After many years of urban cycling, transit, and now-defunct ride share programs, we finally needed a car for medical trips, baby stuff, and seeing friends but opted to say 🙅 to the internal combustion engine. This is a quick summary in case others want to do the same but know as little about cars as I did.

Caveats: this advice is probably very 2025 Bay Area-centric, but probably works for a lot of geographies. If you need a car or are updating your existing one, just don’t get an gas-power car.

As of July 2025, the $7500 discount on EVs is set to expire on October 2025 because of the ‘Big Beautiful Bill’.

Cars

Criteria: DC-fast charging speed (< 30 minutes to 80%), cargo capacity (> 20 cu ft), and range (> 300 miles).

We ended up with a 2024 Hyundai Ioniq 5 SEL RWD, two year / 12k mile lease negotiated with a dealer entirely over text and then delivered to our house. The amortized cost to us has been about $600-$700/month for the life of the lease with about $2k fees and registration at signing. This was my first car ever and if we’d had an existing insurance this entire process (from form to delivery) could have taken an afternoon (which I thought was crazy).

I also looked at the Ioniq 6 (a Tesla-style sedan), which has better high range due to its drag profile, however it only has 11 cu ft of storage, about as much as a typical fridge.

Alternative if you’ve got the stomach for it: great deals on used Teslas.

Other Details (as of 2025)

Leasing and Insurance

EVs, EV power electronics platforms, and EV-related standards improve every year so you may want to consider getting a two year lease and upgrading. In the Bay Area, it’s been been easy through Leasehackr. You can provide a provisional proof of insurance through the website, after which you’ll need the exact vehicle identification number (VIN) from the dealer. Notably, you don’t even need to visit the dealership to see cars: I received photo confirmation via the dealer, did paperwork live over the phone and email, and arranged for delivery.

If you have other insurance you may be able to get a discount by bundling your other insurance through the same provider.. You’ll need to provide the exact VIN to the insurer to get a real policy. All that said, I think our two-driver policy through State Farm is expensive.

Some states have special clean car incentives for car pool lane and tolls. In California, the leased car will arrive preregistered by the dealer, after which you can apply for clean air decals and FasTrak.

Charging and Range

Hyundai currently provides free 30-minutes of DC fast charging through Electrify America and new, non-Tesla EVs are newly compatible with Tesla’s Supercharger network via the NACS standard. If you’re at home, you can easily charge your car overnight by attaching an off-the-shelf 40A ‘Level 2’ charger, like from Grizzl-E or Leviton to a 240V NEMA outlet. If you don’t have this outlet and but space in your circuit breaker panel, it costs about $400 to have a NEMA outlet installed. We don’t drive much, so keep our car top-offed with a regular 120V, 12A Leviton charger running from an exterior outlet to our car.

Things that suck: I actually don’t know what someone who’s renting in a multi-family situation or apartment building should do. I’d love to hear more about what this is like.

Apps: Route Planning, etc.

There’s a ton of fragmentation around car charging networks. Here are some of the good ones that also integrate with Apple CarPlay:

• A Better Route Planner
  • Uses a model of your EV to determine range and do route planning.
• PlugShare
  • Unified view of all available public chargers across all networks.

There’s some interesting stuff! For example, the City started doing more preventative maintenance in 2015 (assuming this attribution is correct):

Caveat: these numbers do differ slightly from the recent numbers reported by the City here.

Renovations ebb and flow, but recent, major repaving work seems to average ~3miles/year. I have no sense of whether that typical performance for a city Berkeley’s size.

Background

To win an election, you must first register a lot of voters. Elections also matter: they have real practical outcomes for voters and so voters should care about voter registration and the momentum around it. Southern states have structural advantages in the Electoral College, so their electoral outcomes are frequently national ones. Some states (like NC) may be diversifying demographically and politically faster than others, but are affected by the crazy gerrymandering of the state legislative districts.

To recap 2020, Biden lost NC to Trump by about 74k votes, or about 1.3% of votes cast. State Democrats also lost the state Supreme Court, lost State House seats, and lost the opportunity to push fairer electoral maps following the 2020 Census. I donated to the Long Leaf Pine Slate, which lost 15 of 17 races they backed! Nationally, the Democratic Party’s ground organizing campaign focused on mail-in voting (which was successful) but prohibited in-person organizing until September 2020 due to obvious COVID-19 concerns. The NC GOP, as far as I could tell, simply adopted social distancing and masking policies for in-person registrations well before the Democratic Party.

You can kind of see this in the accumulated voter registration totals leading up to the election with the Democrats lagging new registrations into Election Day:

2024 … As of December 2023

For additional commentary on this, see the writeup from September 2023!

New registration totals since November 2020:

| Party      | All    | w/o Meck/Wake |
|:-----------|-------:| -------------:|
| DEM        | 329942 |        240560 |
| REP        | 338680 |        292778 |
| THIRD      |  17816 |        14002  |
| UNA        | 612784 |        472383 |

Initial Look: 2024 … As of September 2023

New registration totals since November 2020:

| Party      | All    | w/o Meck/Wake |
|:-----------|-------:| -------------:|
| DEM        | 307628 |        223729 |
| REP        | 309917 |        267982 |
| THIRD      |  13252 |        10428  |
| UNA        | 561085 |        432133 |

Unaffiliated voters still dominate the voter registrations (as they have for some time), but new Democratic and GOP voter registrations since December 2020 are approximately even! The gap between newly registered Republicans and Democrats is less than 3k.

Wake and Mecklenburg County still make up a huge fraction of the new electorate and they are the biggest urban Democratic counties in the state. As of August 2024, the ‘No Labels’ Party is official in NC, with about 400 people registered, of which 52 are in Wake County.

If you exclude Wake and Mecklenburg the registration gap grows but isn’t that huge (around 43k). I would have expected something much larger, given the past organizing resources thrown at these larger counties.

No notable registration bumps following June 2022, which is when the Dobbs Supreme Court ruling overturned Roe. Registration bumps coincide with state primaries and during the midterms, which is not surprising.

Methods

I love that it’s possible for anyone to do this kind of analysis at home using open source tools. To generate these graphs above I used Pandas and a Jupyter notebook and you can follow a similar process for your state. In the following, I’ll talk through the process using the NC voter file, using similar analysis to the one I posted here in 2020.

Getting the Data

The NC Board of Elections maintains a page where you can download the public voter file. Some quick links:

• Voter file
• Voter file schema

If you’re setup with AWS and use the terminal, you can download these files directly from where they’re hosted:

$ aws s3 cp s3://dl.ncsbe.gov/data/ncvoter_Statewide.zip .
$ aws s3 cp s3://dl.ncsbe.gov/data/layout_ncvoter.txt .

Normalize the Data

The statewide voter file is big: about 3.6G and almost 8.4M rows (as of 09-03-2023). Like voter files everywhere, it contains a huge amount of identifying information that would be considered private in any other context. I didn’t want to exhaust my RAM while plotting data, so I parsed and simplified this dataset to something smaller and less creepy: new, active registrations since the last general election, voter demographics, and city/county data.

import pandas as pd

# Parse the data (drops a few records)
df = pd.read_csv('ncvoter_Statewide.txt',
                   delim_whitespace=True,
                   on_bad_lines='skip',
                   encoding_errors='ignore',
                   parse_dates=['registr_dt'],
                   infer_datetime_format=True)

# Normalize and save subset of interest
to_keep = ['registr_dt', 'voter_status_desc',
           'birth_year','race_code', 'ethnic_code', 'gender_code',
           'party_cd', 'county_desc', 'res_city_desc']
start_date = '2020-12-01'
newly_registered = df[to_keep][df['registr_dt'] > start_date]
newly_registered = newly_registered[newly_registered['voter_status_desc'] == 'ACTIVE']

# Group all third parties, including 'No Labels' (!)
newly_registered['party_cd'] \
    = newly_registered['party_cd'].replace(['LIB', 'NLB', 'LIB', 'GRE'], 'THIRD')

# Save the subset
newly_registered.to_csv('ncvoter_Statewide_post-2020-12-01.csv', index=False)

The layout_ncvoter.txt includes a schema for the voter file. This is the subset for the columns I selected above:

------------------------------------------------------------------------------------
name                    data type          description
------------------------------------------------------------------------------------
county_desc             varchar(15)        County name
voter_status_desc       varchar(25)        Registration status description
registr_dt              char(10)           Registration date
res_city_desc           varchar(60)        Residential city name
race_code               char(3)            Race code
ethnic_code             char(3)            Ethnicity code
party_cd                char(3)            Registered party code
gender_code             char(1)            Gender/sex code
birth_year              char(4)            Year of birth

The resulting table newly_registered is much smaller: about 1.2M records for a 62MB CSV file. The contents (from newly_registered.head()) looks like this:

|    | registr_dt          | voter_status_desc   |   birth_year | race_code   | ethnic_code   | gender_code   | party_cd   | county_desc   | res_city_desc   |
|---:|:--------------------|:--------------------|-------------:|:------------|:--------------|:--------------|:-----------|:--------------|:----------------|
|  0 | 2023-05-09 00:00:00 | ACTIVE              |         1945 | W           | NL            | M             | REP        | ALAMANCE      | GRAHAM          |
|  1 | 2023-05-17 00:00:00 | ACTIVE              |         1967 | W           | NL            | F             | REP        | ALAMANCE      | HAW RIVER       |
|  2 | 2021-04-15 00:00:00 | ACTIVE              |         2003 | W           | NL            | M             | REP        | ALAMANCE      | GIBSONVILLE     |
|  3 | 2021-05-21 00:00:00 | ACTIVE              |         1981 | B           | NL            | M             | UNA        | ALAMANCE      | BURLINGTON      |
|  4 | 2022-10-07 00:00:00 | ACTIVE              |         1992 | W           | NL            | F             | REP        | ALAMANCE      | MEBANE          |

Analysis: County Aggregates

To get the county aggregates you:

• group by county name and party registration
• count those groups
• pivot counts by party registration and total for each county
  

# Color coding
COLORS = {'DEM': 'blue', 'REP': 'red', 'THIRD': 'green', 'UNA':'purple'}
parties = ['DEM', 'REP', 'THIRD', 'UNA']

# Group, count, and pivot
# Note: following .count(), all columns have the same value, so choose any one and rename to count
groupby = ['county_desc', 'party_cd']
total_by_county = newly_registered.groupby(groupby).count()['birth_year'] \
   .reset_index() \
   .rename(columns={'birth_year': 'count'})\
   .pivot_table(values='count', index='county_desc', columns='party_cd')
total_by_county['ALL'] = total_by_county.sum(axis=1)
total_by_county = total_by_county.sort_values(by='ALL')
total_by_county

ax = total_by_county[parties.plot.bar(stacked=True, color=COLORS)
ax.set_title('Statewide New Voter Registrations Since 12-01-2020',)
ax.set_xlabel('County')
ax.set_ylabel('Registrations (count)')

Some sample data:

| county_desc   |   DEM |   REP |   THIRD |   UNA |   ALL |
|:--------------|------:|------:|--------:|------:|------:|
| TYRRELL       |    65 |    65 |     nan |   104 |   234 |
| HYDE          |    93 |   131 |       2 |   160 |   386 |
| GRAHAM        |    73 |   402 |       9 |   287 |   771 |
| GATES         |   204 |   330 |      18 |   428 |   980 |
| WASHINGTON    |   283 |   255 |       9 |   438 |   985 |

Analysis: Monthly Registrations

This will produce demographic statistics similar to what you’d find on the NC Registration Statistics page. To get these demographics you group by registration week and demographic feature (race/ethnicity, gender/sex, and party registration) and then count.

groupby = [pd.Grouper(key='registr_dt', freq="M"), 
          'race_code', 'ethnic_code', 'gender_code', 'party_cd']
monthly_voter_stats = newly_registered.sort_values('registr_dt') \
                      .groupby(groupby) \
                      .count()['birth_year'] \
                      .reset_index() \
                      .rename(columns={'birth_year': 'count'})
monthly_voter_stats

Some sample data:

|    | registr_dt          | race_code   | ethnic_code   | gender_code   | party_cd   |   count |
|---:|:--------------------|:------------|:--------------|:--------------|:-----------|--------:|
|  0 | 2020-12-31 00:00:00 | A           | HL            | M             | UNA        |       2 |
|  1 | 2020-12-31 00:00:00 | A           | NL            | F             | DEM        |      53 |
|  2 | 2020-12-31 00:00:00 | A           | NL            | F             | REP        |      25 |
|  3 | 2020-12-31 00:00:00 | A           | NL            | F             | THIRD      |       2 |
|  4 | 2020-12-31 00:00:00 | A           | NL            | F             | UNA        |      72 |

You can then pivot with:

monthly_voter_stats.groupby(['registr_dt', 'party_cd'])['count'].sum() \
  .reset_index()\
  .pivot_table(values='count', index='registr_dt', columns='party_cd')

to get:

| registr_dt          |   DEM |   REP |   THIRD |   UNA |
|:--------------------|------:|------:|--------:|------:|
| 2020-12-31 00:00:00 |  5950 |  6756 |     288 | 10279 |
| 2021-01-31 00:00:00 |  6601 |  6621 |     287 | 11026 |
| 2021-02-28 00:00:00 |  6834 |  6700 |     314 | 11104 |
| 2021-03-31 00:00:00 |  8395 |  8672 |     336 | 13642 |
| 2021-04-30 00:00:00 |  7892 |  7985 |     325 | 12829 |

Changelog

• 12-28-2023. Updated to include registration numbers through 2023.
• 09-04-2023. Initial post.

2022 - A Year in Review

2022 was a dense, eventful year, interleaved with Omicron and COVID-related logistics that became less intense here as the year went on. If 2020 lasted an eternity, 2021 and 2022 both rolled by in such a breathless flash I haven’t really chance to fully appreciate everything that happened.

Projects

Backyard wedding paella (!) from Venga Paella Oakland.

After moving into our new place, we replaced a foundation wall and replaced a big stretch of a concrete driveway with gardening space: Zmagg designed the walkway and landscaping layout and is writing up a post about how we went about landscaping our space.

How it started:

Foundation: Before and during removal

How it’s going:

Garden plan and late 2022

Travel

Sea Ranch, CA

Post-wedding travel: V+A Museum and Chelsea Flower Show in London; climate, geography, and sherry as told by Bodegas Tradición, Jerez de la Frontera, Spain

North Vancouver, BC: Capilano River Regional Park

I wrote up the Vancouver trip and you can read about it here!

Work: Span Drive

My big work project this year was a soup-to-nuts design, implementation, and certification of the “functionally safe” embedded system system powering Span Drive, an EV charger that dynamically adjusts charging rate based on your home’s electrical load.

Warning: flagrant generalizations and lack of references ahead.

My lesson from the last year: there are remarkably few consumer hardware companies and startups working to directly replace fossil-fuel-based consumer technologies with electric ones (“electrification”).

These include creating both electric equivalents (EVs and e-bikes vs internal combustion engines, cheaper induction stoves and heat-pump water heaters, etc.), introducing new new product categories for microgrid management, making more reliable public EV charging infrastructure, or supporting the booming US residental solar industry. If you’re a Bay Area software engineer looking to join the clean energy industry, your options are largely behemoths like Tesla, SolarEdge, Enphase, Rivian, etc.

Given this — as well as the passage of the Inflation Reduction Act — it’s actually really surprising how many climate-focused startups are working on really indirect solutions_ such as creative carbon accounting, carbon finance, monitoring, and monitoring of electric assets (like EVs). Some of these efforts seem really tenuous ground truthing to real climate-related outcomes.

On a contrasting note, check out this amazing EVSE built-into a London street lamp:

Media, Books, Reading, etc.

I read a lot fewer books than Zmagg. I don’t keep a GoodReads or other list of books I read, but these are the ones I remember finishing:

• Silas Marner (George Eliot)
• Means of Ascent (Robert Caro)
• A Place of Greater Safety (Hilary Mantel)
• Laudato Si, Encyclical on Climate Change and Inequality: On Care for Our Common Home (Pope Francis)

My technical reading this past year was almost entirely related to work:

• The Definitive Guide to ARM Cortex-M0 and Cortex-M0+ Processors (Joseph Yiu)
• Understanding Digital Signal Processing (Richard Lyons)
• Operating Systems: Three Easy Pieces (Arpaci-Dusseau)
• Rust Programming Language (Blandy, Orendorff, Tindall)
• Renovation, 5th Edition (Litchfield and Harley)

Most played music and radio:

• Northwest Passage (Stan Rogers)
• Good Food (Evan Kleinman @ KCRW)

Movies, etc.:

• NOPE
• Glass Onion
• Stranger Things (Season 1)
• Andor (Season 1)

Food

I’m excited to do a lot more new cooking in 2023 than I did in 2022. After three years, we finally got off the wait list for the Rancho Gordo Bean Club. I signed up in 2022 and apparently totally forgot 😅! If you want to come over and eat beans in 2023, send me a message!

Notable household experiments and staple from 2022 included:

• Homemade Tempeh
• Eggplant Lasagna
• Cardoon Gratin

My favorite restaurant meal of 2023 was Mari Vega’s Mesoamerican tasting menu at the Lord Stanley Turntable. My favorite things from this were a plum chamoy and this amazing cod baked into a soft clay mold of a fish.

They baked the molds at a pottery studio nearby and you had to pull it apart to get at the fish!

October 2022 Vancouver Trip

After a somewhat intense year for work, we took a one-week vacation to Vancouver and Vancouver Island to eat and look at some big trees. We got pretty lucky – certainly weather-wise – on this trip! By the time we arrived on October 21, it had only just started raining after what I heard was a six-month dry spell and an accumulation of wildfire smoke. It was damp, but otherwise dry, for the first half; the remaining days were simply the wettest weather I’ve ever experienced.

All that aside: Definitely go! The food is excellent, it’s easy to get around, and everyone we met was super-friendly.

Vancouver-area trip highlights and area recommendations:

• Aberdeen Centre food court, Richmond, BC generally
  • Mandatory stop – just off the Canada Line – after leaving YVR. Within 1km off of Aberdeen Center there are at least three to four food courts and strip malls, covering the Cartesian product of [Persian, Japanese, Korean, Sichuan, Cantonese, Hunanese, Taiwanese, Malay] and [Vegetarian, Vegan, Gluten-Free].
• Riley Park / Mount Pleasant. We stayed at an AirBnb in this neighborhood between Ontario and Main Street and it was Pretty Nice. Accessible on transit via TransLink 8 bus and the Canada Line. Some standouts:
  • Liberty Cafe and Bakery
  • 33 Acres Brewing
  • Sushi Hil
  • Hawkers Delight
• Bill Reid Museum and Biking Stanley Park
  • We stopped at the Bill Reid museum to see some Haida First Nation art before heading over to Stanley Park for a rental bike ride.
• North Vancouver: Hiking in Capilano River Regional Park (skip the suspension bridge and hike to the salmon hatchery instead) and Central Lonsdale (fantastic Persian food and markets).

After about six days in Vancouver, we rented a car at the airport and took the BC Ferry from Tsawwassen to Victoria and then drove up to Port Renfrew for a two day stay. Some highlights:

• Ferry ride from Tsawwassen to Victoria (Swartz Bay), winding through the San Juan Islands ($100 for the car, $18/per person).
• Visiting Eden Grove and Big Lonely Doug with Day Trip Drea.

We wanted to, but didn’t get:

• 太二 酸菜鱼 (“Tai Er Chinese Sauerkraut Fish”), or actually more regionalized Chinese food.
  • This place ferments their own mustard greens and apparently has an outpost at Changi Airport too. We showed up on a Tuesday night, and learned they had sold out a full hour before closing.
• Carmanah Walbran Provincial Park
  • This is an old-growth temperate rain forest on Vancouver Island, about a five-hour drive on rough roads from Port Renfrew. Shockingly, many of these ‘Crown Lands’ aren’t protected from logging in the same way similar landmark forests are protected in the US. We weren’t able to visit since lumber companies blocked the more convenient roads after the Ferry Creek Protests. Read more about the preservation work around this forest at the Ancient Forest Alliance.
• Indian food
  • When I made my way to Punjabi Market, it was apparently Diwali (and Rishi Sunak’s installation as UK’s PM). Two sweet shops east of Punjabi Market had pretty long lines for freshly fried (!) jalebi.
• Salmon season
  • Salmon are apparently spawn shortly after the rain starts, since the streams fill back up. Unfortunately, this is worst time to eat them (unless you’re a hungry bear).
• Sun Yat-Sen Classical Chinese Garden, and really, any public gardens in Vancouver and Vancouver Island.

We saved locations both visited and not on a Google Maps list here. Places change! Restaurants and shops in all cities turn over eventually, so best of luck finding these places in the future.

Some tips:

• Getting around. You don’t need a car unless you’re leaving the metro area. Walking, public transportation via Skytrain/bus/Seabus are pretty good and Vancouver’s not that big. If you’re planning to hike in North Vancouver, you can take TransLink buses (240 and 246) from Downtown Vancouver to great public parks, and then return via the SeaBus.

  You will need a 4WD SUV if planning any backcountry visits on Vancouver Island.

• Food. Like Los Angeles, Vancouver has a wide variety of really great food in food courts. Some of these food courts are in unexpected places. Unlike Los Angeles, you don’t have to drive for hours to get to them. Compared to the SF Bay Area, it felt like there was a lot more experimentation, more broadly, as well as more foreign imports (like Yuan’s Chuan Chuan Xiang from Chengdu). Typical meals were cheaper than the Bay Area and LA, but not by much.

• Ferry logistics. You can rent a car in Vancouver to drive on Vancouver Island, but you’ll need to transport that car via the ferry (fare is $18/person, $100/car). You should definitely reserve a spot on the ferry as soon as you’re sure about your plans, and you can check the BC Ferries website for surprisingly detailed information and boat space allocations and reservation utilization. We were one of the last dozen or so cars let onto the ferry without a reservation.

  If returning from a distant part of Vancouver Island for a Sunday afternoon or evening flight, time budget for cross-island driving, potential rain and ferry complications, as well as weekend traffic going into Richmond. We avoided Sunday traffic through Victoria by returning through Nanaimo.

• Rain. If you’re planning to explore on foot during the rainy season, bring a drybags and/or waterproof backpack (…. or borrow a giant umbrella).

(🙏 Thanks to coworkers, friends, and family who gave trip planning advice!)

16 March 2018 - San Francisco

The Metric Geometry and Gerrymandering Group (MGGG) hosted a workshop in SF around redistricting, gerrymandering, and voting rights. My notes from the event are below.

Errors, omissions, and inaccuracies here are entirely my own! If you have any errata, please let me know.

Thursday, March 15

09:30 - 10:20AM - Moon Duchin (Tufts) - slides

Math, Politics, and Law in the Study of Gerrymandering

• Techniques to identify, adjudicate, and remediate
• MGGG has been putting on these workshop
• Redistricting is a mathematical partitioning problem (smash or scalpel)
• Cycling of redistricting: enumeration, apportionment, partition (litigation)
  • Two federal bodies:
    • Senate: 2 per state
    • House: % share of US population, adding up to 435.
  • States are required to redistrict after each Census
  • Officials set their own process
• Six redistricting principles
  • Mathematical
    • Equal population (number) - All districts must be equal to each other +/- one.
    • Compactness (shape) - “Reasonable” shape. Geometry here is all 1900
    • Contiguous (shape) - All districts must be connected
      • From these Cohen-Addad-Klein-Young (picture from Texas).
  • Political
    • Respect for county/city boundaries (political geography) - Don’t split up a city.
    • Respect for “communities of interest” (sociology) - More on this in the future. Hard to make algorithmic.
    • Compliance with Voting Rights Act (race) - Need to give minority communities the ability to elect a candidate of their choice.
  • Other stuff
    • Other things not on test, but seem intuitive: Proportionality? Competitiveness? Responsiveness? Stability? Voice for the marginalized? Different states can adopt different combinations of rules.
• Concepts
  • Draw diagrams. Plurality election in each vote.
  • Efficient majorities for you = Packing and Cracking for your opponents. Cracking is dispersed among losing districts. Opponent is packed into smaller number of districts. Theoretically it’s possible to double your representation from your votes.
  • Examples from WI, PA, and NC with outstanding congressional lawsuits. Half the votes, 63/99, 10/13, and 10/13 seats.
  • No legal framework around political gerrymandering
  • Parties gerrymander when they have control to do so.
• A redistricting plan meets a distribution of voters
  • You have a distrusting plan and compare them to a distribution of voters (D, delta).
    • Delta determines V, the vote share for the party
    • D determines compactness
  • (D, delta) pair gives (V1, ... , Vn) the vote share by district
  • (V1, .... , Vn) gives S, the seat share for a party (how many Vi > 1/2).
  • Fix Delta, very D: How do they perform under various scenarios?
• Menagerie of districts. tl;dr: Important to look at shape and the underlying demographics.
  • Florida-5
    • Created after 1990 from 2013 - 2017
    • Packing
    • Change opposed by packed incumbent
  • Maryland-3
    • Drawn by Democrats (opposed by civil rights groups)
  • Pennsylvania-7
    • Barely contiguous. Strung together by only two buildings.
  • North Carolina-12
    • After 1990 census: billed as a voting rights district. Race is used as a proxy for map design.
  • Illinois-4
    • Two Hispanic communities linked by a highway
    • Accidental packing
• Race and the Voting Rights Act of 1965
  • Aimed at eliminating barriers to the black vote
  • Only supposed to last for four or five years.
  • Any states with problematic regions needed preclearance with all new plans.
  • Still best tool for addressing redistricting fairness.
• Race and party = ‘conjoined polarization’
• Eyeball compactness test is dead
• Dual graph representations describe both population density and connectedness.
• See likelihood analysis
• Q/A: ‘Democrats pack themselves and therefore underperform’: counterexample is actually Massachusetts (Republicans uniformly spread out, but no seats).

10:30 - 11:20AM - Kris Tapp (St Joseph’s) - slides

Measurements of Political Gerrymandering

• Mathematics of elections outcomes and gerrymandering
• Gerrymandering is detected by:
  1. Geometry (‘odd shapes’)
  2. Statistics (‘more votes than average map’)
  3. Symmetry (‘if voter opinions shift to 50-50, A would have one more seats’)
  4. Wasted votes (‘party B wasted more votes than party A’)
• Presentation focuses on: 3 and 4 are computed using only district outcome data.
• Idea of Symmetry Methods (Gelman, King, and Grofman)
  • Simulate a uniform shift in voter opinion
  • Can produce a simulated seats-votes graph
  • A fair outcome should be symmetric about the origin. Unfair outcome has a non-zero bias.
  • If tables turned so blue got 60% of the vote, they wouldn’t have received the same award.
• Weakness of symmetry methods
  • Rooted in speculative, counterfactual simulations
  • Methods never impressed Supreme Court, but Kennedy suggested alternatives
  • Led to work in other areas.
• Wasted vote methods (Eric McGhee, 2014)
  • Efficiency gap (Excess vote, losing vote)
  • Always between -0.5 and 0.5 (up to half of the votes are wasted)
  • Assumes equal turnout (depends only on seat margin and vote margin)
  • Not covering unequal turnout (see Veomett, 2018).
  • Efficiency gap has slope 2 on the symmetry method parallelogram
  • With 60% of vote, you deserve 70% of seats.
  • With 75% of vote, you deserve 100% of seats. With more lopsided elections, EG is useless.
  • Is “winner’s double bonus” normative? Fit line for 1972 state house election data closely matches EG.
• Shortcomings in efficiency gap
  • Depends on election outcome and volatile in competitive races.
  • Flags demographic packing that isn’t related to gerrymandering
  • See Gill v. Whitford
  • Penalizes packing and cracking? Yes, but depends on cutoff of the percentage of the vote.
• Variations on the EG formula - Weight (Nagle)
  • Excess votes counted by weight lambda
  • lambda=1 counts votes in excess of what’s needed for majority
  • lambda=2 counts votes in excess of what losing party received (generalizes to multi-party elections).
  • A district won with 66-75% of the vote is considered packed.
• Variations on the EG formula - Relative (Nagle)
  • Measures the difference between proportions of votes that parties wasted.
  • Parties wasted same proportions, but not the same number.
  • Grounded in voter fairness as opposed to party fairness. A randomly selected voter from party A is equally likely to have wasted their vote as someone for party B.
  • Court has already ruled that lack of proportionality is not a political right.
• Variations on the EG formula - Weighted-Relative (Cover)
  • Fails any efficiency principle

11:30 - 12:20PM - Santiago Olivella (UNC) - slides

Other Frontiers: Voting Precincts and Associated Voting Costs

• Not about redistricting
• Lines matter: How we discretize the geographic distribution of preferences matters for electoral outcomes (social science “path dependence”). More perspectives are coming into play, with many academic disciplines bearing on the problem.
• Districts matter: smallest units of votes. Constraints are common across districts.
• Redistricting has gotten increasingly political.
• Costs of participating can be asymmetric on different groups, as a consequence of seemingly technical decisions around electoral administration: where to place polling places, how many voting machines, how much staff, etc. (bread and butter of applied mathematics).
• Questions to ask to measure electoral quality of service
  • How long did it take to vote? (varies by race)
  • Registration times/locations
  • ID/requirements
  • Learning about candidates
  • Understanding ballot use
• If costs are different across population groups, then we say there is discrimination. These may affect registration and turnout rates. Number of eligible voters is not known. p* Issues also affect other countries, not just the US (e.g., population centers and polling places in Hungary).
• Distribution of outcomes are not a function of asymmetry or simply disinterest. Interests are difficult to observe.
• Bayesian latent variable model of participating and turnout (joint work with Kelsey Shoub)
  • Use it to determine hypothetical interaction between registration, turnout, and turnout given registration.
  • Excellent use of STAN
  • Simulated
  • Apparent discrimination for Hispanic voters in NC is quite high, especially during midterm elections.

02:00 - 02:50PM - Eric McGhee (Public Policy Institute of California) - slides

​Partisan Gerrymandering and the Supreme Court

• Step back and talk about the legal issues
• Legal status quo
  • Davis v Bandemer (1986). Court decided that partisan gerrymandering is justiceable.
  • Justice White: we believe partisan gerrymandering exists, but we have no means of actually measuring it. The court also doesn’t want to interfere, because there’s no clear definition or criteria on gerrymandering.
  • Vieth v Jubelirer (2004). Polarized court. Court is split, but Kennedy says that one has not yet been provided: if a standard emerges, it should be used.
  • LULAC v Perry (2006). Still split, but ‘Gelman-King bias’ (i.e., symmetry) is provided to the court. Kennedy suggests that technologists may come up with a clear means of shedding light on this measurement question.
• The measurement problem is difficult
  • System was designed for geographic/district representation, but parties dominate American politics (i.e., an aggregate across geography). Parties do the work, but are defined by geography. The votes-seats relationship is not codified in law, and parties are not fundamentally covered through law.
  • Geographic system but with parties laid on top: impedance mismatch is the source of the Court’s difficulties.
• Consequence: we now have an opening for mathematics and social science
  • It’s a real challenge for the court, but requires a transparent solution
• Metrics before the court
  • Gelman-king bias (Gelman & King, 1994)
    • Principle: equal parties should be treated equally (‘symmetrically’). Same votes, same seats.
    • Advantages: grounded in core idea of majority rule.
    • Disadvantages: inconsistent and weird for competitive states. May suggest gerrymandering where it may not exist. Advocates have not offered thresholds for when a map is gerrymandered
  • Mean-median difference (McDonald and Best, 2015)
    • MMD = average district outcome (mean) - outcome dividing in half (median)
  • Efficiency gap (McGhee and Stephanopoulos, 2015)
    • Wasted votes don’t apply to a win.
    • Advantages: works for competitive and uncompetitive states (up to a point), targets efficiency. 2-1 seats-to-votes fits historical pattern, and this basis is used on historical data. Can also support turnout variations and multiple parties. Good for simulation: allows to measure intent.
    • Disadvantages: untested! Is it the only or best way? Requires commitment to equal wasted votes and can vary from one election to the next. Bad for simulations: does it really cover all of them?
  • Simulation generally: does it matter if a plan is improbable?
• Gill v. Whitford
  • 2011 Wisconsin Assembly Plan: aggressive seat maximization plan (48.6% of vote yields 60/99).
  • Metrics agree for a competitive state.
• Culture clash
  • Complex concepts, such as the EG, will fail Justice Robert’s ‘man on the street’ test.
  • Can the court solve this problem, a complex measure, without compromising its legitimacy?
  • Each measure has a weakness against the court.
• New measures
  • Determines intent
  • Declination (Warrington, 2017)
  • Sampling as sensitivity testing (Chikini, Frieze, and Pegden, 2017).
  • Aggregate into simple proportionality (Chambers, Millers, and Sobel, 2017).

03:00 - 03:50PM - Michal Migurski (PlanScore)

Making Gerrymandering Boring

• Works on Planscore making gerrymandering modeling more useful and hopefully back into a niche problem we don’t have to worry about (i.e., something predictable, uncontroversial, and low-reward).
• Example: WalkScore made walkability mainstream
• Current visualizers use past election results (Trump v Clinton) as a proxy for assessing redistricting plans.
• 1000-district model
• New data sources:
  • Open Elections
  • District shapefiles
• Kennedy on metrics: limit weak, but non-frivolous, metrics. Metric should ensure that elected representation reflect real voter intent.

05:00 - 05:50PM - Bruce Cain (Stanford)

By Form or By Outcome? The Fifty Year Quest for Redistricting Fairness

• Currently revisiting some ideas from early discussions
• Discussion around putting partisan and racial gerrymandering on comparative terms. How aggressive can you be in remedying racial injustice through redistricting?
• Nothing in the Constitution asserts proportionality, but does protect against racial protections (because of prior history of discrimination).
• “A measurement is not the same thing as a standard”. It is necessary, but it is not sufficient. The thresholds here need to be defined, and should it be done at the state level or the national level?
• A solution for one state may not work for another
• Homopoly
• Lots of interesting stories about being on redistricting boards.

06:00 - 07:30PM - Justin Levitt (Loyola Law), Moon Duchin (Tufts), Michal Migurski (PlanScore)

Panel Discussion - “Technology and Law in Redistricting”

• Justin Levitt: Coming up with the best redistricting plan through technology is like coming up with the ideal tax system through programming. There are, however, many ways we can use technology to alleviate the problems.
• Moon Duchin:
  • Discretized Polsby-Popper (used because it’s referenced and available to use in Maptitude).
  • Technological intervention: How extreme is your intervention?
  • A good plan doesn’t have to sit at the highest bar
  • These methods are for analysis, not generating plans
  • Both sides of the bell curve are not created equally.
• Justin Levitt: must start simulations with the governing law of the state (e.g., laws don’t require compactness, for example). Not having these constraints mean that your simulations don’t match the political will of the state.
• Michal: It’s only a matter of time for the block chain people to show up.
• Moon Duchin: Lots of clever ways of doing things like encoding the Voting Rights Act, communities of interest, etc. into redistricting sampling.
• Justin Levitt: Lots of opportunities to computationally verify the Voting Rights Act.
• Moon Duchin: See Justin Levitt’s website about state-level redistricting rules. Frequently, laws don’t actually make that much sense which makes operationalizing them very difficult. Also, random assignment doesn’t work (via the Central Limit Theorem).

Friday, March 16

09:30 - 10:20AM - Lisa García Bedolla (UC Berkeley) - slides

Why the Lines Matter: Race and Representation

• NOTE(mookerji): Missed the beginning of the presentation
• Commentary on disparities across race on political representation.
• The problem: the lack of outreach across races. Candidates don’t reach out to constituents they don’t need to (e.g., safe majority-minority districts, ones that may not contribute to a win).
• In studies (on white voters), children get political preferences from parents. Immigrants pick up preferences from children, who learn civic engagement from (public) school.
• Why does this matter? Low engagement in US is structural and ultimately affects the quality of public policy.
• Please don’t say ‘identity politics’: race matters because it affects their material experience of reality. People with similar racial experiences likely experience the same structures.
• Examples from American Values Survey
• Turnout affects who candidates speak to, and some candidates may only target specific groups and ignore others (or self-identify on certain positions that may not be universally felt across groups; e.g., AVS: is one group systematically targeted by the police?)
• tl;dr: structure inequality has political consequences. You can have different interpretations of current events because of your race-consequent social position.
• Q/A: Compulsory voting improves voting rights.
• Q: What line drawing principles could increase participation? A: Hard question. Anything that isn’t drawn to protect incumbents.
• Q/A: A lot of political money is spent through ineffective reach out (mailers, TV ads) in national campaigns because of compensation structure around political consultants.

10:30 - 11:20AM - Greg Herschlag (Duke) - slides

Quantifying Gerrymandering: Revealing Political Structure with Sampling

• Duke team’s work:
  • Gill v. Whitford (WI State Assembly)
  • Common Cause v. Rucho (NC Congressional)
  • NC vs Covington (NC State Assembly)
  • See group page
• Q: When is a map fair? When is it typical?
  • What if we drew the districts randomly? (No regard to party or demographics)
  • Can use algorithmicly-generated maps to benchmark plans?
  • All use MCMC
  • Recipe: generate compliant random redistricting plans, generate a winner
  • NC redistricting simulation (distribution of outcomes is between 4 to 9 for 13 seats). Judge panel is a typical result, per the MCMC ensemble. Note that via this method the WI State Assembly plan would seem typical.
  • Gerrymandering can occur in nicely-shaped districts, and skewed results don’t necessarily indicate that it has occurred.
  • Gerrymandering index. See Eric Lander amicus brief in Gill v. Whitford.
  • Plot series of ensembles: Parametric on Fraction of Opposing Party Vote // Likelihood vs Number of Seats for Opposing Party
  • Sampling can be used to recover these structural geopolitical biases
• Q: Can you surface gerrymandered districts?
  • Fix precinct. Determine outlier precincts compared to the generated outcomes of the its containing district. You can then shade (red or blue) the district by their tendency relative to district (i.e., deviation from the median).

11:30 - 12:00PM - Taren Stinebrickner-Kauffman and Kelsey Kauffman

Modern Malapportionment: Prison Gerrymandering and Local Redistricting

• Nobody is tracking whether local governments are redistricting when they are required to do so.
• All government bodies with single-member electoral districts in the US are required to redistrict every 10 years, including local governments with single member districts.
  • 90k local governments in the US
  • No central database of how/when/where how this is happening.
  • Failure to redistrict systematically advantages entrenched parties, affects the pipeline to higher office.
  • Lots of egregious cases of maldistricting and refusal to redistrict.
  • See the webpage
  • Teachers: identify local governments that are not redistricting. Present data, sue governments that aren’t in compliance with the law.
• Shifts to prison incarceration
  • Prison population increases dramatically in 1970, not long after the population of the voting rights act.
  • How is representational equality affected by incarceration?
  • Q for survey with state representative: which inmate would you feel was more truly a part of your constituency? Seems that the current state of counting prisoners in population base, may be unconstitutional.
  • Census directors ultimately decide on population base, even though it’s not in their job description.
  • Possible solutions (maybe Constitutional):
    • Courts: Maybe remove prison inmates from population base?
    • Reallocate prison inmates to their hometowns (or next of kin, most recent address) for the purposes of redistricting.
  • See the webpage.
  • Stuff here at state-level: inmates can vote in some places, get reallocated.
  • See article: Counting Matters: Prison Inmates
    • Link
• See the page from Common Cause

02:00 - 02:40PM - Tanya Cecena Pellegrin (MALDEF) and Jonathan Stein (Asian Americans Advancing Justice) - slides

The California Voting Rights Act

• Voter suppression techniques
• At-Large Elections: found to be discriminatory by the courts. Always dilutes the (numerical) minority vote.
• 100 US Senators elected at-large. Has effect of suppressing minority representation.
• Voting Rights Act of 1965 (VRA)
  • How do you prove a district is legally mandated under Section 2.
  • Bottom line: does the minority group have less opportunity have less opportunities to elect representatives of its choice? A demographer looks at this.
  • Senate factors: history of voting-related discrimination. Other factors that dilute the minority vote.
• California Voting Rights Act (CVRA)
  • Less costly, need not demonstrate the ability to draw
  • Only applicable in challenging at-large voting systems.
  • Requires pre-litigation demand letter. A jurisdiction then has a period of time to respond.
  • Caps pre-litigation fees at $30k.
• Voting History
  • 1790 - Immigrant (White) men
  • 1848 - Mexicans in acquired territories
  • 1870 - Black men - 15th amendment
  • 1920 - Women - 20th amendment
  • 1943 - Asian-Chinese immigrants - repeal of exclusion act
  • 1945 - Asian-Indian immigrants
  • 1952 - All Asian immigrants - capped immigration
  • 1965 - Protected right to vote - Voting Rights Act

02:50 - 30:40PM - Ellen Veomett (St. Mary’s) - slides

​The Efficiency Gap, Voter Turnout, and the Efficiency Principle

• In search of a metric to define gerrymandering?
  • Geometric tools and their issues
    • Boundary length to contained area (isoperimetric inequality). Issues with coastlines or other boundaries. Map zooming (Duchin and Tenner)
    • Compare district to some desired shape. Bad shapes are very similar to nice ones.
• Efficiency gap. A measure of partisan gerrymandering must satisfy two principles. S(D, delta) = seat share for party A, V(delta) = vote share for party A. G(D, delta) number such that the higher value means that the redistricting plan is more favorable to party A (this is the tool that does measurement).
  • EP1. No wasted votes
  • EP2. No not-wasted votes
• Facts/concerns about EG
  • If we consider only a single-district, it’s volatile
  • Seems like it wouldn’t work well for states with few districts
• EG doesn’t support EP2, or EP1 for competitive elections (i.e., doesn’t work for unequal turnout).
• For any rational numbers, 1/4 < V < 3/4 and 0 < S < 1, there exists an election outcome data with vote share V, seat share S and EG=0.
• Don’t compare EG across states or across different elections (historical time periods) for the same state.

03:50 - 04:40PM - Karin Mac Donald (SWDB and UC Berkeley) and Denise Hulett (MALDEF)

Implementation Challenges and Race Discrimination in the Era of Algorithms

• How is redistricting actually implemented?

06:00PM - 07:30PM - Colleen Mathis (AZ Commission Chair), Wendy Underhill (National Conference of State Legislatures Redistricting Coordinator), California Commissioners: Angelo Ancheta (D), Vince Barabba (R), Cynthia Dai (D), and Michelle DiGuilio (I)

Panel Discussion - Independent Commissions and Redistricting Reform

• Wendy
  • NCSL
  • Overview of states that use independent commissions or legislatures to handle redistricting.
  • Overview of redistricting commission adoption after each Census
  • Overview of states with citizens’ initiatives
  • Overview of states with legislature introduction of commissions.
  • Reform - Beyond Commissions
    • Mixed record on independent commissions
    • Each state: Criteria, principles, guard-rails (all available on website)
    • Transparency, accountability, citizen input
• Colleen
  • Overview of Arizona Independent Commission (created in 2000)
  • Citizen initiative to take away power from legislature and have an independent redistricting commission (Proposition 106).
  • 25% area of the land is Native American. Gerrymandered here to preserve the Hopi as a community of interest.
  • AZ is the only state with competitiveness as a criteria.
  • Five commissioners, ~6 million people, 2 Democrats, 2 Republicans, 1 Independent. Independent is chosen and is the chair.
  • Public hearing process to hear about criteria and the draft map. Very public website.
  • What does Independent mean? It means that it’s independent of the legislature. Provision in the Constitution for the legislature to remove a commissioner. Party line vote removed her from the position, but she was then reinstated by the State Supreme Court.
  • Map needs to pass preclearance by the Department of Justice.
  • AZ is now looking to expand the number of seats on the commission (possibly to include more independents).
  • “If you want a friend in redistricting, get a dog”
• Angelo
  • California Citizens Redistricting Commission
  • California: 5 Democrats, 5 Republicans, 4 Neither
  • Filled a lottery seat
  • Ranked constitutional criteria. Not included: competitiveness. Prohibitions: incumbency, candidacy, and party criteria.
• Michelle
  • Qualifications: impartial, appreciation for California’s diversity, analytical skills, no conflicts of interest.
  • Long extensive interview process with veto option over names by the state legislature. Lottery process and then selection.
  • Description of listening tour in CA. Decisions are final and cannot be appealed by the legislature.
• Vince
  • Census director under Nixon and Carter
  • Fairness in redistricting; as well as an overview of the process the committee undertook.
  • Listening tour in Marysville: definitely included a large community of interest (Sikh’s and Hmong), that otherwise would not have been uncovered and included without the tour (for at least one reason: religion is not included in the Census).
• Cynthia
  • Review: Assessing California’s Redistricting Commission (from McGhee)
  • 2-1 voter approval of commission’s work (via direct ballot)
  • CA independent redistricting has dramatically improved approval of the state legislature.

libdataframe is an in-progress library for performing in-memory, structured data analysis: essentially a column-store equivalent to sqlite for use in ad-hoc data analysis and visualization. We started project after finding no easy portable equivalent to Pandas or R dataframes.

Basic Goals:

• Easy-to-read structured datasets and analysis from some on-disk format, such as CSV.
• Robust and natural timeseries analysis
• Easy to decoding/encoding to other formats (CSV, HDF5, Parquet, Apache Arrow), but as separate libraries.
• Fast, performant, well-tested, and well-understood
• Written to C++, but portable to other languages: easy FFI for compiled and interpreted languages via C bindings (as a starting point: Python and Javascript)
• FFI-based interpreted language bindings should be usable from interactive programming environments such as Jupyter notebooks.

Distant, or Explicit Non-Goals:

• Avoiding crazy C++ template metaprogramming and smart expression template techniques
• No hipster technologies or languages
• Streaming writes or analysis on streams
• SQL standard, SQL interpreter interface
• Protocol Buffer, Thrift, JSON, etc. ingest interfaces
• Interop with other systems (i.e., Apache Arrow’s approach to zero-copy between differing Apache big data projects., etc.)

See the rest on Github.

sca_tools analyzes software scalability benchmarking experiments, specifically measurements of throughput as a function of applied load or concurrency. It is a Python library and set of command line utilities and that will help you:

• Model and quantify throughput bottlenecks in your application
• Capacity plan
• Compare performance benchmarks for regressions

sca_tools relies on on Neil Gunther’s Universal Scalability Law as a model and lmfit to perform model fitting. sca_tools differs from existing implementations (see Related Work) in a few ways:

• Emphasis on robust parameter estimation
• Support for propagation of parameter uncertanties when computing derived quantities, such as latency, queue size, etc.
• Support for experimental measurement uncertainty
• Written using Scientific Python libraries

See the rest on Github.

The Icewatch

03 January 2017 - Errera Channel, Antarctic Peninsula

Anchored at 64 degrees (and change) South, it never gets quite dark at midnight. The simmering sea in the bay, the sedimentary rock on land both near and far, and the sea ice all between are monotone black, grey, and white, penetrated by smoky fog. The only color off the deck, the occasional blue of the drifting glacial ice and the odd orange ice tip scraped along from the hull. The only sound, the cracking of distant glaciers and the lapping of sea water on the skirts of those drifting nearby.

Henry, a dog owner from the States, has seen enough Weddell seals at this point and is happy to be pacing about the mizzen in pajamas during a dangerous anchor watch. At 3 o’clock (starboard!), a cracking iceberg calves loudly into the sea, and three squeaking Gentoos, rudely interrupted from important penguin work, surface and purpoise off and out of sight. A thin layer of fresh snow covers the wooden poop deck rails, just bare from a balmy Austral Summer day.

At 9 o’clock (port!) from the poop deck, a low-lying, tentacled iceberg the size of a Toyota Prius is sliding its way towards the back of the hull. Gijs and I are hunched over the rail, eyes fixed three meters below us as its foot slips out of view of the hull, followed by its top, and then (and then!), a mild echoing thud of ice against steel.

“That’s not very nice there, Mr. Iceberg, eh?”.

Ten minutes later, our anchor watch shift is over, and Mike and Wayne are taking in the scene, the snow picking up and the ice hemming in. The Prius’ new momentum carries it into collision with another floater, settling within jumping distance from the lower sloop deck.

“Ok, guys, we are on iceberg watch tonight. If any of the icebergs come much faster at the ship, let us know. If that iceberg aft comes much closer to the height of the dolphinstriker or near the propellers, let me know. If that rock comes any closer to the ship,” gesturing from the wheelhouse at the small mountain ahead of us, “wake up Klaas.”

The scent of fresh baked bread (best eaten vigilantly in cold weather) from the galley’s exhaust pipe wafts by.

Originally published on the Bark Europa logbook.

The Perils of Codesharing Agreements

22 December 2016 - CDMX

Terminal 2. Dropped off M at the Aeromexico Terminal 2 and promptly went to the International Terminal 1, arriving two hours before departure.

Terminal 1, 12PM. Went to the checking desk of the wrong Latin American holding company, LATAM (my flight was with Transamerica Airlines). They were like, “we don’t have a 2PM flight to Lima, also, this is an Aeromexico flight, you need to go check in in the other terminal” (it’s true, flight had an AM call sign). The departure gate was still in Terminal 1.

Terminal 1, 12:05PM. In a panic, I run to an elevator and take it to street level, run to the taxi booth to get a taxi ticket and pay $135 pesos (US$6) to be driven back to the other terminal. On the way, the driver gets into an argument with an adjacent driver in Spanish (maybe they were talking smack), at which point, the other guy spits into the car, spraying me and my bag. I roll up the window. The driver and I have a surprisingly long conversation in broken middle school Spanish about why I’m going to the Aeromexico terminal (“hay todo en terminal dos, domesticos y internacionales”).

Terminal 2, 12:10PM. I’m still panicking and without coffee. I run to the Aeromexico terminals to try and check in, skip an offer of complementary coffee and what can only be described as a care package of free potato chips (welcome to Mexico). Everyone takes their sweet time with baggage and ten minutes pass. An agent at the counter looks at my Gmail itinerary print out and actually can’t figure out what carrier I’m taking or why it has a AM call sign. “We don’t even have any Airbus planes.” After some looking around she lists two different Peruvian airlines I should look into, and that they’re federated under Avianca. “You have plenty of time, an hour and 20 minutes! Just take the air train.”

Terminal 2, 12:40PM. Running is hard at Mexico DFs altitude, and seeing my heavy exhales, the guard at the airtrain lets me on. Two guys with the heaviest American accents I’ve ever heard are leaning over their camo bags (they did not vote for Hillary) and gawking at the tin roofs passing underneath the train. This is just like every airport train you’ve ever been on; the quesadillas are much closer though, but still out of reach.

Terminal 1, 12:55PM. Run to the Avianca check-in, and it seems like I’ve actually finally made it. A lady in a red jacket takes my passport and does a lot of computer things. I have no idea what she’s doing or what her UI looks like and it’s absolutely killing me, a silent drama playing out in the back of my mind. She looks and looks and looks and after much typing she goes over to some other people with my passport and words are exchanged. She comes back, speaking quickly (unlike everyone in CDMX) and then realizing that I can’t speak Spanish, indicates in English that she can book me on a flight to Bogota and that I’ll still be able to make my connection in BA. The original flight check-in is closing and the flight is full; no seats are available.

Terminal 1, 1PM. Her supervisor has decided that she can’t do this, because I came to the airport late (no sob stories about the fact that no one knows what Transamerica Airlines is. It turns out it’s known as TACA.) and issues me a standby ticket, saying I can go through immigration and hope that someone has bailed on their seat, a seat on a full international flight just before Christmas (“maybe they are stuck in the bathroom for something”).

Terminal 1, 1PM - 2PM. Whereupon someone at United says that they cant help me and hands me a real physical goddamn phone with the actual United customer service with that eternal Gershwin piano remix. I bounce around, nervously flicking Chrome seeing one-way tickets to USH from MEX. Lots of anxious phoning. Avianca can’t rebook because the ticket is owned by United, and United can’t rebook because they can’t endorse other airlines tickets (which turned out to be bullshit).

Terminal 1, 2:15PM. Holding a pending business class one-way flight on Copa, I make my way back to United, with the some more air travel jargon under my belt. Standing in line, an ANGEL DESCENDS FROM THE HEAVENS, crop cut and brusque, sounding like she grew up in Brooklyn but lived in DF, the weirdest accent ever. She has the air of someone who’s used to talking to the frustrated and helpless but still has a palpable passion for air travel, and ten minutes later, decides that I can fly back to the US and THEN to Argentina (“oh there are two airports in BA. I didn’t know that. That’s so cool!” “Wow, this is the reverse of you return flight. Neat!”). Lots of technical airline drama ensues, and at some point, I’m holding one real phone waiting so she can make another call on another phone so that two people can agree IRL about changing my ticket. Very analog. The guy who refused to help me earlier is sloshing about with his tail between his legs. They exchange harsh words, and a half hour later, exhausted from travel bureaucracy bloodsport and shuffling about on different phones, she returns with boarding passes: “Are you checking any baggage with us today?”

I haven’t hugged a stranger that hard in quite a while.