introductory concepts, models and theory, observational studies, controlled experiments, and the editors’ concluding reflections.

Chapters with pages in this wiki include Hui et al. on detecting viral memes from diffusion patterns, Nematzadeh et al. on optimal modularity in Complex contagion, and the editors’ closing reflections.

@book{bagrow2024working,
    author = {James Bagrow and Yong-Yeol Ahn},
    title = {Working with Network Data: A Data Science Perspective},
    publisher = {Cambridge University Press},
    doi = {10.1017/9781009212601},
    isbn = {9781009212601},
    year = {2024},
}

A textbook on the practical side of Network science—working with network data—written for data scientists. The book follows the life cycle of a network study, from conceptualization through statistical modeling and analysis, and is organized in three parts: Background (network science fundamentals and data ethics), Applications and Tools (gathering data through machine learning), and Fundamentals (mathematical and computational foundations).

The network cards proposal for concise, readable summaries of network data comes from the same authors.

@article{ahn2006celegans,
    author = {Yong-Yeol Ahn and Hawoong Jeong and Beom Jun Kim},
    title = {Wiring cost in the organization of a biological neuronal network},
    journal = {Physica A},
    volume = {367},
    pages = {531--537},
    doi = {10.1016/j.physa.2005.12.013},
    eprint = {0505009},
    archivePrefix = {arXiv:q-bio},
    year = {2006},
}

Studies wiring cost as an organizing constraint on the C Connectome, treating neuronal placement as a Network optimization problem.

@article{ahn2006twospecies,
    author = {Yong-Yeol Ahn and Hawoong Jeong and Naoki Masuda and Jae Dong Noh},
    title = {Epidemic dynamics of two species of interacting particles on scale-free networks},
    journal = {Physical Review E},
    volume = {74},
    number = {6},
    pages = {066113},
    doi = {10.1103/PhysRevE.74.066113},
    eprint = {0608461},
    archivePrefix = {arXiv},
    year = {2006},
}

Analyzes a two-species Epidemic model in which one species induces the other while the second suppresses the first, producing a degree-dependent Phase transition on Scale free networks.

@inproceedings{ahn2007cyworld,
    author = {Yong-Yeol Ahn and Seungyeop Han and Haewoon Kwak and Sue Moon and Hawoong Jeong},
    title = {Analysis of Topological Characteristics of Huge Online Social Networking Services},
    booktitle = {Proceedings of the 16th International World Wide Web Conference},
    pages = {835--844},
    doi = {10.1145/1242572.1242685},
    year = {2007},
}

Compares the topology of Cyworld, MySpace, and Orkut, and evaluates snowball sampling against complete Cyworld data as a test of Network sampling in Online social networks.

@article{ahn2013flavor,
    author = {Yong-Yeol Ahn and Sebastian E. Ahnert},
    title = {The Flavor Network},
    journal = {Leonardo},
    volume = {46},
    pages = {272--273},
    doi = {10.1162/LEON_a_00569},
    year = {2013},
}

Flavor-compound and recipe networks reveal contrasting ingredient-pairing patterns in Western and Eastern cuisines, presenting the Flavor network as both a Computational gastronomy result and a piece of Network art.

@article{ahn2014drugtarget,
    author = {Yong-Yeol Ahn and Deok-Sun Lee and Henry Burd and William Blank and Vinayak Kapatral},
    title = {Metabolic Network Analysis-Based Identification of Antimicrobial Drug Targets in Category A Bioterrorism Agents},
    journal = {PLOS ONE},
    volume = {9},
    number = {1},
    pages = {e85195},
    doi = {10.1371/journal.pone.0085195},
    year = {2014},
}

Flux balance analysis across multiple bacterial strains identifies shared enzymes in the Metabolic network as candidate antimicrobial targets for three Category A bioterrorism agents, an approach to Drug discovery.

@article{bae2016landscape,
    author = {Arram Bae and Doheum Park and Juyong Park and Yong-Yeol Ahn},
    title = {Network Landscape of Western Classical Music},
    journal = {Leonardo},
    volume = {49},
    number = {5},
    pages = {448},
    doi = {10.1162/LEON_a_01268},
    year = {2016},
}

A network view of collaboration and influence among figures in western classical Music, applying Scientific collaboration network analysis and presented as Network art.

@article{bae2016multi,
    author = {Arram Bae and Doheum Park and Yong-Yeol Ahn and Juyong Park},
    title = {The Multi-Scale Network Landscape of Collaboration},
    journal = {PLOS ONE},
    volume = {11},
    number = {3},
    pages = {e0151784},
    doi = {10.1371/journal.pone.0151784},
    eprint = {1601.07154},
    archivePrefix = {arXiv},
    year = {2016},
}

The paper develops Multiscale network methods for analyzing a Scientific collaboration network of western classical musicians, mapping the structure of Music communities.

@article{bagrow2015cperc,
    author = {James P. Bagrow and Sune Lehmann and Yong-Yeol Ahn},
    title = {Robustness and Modular Structure in Networks},
    journal = {Network Science},
    volume = {3},
    pages = {509--525},
    doi = {10.1017/nws.2015.21},
    eprint = {1102.5085},
    archivePrefix = {arXiv},
    year = {2015},
}

Studies how Modularity and overlapping community structure affect Network robustness under failure, a question of Percolation on modular networks.

@inproceedings{cha2007youtube,
    author = {Meeyoung Cha and Haewoon Kwak and Pablo Rodriguez and Yong-Yeol Ahn and Sue Moon},
    title = {I Tube, You Tube, Everybody Tubes: Analyzing the World's Largest User Generated Content Video System},
    booktitle = {Proceedings of the ACM Internet Measurement Conference},
    pages = {1--14},
    doi = {10.1145/1298306.1298309},
    year = {2007},
}

A large-scale study of YouTube measures video Popularity life cycles, request distributions, and content duplication, with implications for User generated content delivery.

• GitHub Workflow - 2026-09-01
• Traffic projection - 2026-08-31
• Traffic calming - 2026-08-30
• Figure guide - 2026-08-29
• Traffic calming is cheap - 2026-08-24
• A letter to city council: Inability to execute is not a justification for inaction - 2026-08-20
• Git - 2026-08-17
• LaTeX Pet peeves - 2026-08-17
• Induced demand - 2026-08-16
• Quarto - 2026-08-16
• YNAB - 2026-08-15
• Ebike - 2026-08-14
• (Kinda) Leaving Overleaf - 2026-08-14
• Message passing - 2026-08-14
• Ahn et al., "Link communities reveal multiscale complexity in networks" (2010) - 2026-08-14
• Ahn et al., "Flavor network and the principles of food pairing" (2011) - 2026-08-14
• Aiyappa et al., "Can we trust the evaluation on ChatGPT?" (2023) - 2026-08-14
• Aiyappa et al., "Emergence of simple and complex contagion dynamics from weighted belief networks" (2024) - 2026-08-14
• Aiyappa et al., "Implicit degree bias in the link prediction task" (2025) - 2026-08-14
• Aiyappa et al., "Zero-shot stance detection in practice: insights on training, prompting, and decoding with a capable lightweight LLM" (2026) - 2026-08-14
• An et al., "SemAxis: A Lightweight Framework to Characterize Domain-Specific Word Semantics Beyond Sentiment" (2018) - 2026-08-14
• An et al., "Predicting Anti-Asian Hateful Users on Twitter during COVID-19" (2021) - 2026-08-14
• Bagrow & Ahn, "Network Cards: concise, readable summaries of network data" (2022) - 2026-08-14
• Bento et al., "Evidence from internet search data shows information-seeking responses to news of local COVID-19 cases" (2020) - 2026-08-14
• Blythe et al., "Massive Multi-agent Data-Driven Simulations of the GitHub Ecosystem" (2019) - 2026-08-14
• Bollenbacher et al., "On the challenges of predicting microscopic dynamics of online conversations" (2021) - 2026-08-14
• Cha et al., "Prevalence of Misinformation and Factchecks on the COVID-19 Pandemic in 35 Countries: Observational Infodemiology Study" (2021) - 2026-08-14
• Constantino et al., "Representing the disciplinary structure of physics: A comparative evaluation of graph and text embedding methods" (2025) - 2026-08-14
• DeVerna et al., "Modeling the amplification of epidemic spread by individuals exposed to misinformation on social media" (2025) - 2026-08-14
• Frank et al., "AI exposure predicts unemployment risk: A new approach to technology-driven job loss" (2025) - 2026-08-14
• Gates & Ahn, "CluSim: a python package for calculating clustering similarity" (2019) - 2026-08-14
• Gates et al., "Element-centric clustering comparison unifies overlaps and hierarchy" (2019) - 2026-08-14
• Grabe et al., "The Social Contagion Potential of Pro-Vaccine Messages on Black Twitter" (2023) - 2026-08-14
• Gu et al., "Principled approach to the selection of the embedding dimension of networks" (2021) - 2026-08-14
• Hasan et al., "The Impact of Viral Posts on Visibility and Behavior of Professionals: A Longitudinal Study of Scientists on Twitter" (2022) - 2026-08-14
• Hébert-Dufresne et al., "One pathogen does not an epidemic make: a review of interacting contagions, diseases, beliefs, and stories" (2025) - 2026-08-14
• Jiang et al., "Social inequality and cultural factors impact the awareness and reaction during the cryptic transmission period of pandemic" (2025) - 2026-08-14
• Jing & Ahn, "Characterizing Partisan Political Narrative Frameworks about COVID-19 on Twitter" (2021) - 2026-08-14
• Jing et al., "Sameness entices, but novelty enchants in fanfiction online" (2025) - 2026-08-14
• Jo et al., "Extracting hierarchical backbones from bipartite networks" (2020) - 2026-08-14
• Ke et al., "A systematic identification and analysis of scientists on Twitter" (2017) - 2026-08-14
• Kianersi et al., "Development and validation of a model for measuring alcohol consumption from transdermal alcohol content data among college students" (2023) - 2026-08-14
• Kim et al., "Labor Space: A Unifying Representation of the Labor Market via Large Language Models" (2024) - 2026-08-14
• Kim et al., "Uncovering simultaneous breakthroughs with a robust measure of disruptiveness" (2026) - 2026-08-14
• Kojaku et al., "The effectiveness of backward contact tracing in networks" (2021) - 2026-08-14
• Kojaku et al., "Residual2Vec: Debiasing graph embedding with random graphs" (2021) - 2026-08-14
• Kojaku et al., "Network community detection via neural embeddings" (2024) - 2026-08-14
• Kolchinsky et al., "The Minor fall, the Major lift: inferring emotional valence of musical chords through lyrics" (2017) - 2026-08-14
• Kwak et al., "A Systematic Media Frame Analysis of 1.5 Million New York Times Articles from 2000 to 2017" (2020) - 2026-08-14
• Kwak et al., "FrameAxis: characterizing microframe bias and intensity with word embedding" (2021) - 2026-08-14
• Lee et al., "Substitution of Nonpharmacologic Therapy With Opioid Prescribing for Pain During the COVID-19 Pandemic" (2021) - 2026-08-14
• Lee et al., "Systematic Evaluation of State Policy Interventions Targeting the US Opioid Epidemic, 2007-2018" (2021) - 2026-08-14
• Lee et al., "Social signals predict contemporary art prices better than visual features, particularly in emerging markets" (2024) - 2026-08-14
• Lee et al., "A semantic embedding space based on large language models for modelling human beliefs" (2025) - 2026-08-14
• Menczer et al., "Addressing the harms of AI-generated inauthentic content" (2023) - 2026-08-14
• Miao et al., "The latent structure of global scientific development" (2022) - 2026-08-14
• Mišić et al., "Cooperative and Competitive Spreading Dynamics on the Human Connectome" (2015) - 2026-08-14
• Moriano et al., "Community-Based Event Detection in Temporal Networks" (2019) - 2026-08-14
• Muralidharan et al., "Federating Governance: How Community Rules Scale with Mastodon Instances" (2026) - 2026-08-14
• Murray et al., "Unsupervised embedding of trajectories captures the latent structure of scientific migration" (2023) - 2026-08-14
• Nakandala et al., "Gendered Conversation in a Social Game-Streaming Platform" (2017) - 2026-08-14
• Nematzadeh et al., "Optimal Network Modularity for Information Diffusion" (2014) - 2026-08-14
• Nematzadeh et al., "Information overload in group communication: from conversation to cacophony in the Twitch chat" (2019) - 2026-08-14
• Newman, Message passing methods on complex networks - 2026-08-14
• Park et al., "Global labor flow network reveals the hierarchical organization and dynamics of geo-industrial clusters" (2019) - 2026-08-14
• Park et al., "How to Understand a City via Mobility" (2021) - 2026-08-14
• Park et al., "Rap as a social reflection" (2026) - 2026-08-14
• Parmer & Ahn, "Evolution of the Informational Complexity of Contemporary Western Music" (2019) - 2026-08-14
• Peng et al., "Neural embeddings of scholarly periodicals reveal complex disciplinary organizations" (2021) - 2026-08-14
• Perry et al., "Co-prescription network reveals social dynamics of opioid doctor shopping" (2019) - 2026-08-14
• Rodriguez et al., "Collective Dynamics of Belief Evolution under Cognitive Coherence and Social Conformity" (2016) - 2026-08-14
• Singh et al., "Misinformation, Believability, and Vaccine Acceptance Over 40 Countries: Takeaways From the Initial Phase of The COVID-19 Infodemic" (2022) - 2026-08-14
• Sriram et al., "The interplay of sex and genotype in disease associations: a comprehensive network analysis in the UK Biobank" (2025) - 2026-08-14
• Sugimoto et al., "Factors affecting sex-related reporting in medical research: a cross-disciplinary bibliometric analysis" (2019) - 2026-08-14
• Weng et al., "Virality Prediction and Community Structure in Social Networks" (2013) - 2026-08-14
• Weng et al., "Predicting Successful Memes Using Network and Community Structure" (2014) - 2026-08-14
• Wright et al., "Cognitive Linguistic Identity Fusion Score (CLIFS): A Scalable Cognition-Informed Approach to Quantifying Identity Fusion from Text" (2025) - 2026-08-14
• Wu et al., "Metrics and mechanisms: Measuring the unmeasurable in the science of science" (2022) - 2026-08-14
• Yang et al., "BiRank: Fast and Flexible Ranking on Bipartite Networks with R and Python" (2020) - 2026-08-14
• Yang et al., "Use of and Comorbidities Associated With Diagnostic Codes for COVID-19 in US Health Insurance Claims" (2021) - 2026-08-14
• Yang et al., "Comparing measures of centrality in bipartite patient-prescriber networks: A study of drug seeking for opioid analgesics" (2022) - 2026-08-14
• Yoon et al., "Persona2vec: a flexible multi-role representations learning framework for graphs" (2021) - 2026-08-14
• Yu et al., "Unique in What Sense? Heterogeneous Relationships between Multiple Types of Uniqueness and Popularity in Music" (2023) - 2026-08-14
• Yuan et al., "Implications of COVID-19 vaccination heterogeneity in mobility networks" (2023) - 2026-08-14
• Zhao et al., "Discovering Collective Narratives Shifts in Online Discussions" (2024) - 2026-08-14
• Parking minimum - 2026-08-14
• Claude Code: Tips - 2026-08-09
• Herdr - 2026-08-09
• Motornomativity - 2026-08-08
• AI Agents: How I Use Them - 2026-08-08
• Claude Code - 2026-08-08
• Don't ask to ask, just ask - 2026-08-08
• Motornomativity: How Social Norms Hide a Major Public Health Hazard - 2026-08-08
• Traffic evaporation - 2026-08-08