Sometimes we can learn a great deal about how to do things right from the times when things went wrong. Welcome to a new occasional series: The Bug Parade!
Gandhi attacks
The Civilization series is one of the most enduringly successful computer strategy games of all time. The games let you control a human civilization, starting in 4000 B.C. and guiding them into the 21st century. You control all aspects of your civilization, from military expansion to scientific advancement to keeping your citizens happy. Along the way, you encounter other civilizations and their leaders. You can go to war with these civilizations, make peace, manage alliances, and negotiate with their leaders. The series is one of the most incredibly addictive, just-one-more-turn strategy games ever created.
It all started in 1991 with the original game of that name, now informally known as Civ 1, created by legendary game designer Sid Meier. Nearly 30 years later, the original game still has a devoted fanbase, who not only still play the game but also dig deep into the game’s code to create fan mods to play and share.
Creating a game that people still enjoy after 30 years is an amazing achievement on its own, but it becomes absolutely mind-blowing when you realize that the entire game takes up just 2.6 megabytes of space, less than a single song stored in MP3 format.
Size comparison: the entire game (2.4 Megabytes) against a single MP3 (3.4 Megabytes)
Creating such a complex game in such a tiny amount of storage space necessarily meant that many compromises had to be made – and from one of these compromises came one of the most famous software bugs in history.
In 1991, the Internet had yet to penetrate into public consciousness, and so the idea of playing a game online was still new. Meier knew this, and knew that Civilization presented enough of a design and programming challenge already, so he decided to keep it firmly a one-player strategy game. And that meant that the quality of the game would depend on the quality of the player’s computer-controlled artificial intelligence (AI) opponents. If the AI opponents made good decisions, the single human player could respond to the challenges presented with decisions of their own.
But that was the challenge – how could the game developers make convincing AI for opposing world leaders while making the entire game fit into an absurdly low 2.6 megabytes? All opposing leader decisions where based on a set of simple formulas. Will Hammurabi offer a peace treaty? Check the formula. Will Mao Tse Tung launch a nuclear weapon? Check the formula. And to account for differences between leaders, the input to the formula was a simple scale from 1 to 5, ranging from most peaceful to most warlike: Mahatma Gandhi scored 1, Genghis Khan scored 5.
But there was another game design point to consider. One of the things that made Civilization so groundbreaking was that it had multiple ways to win. You can win by conquering the entire world, by winning the space race, or by having the highest score when the game ends in 2100 AD. To reward the player for sticking around for the modern era, – as well as to reflect historical experience – opponents become less aggressive. Specifically, each leader’s aggression rating drops by two.
But if aggression drops by two, what happens to Gandhi, who already had an aggression rating of one? Logically, his aggression rating should drop to -1, making him even more peaceful than he was already. But if logic prevailed, we wouldn’t be talking about this in a recurring series called The Bug Parade.
Remember that with only 2.6 megabytes to work with, programmers were desperate to save space, so they used every strategy they could think of to make their code use memory efficiently. Since they knew the aggression rating went from 1 to 5, they used only one byte of memory to store the value. One byte of memory can store values from 0 (in binary, 00000000) to 255 (in binary, 11111111). You can probably already see the problem with this.
If the program could only store values from zero to 255, and it had to calculate 2-1, what was it supposed to do? The only thing that made any sense at all was for the odometer to turn over, for the calculation to wrap around, and to return the result 255.
Illustration of an integer underflow error
Thus, as soon as the player discovers Democracy, Gandhi’s aggression jumps to 255.
Out of five.
And so as soon as Gandhi gains the ability to launch nuclear missiles, he immediately launches them at everyone on the next turn, and on every turn after that until he is dead or everyone else is dead.
The bug was never caught in testing, but of course as soon as the game shipped, millions of players discovered it, and it became legend. The bug was quickly fixed in the sequel, but by that point “Nuclear Gandhi” had become enshrined in popular culture.
Memes weren’t really a thing in 1991, but “Nuclear Gandhi” was an inside joke among computer gaming fans for literally decades. And as soon as memes became a thing, Nuclear Gandhi memes became a thing too. Here’s a mashup of Nuclear Gandhi with another famous meme, Der Untergang subtitles:
Coding the game in such a way as to allow a variable underflow error wasn’t even really a mistake, just a design decision that had consequences that were unexpected – and hilarious.
But that’s the nature of the parade – the bugs show up when you least expect them.
I hope you enjoyed this light-hearted introduction to where software bugs come from and what they can do – because when the Bug Parade returns, the next one will be pure nightmare fuel.
Graphs day 321, pandemic day 328, day 400 since the first cases were diagnosed. That’s right, it has now been 400 days since Dr. Li Wenliang officially diagnosed the first few cases of the new disease now called Coronavirus Disease 2019 (COVID-19). Just over a month later, Dr. Li had died of the disease.
Today’s sad totals:
Total cases of COVID-19 diagnosed worldwide: 102,579,774
Total deaths: 2,206,445
Remember how we said a few months ago that as we get better at diagnosing less severe cases, the case fatality rate (CFR) would get closer to the biological value, which was likely around 2 percent? The case fatality rate is currently 2.2 percent.
Here are the usual graphs, with countries divided into the same categories as on Tuesday. Countries are color-coded and labeled by name. The thickness of the lines and the sizes of the labels correspond to the case fatality rate in the country – thicker lines and larger labels indicate countries where a larger percentage of COVID-19 cases have led to death. Click on any of the images for a larger version.
Countries that have consistently gotten it right
These countries have consistently kept their per-capita cases of COVID-19 at very low levels.
Countries where COVID-19 was stayed mostly contained throughout the pandemic
When we last checked in last Tuesday, Japan seemed to be in a phase of an increased number of cases. Today, it is much clearer that cases are once again declining in Japan, after having peaked at 40 new cases per million people per day about two weeks ago.
Plenty of people have told me they do not believe the case and death numbers coming out of China. That’s fair enough; there are lots of reasons to not believe China.
But do you believe the case and death numbers coming out of New Zealand, Australia, India, and Japan?
Countries where COVID-19 is currently under control(-ish)
Countries where COVID-19 was bad but are now sort of under-control-ish
When we looked on Tuesday, there seemed to be a slight increase in Chile, Peru, and Canada. The daily case load definitely seems to have decreased in Canada, and seems to be leveling off in Chile and Peru.
Countries moving in the right direction(-ish)
These countries, all in Europe, experienced nightmarish COVID-19 disease waves in the fall – some of the worst in the world – bur fortunately for now have seen cases stabilize at a lower level.
Regions where cases are still high but
The good news is that these three European countries seem to be doing well. Cases continue to decrease in Belgium, Switzerland, and Italy.
The bad news is that Italy is experiencing a wave of sometimes-violent protests against the necessary restaurant closings. If the current protests lead to an increase in COVID-19 cases and deaths in Italy over the next two to four weeks. Obviously I hope they will not, but nature cares not for our hopes. I’ll let you know how it goes either way.
Countries where the pandemic is still awful
Lastly, the countries where the pandemic is still infecting and killing lots and lots of people: Spain, the United States, the United Kingdom, France, Sweden, and Brazil.
Regions experiencing a second wave of cases
Fortunately, even in these countries, cases are beginning to decrease; although in the meantime, Spain has moved ahead of the U.S. into first place on this graph in terms of cases per day.
Methods
Sadly, Corona Data Scraper has stopped reporting data in the way I was using it, so I have gone back to using the data from the JHU Coronavirus Resource Center. You can download the data from their GitHub repository. You are welcome to use my Excel spreadsheet (version 9.2).
Graphs day 315, pandemic day 322, day 392 since the first cases were diagnosed.
Between The Most Important Election Of Our Lifetimes (TM), the holidays, and preparing for my first-ever virtual American Astronomical Society meeting, it’s been months since I’ve kept my promise to create daily graphs of the global COVID-19 pandemic.
COVID-19 has been busy, and the resulting graphs are almost unbearably sad.
Total cases of COVID-19 diagnosed worldwide: 99,725,495
Total deaths: 2,140,073
There is no silver lining here, I might as well just show you the numbers. In the immortal words of Stringer Bell:
https://youtu.be/A90TYC0BiwE?t=25
Let’s again take a look at how some different countries are handling the COVID-19 pandemic. As a reminder, graphs show the history of the pandemic from late January 2020 to today – just about exactly one year. Countries are color-coded and labeled by name. The thickness of the lines and the sizes of the labels correspond to the case fatality rate in the country – thicker lines and larger labels indicate countries where a larger percentage of COVID-19 cases have led to death. Click on any of the images for a larger version.
Countries that have consistently gotten it right
These countries have consistently kept their per-capita cases of COVID-19 at very low levels.
Countries where COVID-19 was stayed mostly contained throughout the pandemic
Last summer, we were worried about a wave of new cases in Australia, but looking back on it, per capita cases in Australia are barely even visible on this graph. Joining them on this graph are India (happy Republic Day BTW, enjoy your republic while it lasts!), New Zealand, Japan – and China, even Hubei province where the whole pandemic began.
What do these countries have in common?
You already know the answer to that, even if you like to pretend that you don’t.
Countries where COVID-19 is currently under control(-ish)
Countries where COVID-19 was bad but are now sort of under-control-ish
Last spring and summer we were watching things get bad in Qatar, Chile, and Peru, but they all appear to be doing OK right now – although cases are ticking back up in Chile and Peru. Cases were increasing in Canada but they are now decreasing again.
Countries moving in the right direction(-ish)
Regions where cases are still high but
While we were away, the pandemic got very, very bad in Europe, particularly Belgium. Fortunately, cases are headed down, but they are still far higher than they should be.
Countries where the pandemic is still awful
Lastly, the countries where the pandemic is still infecting and killing lots and lots of people: the United States, Spain, the United Kingdom, Sweden, France, and Brazil.
Regions experiencing a second wave of cases
Fortunately, cases are decreasing everywhere except Spain.
I’ll be here again tomorrow. I’ve promised to keep doing this until the pandemic ends. I hope the pandemic ends.
Methods
Sadly, Corona Data Scraper has stopped reporting data in the way I was using it, so I have gone back to using the data from the JHU Coronavirus Resource Center. You can download the data from their GitHub repository. You are welcome to use my Excel spreadsheet (version 9.2).
I closed Monday’s review of my House predictions with an observation that I amazingly haven’t seen anyone else make: the Republicans actually came very close to capturing control of the House of Representatives. How close?
The current seat count in the U.S. House of Representatives is Democratic 220, Republican 202 – likely to go up to Republican 205 once the final three seats are decided. That means that if eight elections had gone Republican instead, the count would be Republican 213 Democratic 212.
The table below shows the eight closest races in which the Democratic candidate won. Adding up all the margins gives the minimum number of extra Republican votes that would have been required to give control of the House to the Republicans – provided of course all the votes had come in exactly the right places.
District
Democratic
D votes
Republican
R votes
Margin
Pct margin
IL-14
Underwood
203,195
Oberweis
197,818
5,377
1.3%
IA-3
Axne
219,205
Young
212,997
6,208
1.4%
VA-7
Spanberger
230,893
Freitas
222,623
8,270
1.8%
PA-17
Lamb
221,547
Parnell
211,951
9,596
2.2%
WI-3
Kind
199,870
Van Orden
189,524
10,346
2.7%
MN-2
Craig
204,031
Kistner
194,466
9,565
2.4%
MI-11
Stevens
226,128
Esshaki
215,405
10,723
2.4%
TX-15
V. Gonzalez
115,594
De La Cruz-Hernandez
109,005
6,589
2.9%
The eight closest races with a Democratic winner
The final count: 66,674. That’s not very many voters.
As happens every even-numbered year, all 435 seats in the House of Representatives were up for a vote. I predicted the outcomes of them all on a map, and discussed detailed predictions in part 2 and part 3). How did I do in predicting the results of those elections?
What We Still Don’t Know
The answer is, “we don’t know for sure yet,” because there are four results still unknown.
The weird one
Louisiana-5: As with the two Senate Elections in Georgia, we don’t know the winner in Louisiana’s 5th congressional district, which covers the whole northeastern quarter of the state and includes Monroe and Alexandria. Like Georgia, Louisiana uses a “jungle primary” system where all candidates appear on the ballot regardless of party. If one candidate gets 50 percent of the vote, they win the election; if not, the top two candidates face off in a runoff. The top two finishers were Republicans Luke Letlow (33.1% election day vote share) and Lance Harris (16.6%), leaving us in the weird position of knowing that a Republican will represent LA-5, but we still don’t know which one.
The exciting ones
And there are also three house races where both winner is still unknown, and might be from either major party:
California-21 (the southern third of the Central Valley and the east side of Bakersfield): with 95 percent of ballots counted, Republican former representative David Valadao (83,564 votes; 50.5%) leads Democratic incumbent T.J. Cox (81,946; 49.5%). CURRENT RESULTS
California-25 (rural-ish Los Angeles and Ventura Counties, including Palmdale, Santa Clarita, and the San Fernando Valley): with 87 percent of ballots counted, Republican incumbent Mike Garcia (169,060 votes; 50.1%) leads his former special election Democratic opponent Christy Smith (168,660; 49.9%). That’s a difference of just 400 votes, smaller than my high school graduating class. CURRENT RESULTS
Iowa-2 (the southeastern quarter of Iowa, including Davenport and Iowa City): the most mind-blowing one of all… with all votes counted and currently being recounted by hand, Republican Marianette Miller-Meeks (196,862 votes; 50.006%) leads Democratic candidate Rita Hart (196,815; 49.994%) BY JUST 47 VOTES. That’s literally less people than would fit on a single bus! CURRENT RESULTS
Iowa’s second district race is headed to a recount, but recounts almost never change election results, so the most likely outcome is a ridiculously narrow win by Miller-Meeks. What is especially remarkable about this race is that neither I nor anyone else had this on our radar as a close race; Hart was expected to win easily. But so it always goes on election night – there is always at least one race that defies expectations. That’s the nature of trying to predict 435 different things; enough opportunities for outcomes to occur virtually guarantees that there will be a rare outcome somewhere, somehow.
So there are those four uncertainties, but that still leaves 431 certainties. How did I do in my predictions?
Hits and Misses
My top line result for the House of Representatives was that the party count would be a good result for the Democratic Party: 237 seats going Democratic and 198 Republican, for a net swing of five seats Democratic.
In reality, the Democratic majority remains, but the House elections feel very much like a win for the Republicans. Not including the three seats yet to be decided, the count is 230 Democratic, 202 Republican. Republicans have a narrow lead in all three undecided races, meaning the result most likely result is a swing of seven seats in favor of the Republicans.
I have some ideas about why the 2020 House and Senate elections were so good to the Republicans, but that’s a topic for a different post. What matters here is: how right or wrong were my predictions? And for evaluating my predictions, the important question is not the swing, but rather the individual election results.
Here is my the map of the outcome of my predictions. Colors indicate the party of the winning candidate, using the traditional yet arbitrary red for Republican and blue for Democratic. Districts whose calls I missed are shown with their winner’s name in BIG BOLD ALL-CAPS.
Checking my predictions for the 2020 U.S. House of Representatives elections. Click for a larger version.
The map shows that my predictions were correct for 417 of the 432 decided House seats. That’s a 96.5 percent success rate, although it’s less impressive than it sounds when you realize how gerrymandering has made most House elections completely uncompetitive. It’s also less impressive when you realize the systematic offset: I mistakenly predicted one Republican win in a seat that went Democratic, and FOURTEEN Democratic wins in seats that went Republican. If Garcia holds on in California-25 and Miller-Meeks in Iowa-2, the latter number goes up to SIXTEEN.
Here are the districts where I got it wrong. Click the district names to see the district’s Wikipedia page, which includes a map; click on RESULTS see the results from the most official source I could find (and many thanks to the websites of the Secretaries of State of California and Utah for making their official state counts so easy to find and link to):
California-39 (eastern suburbs of LA, including Fullerton and Yorba Linda): Republican state assembly woman Young Kim (173,312 votes; 50.6 percent vote share) defeated Democratic incumbent Gil Cisneros (169,089; 49.4 percent). RESULTS
California-48 (coastal Orange County, including Huntington Beach and Laguna Beach): Republican county councilwoman Michelle Steel (201,660 votes; 51.1 percent vote share) defeated Democratic incumbent Harley Rouda (193,305; 48.9 percent). RESULTS
Colorado-3 (the western half of Colorado, including Aspen, Grand Junction, and Pueblo): Republican bar owner and gun rights acitvist Lauren Boebert (220,502 votes; 51.4 percent vote share) defeated Democratic state representative Diane Mitsch Bush (193,980; 45.2 percent). RESULTS
Florida-26 (southern Miami-Dade County and the Florida Keys, including Homestead and Key West): Carlos Gimenez, the Republican county commissioner of Miami-Dade County (177,223 votes; 51.7 percent vote share) defeated Democratic incumbent Debbie Murcasel-Powell (165,407; 48.3 percent). RESULTS
Florida–27 (cities of Miami, Miami Beach, and Coral Gables): in another race that was absolutely not at all on anyone’s radar, Spanish-language TV news host Maria Elvira Salazar (176,141 votes; 51.4 percent vote share) shocked long-revered Democratic incumbent Donna Shalala (166,758 votes; 48.6 percent). The unexpected result, like Trump’s unexpected advantage in Miami-Dade County, is further evidence that that Trump’s messaging resonated in Miami. RESULTS
Georgia-7 (northeastern suburbs of Atlanta, including Duluth, Norcross, and Suwanee): The Democratic 2018 candidate, Professor Carolyn Bourdeaux (190,900 votes; 51.4 percent vote share) defeated emergency room doctor and Marine Corps veteran Rich McCormick (180,564; 48.6 percent). I predicted this race would be within-one-percent close, but I picked the wrong winner. It was the only race where I pick d a Republican and the seat went Democratic. RESULTS
Indiana-5 (city of Indianapolis and its northern suburbs, including Carmel and Marion): In another election I predicted would be within-one-percent close – but was not – Republican state senator Victoria Spartz (208,212 votes; 50.0 percent vote share) defeated Democratic former state representative Chrstina Hale (191,226; 45.9 percent). RESULTS
Iowa-1 (northeastern quarter of Iowa, including Cedar Rapids and Dubuque): in another race that I thought would not be at all competitive, I was blindsided by a loss for the Democratic incumbent. Republican former Cedar Rapids TV news anchor Ashley Hinson (212,088 votes; 51.3 percent vote share) defeated Abby Finkenauer (201,347; 48.7 percent). RESULTS
New Mexico-2 (the southern part of the state, including the southern part of Albuquerque, Las Cruces, and Roswell): a rematch of the 2018 election goes the other way, as Republican state representative Yvette Herrell (142,169 votes; 53.8 percent vote share) defeated Democratic incumbent Xochtil Torres Small (122,314; 46.3 percent) by a surprisingly large margin. RESULTS
New York-2 (the South Shore of Long Island, including Islip, Massapequa, and Ronkonkoma): New York state assemblyman Andrew Garbarino (164,272 votes; 56.4 percent vote share) won the election to replace the retiring senior Republican Peter King, by defeating Democratic challenger Jackie Gordon (124,071 votes; 42.6 percent). RESULTS
New York-11 (Staten Island and far southwestern Brooklyn): State assemblywoman Nicole Malliotakis (136,382 votes; 56.4 percent vote share) returned New York City’s most conservative district to Republican control, defeating one-term Democratic incumbent Max Rose (99,224 votes; 41.1 percent vote share). RESULTS
New York-22 (the bridge of the nose, including Binghamton and Utica): in a rematch of 2018 with a different result, Republican attorney and publisher Clauida Tenney (149,769 votes; 50.5 percent vote share) unseated Democratic incumbent Anthony Brindisi (140,644; 47.4 percent). RESULTS
Utah-4 (rural central Utah plus the southern edge of Salt Lake City because gerrymandering): As part of 2018’s “blue wave,” Ben McAdams made this the most Republican-leaning district with a Democratic representative – but like the Alabama Senate, it was always going to be tough to hold the seat. Republican candidate and former Jets and Raiders American football safety Burgess Owens (179,688 votes; 47.7 percent vote share) defeated McAdams (175,923; 46.7 percent). Someday I will look up former American football players in Congress and decide an all-star team, but this post is already long and is about to get longer. RESULTS
Margins of victory
District
Margin
NJ-7
1.2%
TX-24
1.3%
UT-4
-1.0%
CA-39
-1.2%
CA-48
-1.2%
MN-1
3.0%
TX-7
3.3%
NE-2
4.3%
AZ-6
4.4%
MI-3
6.0%
MO-2
6.4%
TX-21
6.5%
NC-8
6.6%
FL-26
-3.4%
TX-22
6.9%
OH-1
7.2%
NY-22
-3.8%
TX-23
-4.0%
IL-13
8.8%
AK
9.0%
GA-6
9.2%
AR-2
10.8%
MT
12.8%
MN-7
13.5%
MI-6
15.7%
NY-11
-15.8%
Elections I predicted would be close, and their actual margins
As I had discussed when I evaluated my Senate picks on Saturday, I predicted not only the winner of each contest, but also which would be close (within 5% vote share) and very close (within 1% vote share). How did I do?
I predicted that 26 races would be close – decided by a vote share of more than 1 percent but less than 5 percent. I looked up the actual vote share margin of those 26 races, defined as the percent vote share for the winning candidate minus that for the losing candidate. The table to the left (or above if you’re on a small screen) shows the results of those 26 races.
Because of votes for third-party candidates, the vote shares do not always add up to 100 percent; but that doesn’t matter because I only considered the difference. For a few of the close races I predicted, I missed not only the margin but also the winner. Those races are described in more detail in the Hits and Misses section above, and I somewhat arbitrarily decided to double the penalty for missing the margin. Of my 26 predicted close races…
NINE were actually close, decided by between 1 and 5 percent once I applied the double-penalty for picking the wrong winner. The one that I missed most hilariously was New York-11, described above, where I predicted a close win for Max Rose but Nicole Mallonakis won by a shocking 15.8 percent.
What about the races that I predicted would be very close, within one percent vote share or less? The table to the left (or below if you’re on a small screen) show the results.
District
Margin
CA-25
-0.2
CA-21
1
OK-5
4.2
VA-5
5.2
GA-7
-2.8
NJ-2
5.7
PA-10
6.6
IN-5
-4.1
CO-3
-6.2
NY-24
14.1
NY-2
-13.8
Elections I predited would be very close, and their actual margins
Of the 11 races that I predicted would be very close, two actually were very close – the two in California that are so close we don’t yet know who won. One more (OK-5) was within 5 percent. The other eight were decided by even more than 5 percent. Most hilariously wrong were the elections in NY-2 and NY-24, where I picked the wrong candidate, and the other candidate won by 13.8 percentage points.
Correctly calling of 26 close and 2 of 11 very close races is not great, but even that doesn’t tell the full story, because there were races that I did not expect to be close that turned out to be close. The most obvious example of a miss to that side was Iowa-1, where I picked incumbent Abby Finkenauer to win easily, but actually she lost by 3.6 percentage points. In addition to her and Donna Shalala in Florida-27 (who lost by 2.8 percentage points), there were an additional 17 where my predicted candidate won, but by less than 5 percentage points. Those unexpectedly close races are shown in the table below.
What do all 19 of these districts have in common? They were all occupied by Democratic representatives. Two of them have changed parties and are now represented by Republicans. The other 17, like the House as a whole, remain Democratic, but by a much smaller margin than expected.
District
Margin
IA-2
-0.0
IL-14
1.4
IA-3
1.4
VA-7
1.8
PA-17
2.0
WI-3
2.0
MN-2
2.2
MI-11
2.4
TX-15
2.9
NV-3
3.0
AZ-1
3.2
MI-8
3.6
PA-8
3.6
WA-8
3.6
PA-7
3.8
IL-17
4.0
NV-4
4.9
FL-27
-2.8
IA-1
-3.6
Races that were unexpectedly close, with their actual margins
That was supposed to be the end of this post – an evaluation of how I did in predicting the 2020 House elections. Not great, but not bad for a first attempt. I’ll take it.
But this is not the end of this post, because in fact, doing this comparison has made me realize something that I hadn’t realized before, and I haven’t heard anyone else observe either: the Republicans came very close to winning the House. Once all the seats are decided, it looks like the balance of power will be Democratic 220, Republican 205. If eight elections had gone the other way, the balance would be Republican 213, Democratic 212.
How many additional Republican votes would it have required to swing the eight closest elections in their favor? Those are the districts in the second through ninth row of the table to. the right (or above if you’re on a small screen), from Illinois-14 to Texas-15. I’m calculating that number now, and I’ll let you know on Wednesday.
And if there is one ironclad rule in American elections, it’s that the President’s party suffers significant losses in Congress in the first midterm election. Which means that there is a good chance that the Republicans will control both houses of Congress in 2022. What impact would that have on our system?
Hang on folks, because I think I found a game-breaking glitch in democracy.
