Website Traffic

Meta page describing editing activity, traffic statistics, and referrer details, primarily sourced from Google Analytics (2011-present).
personal, statistics, meta, shell, R, Bayes, Google
2011-02-032020-07-05 in progress certainty: certain importance: 3

On a semi­-an­nual basis, since 2011, I review web­site traf­fic using Google Ana­lyt­ics; although what most read­ers value is not what I val­ue, I find it moti­vat­ing to see total traf­fic sta­tis­tics remind­ing me of read­ers (writ­ing can be a lonely and abstract endeav­our), and use­ful to see what are major refer­rers. typ­i­cally enjoys steady traf­fic in the 50–100k range per mon­th, with occa­sional spikes from social media, par­tic­u­larly Hacker News; over the first decade (2010–2020), there were 7.98m pageviews by 3.8m unique users.

For back­ground about the goals, tools, design, and imple­men­ta­tion of, see the page


October 2010–February 2011

My edit­ing activ­i­ty, as gen­er­ated by darc­s-­graph:

Plot of patch cre­ations (y-ax­is) ver­sus date (x-ax­is): Octo­ber to Feb­ru­ary


“An audi­ence, even an audi­ence of one, is always to be trea­sured and respect­ed.”

Adal­ric Brandl (“Uhl Eharl Khoehng”, Patri­cia A. Jack­son)

Pop­u­lar­i­ty-­wise, reports that over the 150 days between 2010-10-01 and 2011-02-28, there were 4,346 page-views (av­er­age 30/day):

Plot of page-hits (y-ax­is) ver­sus date (x-ax­is)

The most pop­u­lar pages were1:

  1. : 1,180
  2. : 644
  3. : 241
  4. : 108
  5. : 104
  6. : 101
  7. : 96

The rank­ings are not as I would pre­fer (I imag­ine Inter­net archivist feels much the same way about ), but it’s pretty clear that peo­ple enjoy my more prac­ti­cal arti­cles the most.

February 2011–July 2011

darcs-graph for this peri­od:

Plot of patch cre­ations (y-ax­is) ver­sus date (x-ax­is): Repos­i­tory cre­ation to July


“Stream­ing in the wind / the smoke from Fuji / van­ishes in the sky; / I know not where / these thoughts of mine go, either.”

the monk ( XVII: #1615)

Google Ana­lyt­ics reports that over the 124 days between 2011-02-28 and 2011-07-02, there were 42,410 page-views (av­er­age 342/day):

Plot of page-hits (y-ax­is) ver­sus date (x-ax­is)

The most pop­u­lar pages rank­ing changed con­sid­er­ably; while the DNB FAQ main­tained its pre-em­i­nent pop­u­lar­i­ty, 3 new pages bumped out ‘Links’, ‘Spaced rep­e­ti­tion’, ‘The Melan­choly of Kyon’, and ‘Haskell Sum­mer of Code’. I am a lit­tle sur­prised that my 2 Death Note essays seemed to’ve struck a chord, and even more sur­prised that my sloppy & ran­dom & un-rig­or­ous notes about nootrop­ics would be con­sis­tently pop­u­lar:

  1. DNB FAQ: 10,406
  2. : 4,189
  3. Modafinil: 3,231
  4. : 2,633
  5. : 1,779
  6. : 1,366
  7. Nootrop­ics: 2,056
  8. : 706


“They accu­mu­late / but there are none to buy them— / these leaves of words / pil­ing up like wares for sale / beneath the Sumiyoshi Pine.”

; (#180, “Famous Mar­ket Town”, ; trans. Steven D. Carter)2

As a writer, I desire feed­back. I also want to feel that my work has been of use to peo­ple. So while it would be nice if the world beat a path to my web­site, I rec­og­nize that I have to put some effort into mar­ket­ing my work. I’ve tried a num­ber of meth­ods.

  1. Wit­coin: I sub­mit­ted any num­ber of fairly pop­u­lar arti­cles but my total Wit­coin traf­fic over this period was 132 vis­it­s—a traf­fic total I could have got­ten with one slightly pop­u­lar link on Red­dit or a few links in com­ments. While I did­n’t lose any Bit­coins (be­cause my reg­is­tra­tion was funded by Kiba’s dona­tion of $20 and actu­ally prof­ited $55), I have spent at least 6 hours fig­ur­ing out how to use Wit­coin, sub­mit­ting arti­cles, respond­ing to com­ments, and vot­ing. Not the best use of time.
  2. Google : ini­tially dis­ap­point­ing, with after 3010 impres­sion, there were still no clicks! It was funded by the $129 coupon for sign­ing up for Google’s Web­mas­ter Tools. Inter­face is decent given com­plex­ity of task, but deeply frus­trat­ing to have to wait many weeks for the DNB FAQ and Modafinil ads to be approved or reject­ed. Final­ly, almost in June, the DNB FAQ ads were approved and the modafinil ads reject­ed. From 23 March to 2011-07-02, I paid $27.20 for 98,900 impres­sions yield­ing 63 clicks through. (Those vis­i­tors only spent an aver­age of <1.5 min­utes on the site, too.) Again, not a great invest­ment of time.
  3. : with just 3 arti­cles ‘stum­bled’ (in­cluded in the data­base), specif­i­cally DNB FAQ, & Nootrop­ics, Stum­ble­Upon was respon­si­ble for 161 vis­its or 2.77% of all traf­fic in the period I looked at. How much traf­fic could I expect with 30 or 40 arti­cles stum­bled? Quite a bit. SU has no ‘front page’ like other social news aggre­ga­tors so traf­fic is more of a trickle than flood; nev­er­the­less, “Death Note End­ing” clicked with SU read­ers and I got >500 read­ers out of it in a day or two. In total over this time, SU drove 2,257 vis­its. SU tended to give a pretty steady 30–50 vis­its a day with rare spikes when an arti­cle clicked. The down­side is that after look­ing at SU com­ments and at how much time they spend on pages3, I have to agree with Arvind Narayanan’s “Stum­ble­Upon Con­sid­ered Harm­ful”—SUers do not want qual­ity con­tent but quick con­tent, for the dopamine boost.
  4. : “Girl Scouts and good gov­er­nance” made it to the front page, result­ing in 1,727 vis­its & set­ting traf­fic records (it is that giant spike in the traf­fic graph), but appar­ently min­i­mal view­ing of other pages. Fur­ther, while I seem to get a mod­est amount of Red­dit traf­fic from even unsuc­cess­ful sub­mis­sions, HN sub­mis­sions will sink with­out a trace. Kiba calls Hacker News a ‘lot­tery’, but it seems to be one worth play­ing.
  5. Less­Wrong is a nat­ural place to post many of my writ­ings. And per­haps unsur­pris­ing­ly, LW is my sec­ond-largest source of traf­fic, com­ing in after SU with 1,857 vis­its. While few of my sub­mis­sions get upvoted all that high­ly, most of them drove a fair amount of traf­fic even in the Dis­cus­sion ghet­to. (Link­ing in com­ments also dri­ves a sur­pris­ing amount of traf­fic over long peri­ods to my prac­ti­cal arti­cles like on n-back or mela­tonin.) At some point I hope to have a good Arti­cle and see how much of a dis­par­ity there is.

July 2011–December 2011

Res audita per­it, lit­era scripta manet.


darcs-graph for this period (in­clud­ing 2012-01-01):

Plot of patch cre­ations (y-ax­is) ver­sus date (x-ax­is): July 2011 to 2012-01-02

I ran into a cool post by Christo­pher Done on a tool that does detailed analy­sis of patch pat­terns on a Git repos­i­to­ry, Git­Stats, and this spurred me to cre­ate a Git mir­ror of using darc­s-­to-git. Git­Stats pro­duces a whole bun­dle of graphs and fig­ures, some of which I found sur­pris­ing. (I did not expect to see a large spike on Wednes­day and rel­a­tively few patches on Sat­ur­day, or a spike around 5 PM, as opposed to the early morn­ing.) I think I will update the Git­Stats out­put with each out­put, as a (large) adjunct to the darcs-graph plots.


“Ah! let not Cen­sure term our fate our choice, / The stage but echoes back the pub­lic’s voice; / The dra­ma’s laws the dra­ma’s patrons give, / For we that live to please must please to live. / …prompt no more the fol­lies you decry, / As tyrants doom their tools of guilt to die; / ’Tis yours this night to bid the reign com­mence / Of res­cu’d Nature, and reviv­ing Sense; / …Bid scenic Virtue form the ris­ing age, / And Truth dif­fuse her radi­ance from the stage.”

(“Pro­logue at the Open­ing of Drury Lane The­atre”)

Google Ana­lyt­ics reports, that over the 185 days between 2011-07-02 and 2012-01-02, there were 191,015 page-views (av­er­age 1,032/day) by 79,346 vis­i­tors for a total of 115,585 vis­its (av­er­age 624/day). This is bet­ter than I expected and makes me won­der about my pre­dic­tion for <2000 aver­age daily vis­its by 2013 (but it still seems unlikely traf­fic will triple over the next year).

Plot of page-hits (y-ax­is) ver­sus date (x-ax­is)

The main change in page pop­u­lar­ity did not sur­prise me; when I was writ­ing “Silk Road 1”, I knew it would almost cer­tainly be pop­u­lar given how very pop­u­lar the Gawker arti­cle was but also how lack­ing in prac­ti­cal details it was, and I also sus­pected that “Bit­coin is Worse is Bet­ter” would be fairly pop­u­lar as it argued an inter­est­ing and con­tro­ver­sial the­sis (the orig­i­nal and pro­moted ver­sion is hosted on, so its hit-­count ought to be low). I’m sur­prised at how much the page still gets; a sur­pris­ing num­ber of peo­ple must either visit the main page after read­ing another arti­cle or click on my var­i­ous blog com­ments.

  1. Silk Road: 62,167
  2. DNB FAQ: 25,541
  3. home/main page: 16,967
  4. Modafinil: 10,437
  5. Nootrop­ics: 10,219
  6. Spaced rep­e­ti­tion: 6,570
  7. Bit­coin is Worse is Bet­ter: 4,297
  8. Links: 3,490

More inter­est­ing is the other sig­nals of pop­u­lar­i­ty: gave me a free set of head­bands (worth ~$63) because they liked my , a soft­ware engineer/manager con­tacted me to see about recruit­ing me, ThinkGum offered me some of their epony­mous prod­uct for my Nootrop­ics page, and my request for Bit­coin dona­tions has paid off a lit­tle with a few dona­tions $2$20 and one gen­er­ous dona­tion of $398 (worth a bit upwards of $127 at the time; I spent it on modafinil). (This is all intrin­si­cally help­ful but I value it mostly because money speaks louder than word­s.)


I’ve done rel­a­tively lit­tle in this period com­pared with the pre­vi­ous peri­od:

  1. I aban­doned Wit­coin not long after my exper­i­ment with it; and now Wit­coin is dead, pend­ing a pos­si­ble open-­sourc­ing of the code­base.
  2. My AdWords credit is mostly expired. For some rea­son, my click­-through rates kept drop­ping.
  3. Stum­ble­Upon remains a good traf­fic source (1,430 vis­it­s). I con­tinue to ‘stum­ble’ my new arti­cles when I remem­ber to do so.
  4. Hacker News was respon­si­ble for a great deal of my traf­fic in this period (4,175 vis­it­s). Most of it was not my doing, how­ev­er—when­ever I sub­mit links, they do poor­ly.
  5. Less­Wrong remains a major traf­fic dri­ver (6,961 vis­it­s); I con­tinue to see a lot of refer­rals from old posts and com­ments. Nor do all the links seem to be per­ceived neg­a­tively or as self­-pro­mo­tion by the LW com­mu­ni­ty: I posted an arti­cle describ­ing site updates and the arti­cle was received well to my sur­prise, elic­it­ing very favor­able reviews of my writ­ings in gen­er­al. That was nice.
  6. For this peri­od, I did spend a lit­tle more effort sub­mit­ting stuff to Red­dit; and I was hand­somely rewarded with the Silk Road sub­mis­sion sky­rock­et­ing and become one of the all-­time most pop­u­lar arti­cles in the Bit­coin sub­red­dit. Between that and my nootrop­ics arti­cles, Red­dit sent me 20,842 vis­its.

The largest traf­fic sources are Google at 36,625 vis­its and direct/no-referrals at 24,118 vis­its. As I have no idea how to improve these two fig­ures, I ignore it. I write good con­tent, sub­mit it places, sup­ply meta­data, and abide by my hack­erly prin­ci­ples; I hope that that is all the SEO I need.


January 2012–July 2012

“Uproot your ques­tions from their ground and the dan­gling roots will be seen. More ques­tions!”

Frank Her­bert (“Men­tat Zen­sufi admo­ni­tion”, )

darcs-graph for this period (2012-01-03–2012-07-02):

Plot of patch cre­ations (y-ax­is) ver­sus date (x-ax­is): Jan­u­ary 2012 to 2012-07-02


“If it were not for the intel­lec­tual snobs who pay—in solid cash—the trib­ute which philis­tin­ism owes to cul­ture, the arts would per­ish with their starv­ing prac­ti­tion­ers. Let us thank heaven for hypocrisy.”

(“Indian & Burma”, Jest­ing Pilate, 1926)

The Ana­lyt­ics report records traf­fic over those 182 days as being sub­stan­tially increased: to 570,997 page-views (av­er­age 3,137/day) - almost 5 times the pre­vi­ous six month peri­od—by 268,031 unique vis­i­tors in 366,301 vis­its (av­er­age 2012/day). If this rate con­tin­ues, I will likely lose my traf­fic pre­dic­tion (which does­n’t bother me very much). In par­tic­u­lar, the life­time total page-views is now at 809,000, which would seem to imply that I will break the mil­lion page-view mark by the next update! I would be very pleased by that mile­stone.

Plot of page-hits (y-ax­is) ver­sus date (x-ax­is), early 2012

The pop­u­lar­ity rank­ing remains mainly the same. 2 dif­fer­ences from the past stand out: the sud­den pop­u­lar­ity of the Zeo and “Death Note Anonymity” arti­cles. Both owe their inclu­sion to 2 suc­cess­ful fron­t-­page appear­ances on Hacker News. DNA was sub­mit­ted by some­one I don’t know, but I sub­mit­ted the Zeo page at the con­clu­sion of my first Vit­a­min D exper­i­ment where I con­cluded that that Vit­a­min D con­sumed in the evening did indeed dam­age my sleep (I then fol­lowed up with a sec­ond exper­i­ment which found that Vit­a­min D con­sumed in the morn­ing did not dam­age my sleep.) I am proud of these two exper­i­ments and so I was grat­i­fied that Hacker News found them worth­while too.

  1. Silk Road: 308,895
  2. DNB FAQ: 36,488
  3. home/main page: 32,259
  4. Modafinil: 23,801
  5. Nootrop­ics: 21,970
  6. : 15,860
  7. : 15,841
  8. Spaced rep­e­ti­tion: 8,628
  9. Links: 7,431

Dona­tion-­wise, I received ~$17, and a num­ber of Pay­pal dona­tions: $5, $4, $5, and $100 from a par­tic­u­larly gen­er­ous Less­Wronger who wished me to backup my files more securely & remote­ly. Alexan­dra Carmichael was impressed by my Zeo exper­i­ments and asked me to write an ebook on sleep for an upcom­ing Quan­ti­fied Self series of O’Reilly ebooks; we will see how that goes.4


Hacker News, Red­dit, and Less­Wrong remain major refer­ral dri­vers. For recent new pages, I’ve been try­ing a check­list which includes sub­mis­sion to SU, Google+, Hacker News, Red­dit, and Less­Wrong as appro­pri­ate; I am not sure how well it is work­ing since pop­u­lar­ity seems very ran­dom.

July 2012–January 2013

“All I say is by way of dis­course, and noth­ing by way of advice. I should not speak so boldly if it were my due to be believed.”

(“Of Crip­ples”, Essays)

darcs-graph for this period (2012-07-03–2013-01-02):

Plot of patch cre­ations (y-ax­is) ver­sus date (x-ax­is): July 2012 to 2013-01-02


Ana­lyt­ics traf­fic records 758,843 page views by 366,028 unique vis­i­tors over the 184 days for a daily aver­age of 4,124.1 page-views, which is dou­ble the pre­vi­ous half-year aver­age of 2,012 daily page-views; traf­fic growth is clearly slow­ing, though, since the pre­vi­ous half-year had quadru­ple the traf­fic com­pared to its pre­de­ces­sor. My pre­dic­tion of break­ing the mil­lion page-view mark came true, by a very large mar­gin: the life­time total page-views is now 1,568,957 page-views.

Plot of page-hits (y-ax­is) ver­sus date (x-ax­is), late 2012

Pop­u­lar­ity rank­ings have changed a bit: the Death Note essay and my sleep exper­i­ments have fallen out of the top 10 (the for­mer because not many peo­ple are still link­ing it, and the lat­ter prob­a­bly because my lat­est exper­i­ments were rel­a­tively bor­ing), replaced by a side­bar link and one of my ter­ror­is­m-re­lated essays:

  1. Silk Road: 491,934
  2. home/main page: 37,853
  3. Modafinil: 31,047
  4. DNB FAQ: 27,433
  5. Nootrop­ics: 22,015
  6. Drug heuristics/Algernon’s Law: 18,991
  7. Spaced rep­e­ti­tion: 11,693
  8. Links: 8,123
  9. About: 8,075
  10. : 7,369

I am quite sur­prised that my essay does not even make the top 50 pages, given that it deals with an novel the­sis on which there’s many inter­est­ing things to think and which is eas­ily mis­un­der­stood.

Dona­tions: the ebook fell through when O’Reilly decided to can­cel the entire series, which was a dis­ap­point­ment; Carmichael had fin­ished her book on mood and is self­-pub­lish­ing, but I haven’t seen tremen­dous inter­est in sleep and will prob­a­bly just roll my draft mate­r­ial into the exist­ing Zeo page. Pay­pal dona­tions per­formed out­stand­ing­ly: I received $10, $25, $10, $15, $200, & $10 ($270). Bit­coin­ers were not so gen­er­ous: $4, $5, & $17 ($26, or $21 at the 2013-01-02 Mt.­Gox exchange rate).


To Hacker News, Red­dit, and Less­Wrong, I can add as a major refer­rer Wikipedi­a—pri­mar­ily to the Silk Road arti­cle, but also to a few Evan­ge­lion-re­lated pages. Stum­ble­Upon has declined to the 10th largest refer­rer:

  1. HN: 16,886
  2. Red­dit: 15,219
  3. Wikipedia: 7,531
  4. Less­Wrong: 6,332
  5. Face­book: 3,952
  6. 3,733
  7. 2,649
  8. 1,798
  9. 1,694
  10. 1,239

I haven’t spent much time pro­mot­ing my con­tent, but improv­ing the site with meta­data & writ­ing new con­tent: for exam­ple, I tripled the size of my data­base.


January 2013–July 2013

“I was dec­i­mat­ed. To pro­gram any more would be point­less. My pro­grams would never live as long as . A com­puter will never live as long as The Trial. …What if was only writ­ten for ?”

_why the lucky stiff (CLOSURE)

darcs-graph for this period (2013-01-03–2013-07-02):

Plot of patch cre­ations (y-ax­is) ver­sus date (x-ax­is): Jan­u­ary 2013 to 2013-07-02


Ana­lyt­ics traf­fic records 832,415 page-views by 422,285 unique vis­i­tors over 181 days, or 4598 page-views per day. (The life­time total page-views has thus reached 2,403,807.) This rep­re­sents ~11% growth in traf­fic com­pared to the pre­vi­ous 6-month period of 4,124 dailies, con­tin­u­ing the slow­down trend—­pos­si­bly the next half-year won’t see any growth, or a decline.

Plot of page-hits (y-ax­is) ver­sus date (x-ax­is), ear­ly-mid 2013

Pop­u­lar­ity rank­ings have changed: Silk Road lost a lot of Wikipedia traf­fic when, dur­ing an edit­ing dis­pute, an admin­is­tra­tor deleted it from their arti­cle on Silk Road, and it was only restored on 2 July. 2 new sta­tis­ti­cal essays ( & ) enter the top 10, but nei­ther were able to dethrone Silk Road as that dark­net mar­ket con­tin­ued to thrive & receive media cov­er­age:

  1. Silk Road: 396,993
  2. Modafinil: 58,856
  3. Google shut­downs: 49,907
  4. DNB FAQ: 25,979
  5. Death Note script: 25,792
  6. Nootrop­ics: 22,702
  7. home/main page: 20,051
  8. Links: 16,789
  9. About: 14,746
  10. Spaced rep­e­ti­tion: 11,711

Finan­cial: this period was a remark­able peri­od. My con­tract­ing work dried up (largely my own fault) and pressed for mon­ey, I began explor­ing alter­na­tives:

  1. the eas­i­est strat­egy was to turn all my Ama­zon links for books & nootrop­ics & mis­cel­la­neous into affil­i­ate links; they are already there, do not affect read­ers, and cod­ing it up was as sim­ple as append­ing a string to links in hakyll.hs by a sim��ple mod­i­fi­ca­tion of my exist­ing code for eas­ily link­ing to Wikipedia. This worked as far as it went, which was not very far ($47 in Q1 2013 and $130 Q2 2013).
  2. more painful­ly, I decided to try Google’s . While AdWords had­n’t been the most pleas­ant expe­ri­ence in the world, it was fairly decent, and Google’s ads always seemed rea­son­able to me in the search engine (be­fore I learned of AdBlock). AdSense has scary lan­guage in its terms of ser­vice for­bid­ding detailed dis­cus­sion of rev­enue, but I should be safe when I say that the mean CTR was 0.17% & the mean CPC was $0.57, ~40% of vis­i­tors did not have ad fil­ter­ing enabled, and so over the 80 days AdSense was enabled, I earned ~$260. Help­ful, but not ren­t-­pay­ing.
  3. dona­tions made the dif­fer­ence. After “Google shut­downs” hit the front page of Hacker News, an acquain­tance men­tioned I could use some money. I received $408 in 14 Pay­pal dona­tions; through Bit­coin, 1.62btc (then $189) in 4 dona­tions (fol­lowup). I am grate­ful to all the HNers who donat­ed.
  4. Related to that, I was con­tacted in May by Nava White­ford with an offer: he had heard my finan­cial tar­get of $300/month and offered that (in bit­coins) in exchange for a sponsorship/banner link­ing to his graph­i­cal ter­mi­nal HTerm—a com­bi­na­tion purchase/donation. This was both larger & more sta­ble than AdSense, so I accept­ed, and 3 months have thus far passed.


Wikipedia declines in this period due to the afore­men­tioned removal, and more obscure sites become chief refer­rers:

  1. 45,186
  2. 34,225
  3. 16,527
  4. 11,177
  5. Twit­ter: 6,717
  6. 5,675
  7. 5,576
  8. 4,982
  9. 4,496
  10. 4,336

July 2013–January 2014

“The great globe reels in the solar fire, / Spin­ning the triv­ial and unique away. / (How all things flash! How all things flare!) / …Time is the school in which we learn / Time is the fire in which we burn…”

(“Calmly We Walk Through This April’s Day”)

darcs-graph for this period (2013-07-03–2014-01-02):

Plot of patch cre­ations (y-ax­is) ver­sus date (x-ax­is): July 2013 to 2014-01-02


Ana­lyt­ics traf­fic records 770,264 page-views by 340,104 unique vis­i­tors over 184 days, or 4186 page-views per day. (Life­time total: 3,173,172 page-views by 1,472,606 unique vis­i­tors.) Since the pre­vi­ous period was 4598/day, that means traf­fic fell by ~9%. (Traf­fic would have fallen even more if I had­n’t hap­pened to be engaged in a suc­cess­ful cam­paign of Hacker News sub­mis­sions at the time.) My guess is that this is largely attrib­ut­able to the fall of Silk Road in early Octo­ber ren­der­ing my page of much less inter­est, and that was one of my most pop­u­lar pages. It was inevitable, real­ly, but I had hoped SR would last more than <3 years. Traf­fic may fall or increase dur­ing the next peri­od.

Plot of page-hits (y-ax­is) ver­sus date (x-ax­is), mid-late 2013

This rebal­anc­ing is evi­dent in the traf­fic rank­ings, where SR is still #1 and big­ger than #2 (Modafinil), but by a much smaller fac­tor this time (1.26x rather than 6.66x):

  1. Silk Road: 80,233
  2. Modafinil: 63,966
  3. home/main page: 65,203
  4. Mela­ton­in: 50,952
  5. : 36,573
  6. Spaced rep­e­ti­tion: 35768
  7. : 26,825
  8. Nicotine: 25,370
  9. DNB FAQ: 23,169
  10. Nootrop­ics: 20,883

My finances con­tin­ued to improve in part due to White­ford, addi­tional dona­tions (I explored Flattr & Grati­pay; the lat­ter seems to be work­ing out bet­ter), another more-­tar­geted ban­ner on the modafinil page, and a major appre­ci­a­tion in Bit­coin. (Takes a real load off me.) I thank all my dona­tors. I think the results show in my .


  1. 148,940
  2. 84,163
  3. 41,511
  4. 31,967
  5. 20,394
  6. [Twit­ter]: 19,628
  7. 16,808
  8. 11,135
  9. [Google+?]: 10,154
  10. 10,128

Inter­est­ing­ly, despite my Hacker News exper­i­ment (which resulted in dozens of my pages reach­ing its main page), is still beaten out traf­fic-­wise by Red­dit, by a huge fac­tor (al­most dou­ble).


January–July 2014

“A cour­te­ous hand­some young man / well-­versed in the Clas­sics and His­to­ries / peo­ple address him sir / every­one calls him a scholar / but he has­n’t found a posi­tion yet / and he does­n’t know how to farm / in win­ter he wears a tat­tered robe / this is how books fool us”

(#28, The Col­lected Songs of Cold Moun­tain)

Activ­ity graph5 for this period (2014-01-03–2014-07-02):

Plot of patch cre­ations (y-ax­is) ver­sus date (x-ax­is): Jan­u­ary to July 2014


Ana­lyt­ics traf­fic records 370,592 page-views by 162,032 unique vis­i­tors over 181 days, or 2047 page-views per day. (Life­time total: 3,544,489 page-views by 1,628,173 unique vis­i­tors.) Since the pre­vi­ous period was 4186/day, that means traf­fic fell by a dis­as­trous ~50%, defeat­ing my hopes that traf­fic might recover post-Silk-Road & Hacker News exper­i­ment

Plot of page-views (y-ax­is) ver­sus date (x-ax­is), ear­ly-mid 2014

The traf­fic rank­ings show a con­tinue in the plum­met of SR, and its dethron­ing by Modafinil and even the index page, while host­ing some early Satoshi Nakamoto drew con­sid­er­able inter­est & my page was sur­pris­ingly linked on Tyler Cowen’s Mar­ginal Rev­o­lu­tion blog:

  1. Modafinil: 54,992
  2. home/main page: 43,984
  3. Silk Road: 27,998
  4. Spaced rep­e­ti­tion: 27,680
  5. DNB FAQ: 17,833
  6. Nootrop­ics: 15,935
  7. Mis­takes: 15,267
  8. Nicotine: 11,936
  9. : 9,827
  10. : 9,391

On the plus side, it’s hard to see how traf­fic could fall much more.

Finance-­wise, Grati­pay has worked out much bet­ter than I expect­ed. I thank all my dona­tors.


Hacker News turns out to no longer show up in the refer­rals ana­lyt­ics due to strip­ping refer­rals (ap­par­ently related to HTTPS use?), which is unfor­tu­nate; I sus­pect other sites are also “going dark”, given that the num­bers are more than halv­ing:

  1. 24,176
  2. 5,695
  3. 5,504
  4. 4,051
  5. 3,954
  6. 3,051
  7. Twit­ter: 2,771
  8. 2,54
  9. 1,812
  10. 1,625

I also began ; with 300 peo­ple on it, it does­n’t seem to make a big dif­fer­ence to traf­fic but it’s a good way to curate my links and writ­ings, and may be quite use­ful.

July 2014–January 2015

“Writ­ing is in itself a joy, / Yet saints and sages have long since held it in awe.”

“For it is being, cre­ated from a void; / It is sound rung out of pro­found silence. / In a sheet of paper is con­tained the infinite, / And, evolved from an inch-­sized heart, an end­less panora­ma.”

Lu Chi’s (sec­tion 4, ‘The Joy of Writ­ing’, of “The Art of Writ­ing”).

Activ­ity graph6 for this period (2014-07-03–2015-01-02):

Plot of patch cre­ations (y-ax­is) ver­sus date (x-ax­is): July 2014 to Jan­u­ary 2015

Traffic: Late 2014

Ana­lyt­ics traf­fic records 347,482 page-views by 154,599 unique vis­i­tors over the 184 days, or 1888 page-views per day for a daily drop of 159 com­pared to the pre­vi­ous six months. (Life­time total: 3,891,971 page-views by 1,778,197 unique vis­i­tors.) This drop is not nearly as bad before, and may rep­re­sent a bot­tom.

Plot of page-views (y-ax­is) ver­sus date (x-ax­is), late 2014

Page rank­ings:

  1. Modafinil: 44,420
  2. home/main page: 41,499
  3. Nootrop­ics: 41,499
  4. Silk Road: 21,863
  5. Spaced rep­e­ti­tion: 17,620
  6. Causal­i­ty: 15,396
  7. DNB FAQ: 14,463
  8. : 11,313
  9. Nicotine: 9,804
  10. Links: 8,823

Silk Road declines a lit­tle more; the Wei Dai/Satoshi Nakamoto emails & Mis­takes pages drop out after their moments in the lime­light; and we see 2 new entries: I wrote a sta­tis­ti­cal essay on quan­ti­fy­ing how often cor­re­la­tion ≠ causal­ity (a topic of burn­ing impor­tance in almost any health-re­lated research), and an older cryp­tog­ra­phy essay became abruptly more rel­e­vant when one of my pro­pos­als was imple­mented using Bit­coin.

Grati­pay con­tin­ues to work well, and I am grate­ful for all my dona­tors.

Promotion: Late 2014

I did min­i­mal active pro­mo­tion of my con­tent, let­ting my AdWords credit be used up and not spend­ing much effort on Red­dit or Hacker News,

Con­fus­ing­ly, HN is back in my refer­rals:

  1. 27,182
  2. 18,420
  3. 4,583
  4. Twit­ter: 3,559
  5. 2,327
  6. Longecity: 2,143
  7. 2,123
  8. 1,488
  9. 1,346
  10. 1,276

My con­tin­ued, hav­ing reached 456 sub­scribers.


January 2015–July 2015

‘“The hor­ror of that moment,” the King went on, “I shall never never for­get!” “You will, though”, the Queen said, “if you don’t make a mem­o­ran­dum of it.”’

Lewis Car­roll ()

Activ­ity graph for this period (2015-01-03–2015-07-02):

Plot of patch cre­ations (y-ax­is) ver­sus date (x-ax­is): Jan­u­ary 2015 to July 2015

Traffic: Early 2015

Ana­lyt­ics traf­fic records 378,580 page-views by 175,303 unique vis­i­tors over the 181 days, or 2091 page-views per day for a daily increase of +203 com­pared to the pre­vi­ous six months. (Life­time total: 4,270,551 page-views by 1,949,700 unique vis­i­tors.) This is bet­ter than the pre­vi­ous drop but may be due mostly to some addi­tional spikes from HN.

Plot of page-views (y-ax­is) ver­sus date (x-ax­is), early 2015

Page rank­ings:

  1. Modafinil: 48,259
  2. home/main page: 48,614
  3. : 20,312
  4. Spaced rep­e­ti­tion: 17,522
  5. DNB FAQ: 16,076
  6. Silk Road: 15,116
  7. LSD micro­dos­ing: 13,673
  8. Nicotine: 9,804
  9. Links: 9,234
  10. Death Note Anonymi­ty: 9,234

The new entry here, “DNM arrests”, is a long-term project in which I com­pile all the sev­eral hun­dred known and infer­able arrests/prosecutions related to Silk Road 1 and its suc­ces­sor DNMs; it can be very dif­fi­cult to find details for indi­vid­ual cases (often requir­ing hours read­ing court doc­u­ments for a key detail) or keep track of them, and only recently was I able to clear out my back­log and present it to the com­mu­ni­ty. But it’s use­ful in giv­ing peo­ple an author­i­ta­tive idea of the risks and what peo­ple are arrested for, and I’m pleased to see that it’s being read & con­sult­ed. “LSD micro­dos­ing” seems to be get­ting more traf­fic due to some recent media cov­er­age of the idea and peo­ple link­ing to my writeup as (still) the only blinded self­-­ex­per­i­ment on the top­ic.

Grati­pay, unfor­tu­nate­ly, has been forced to change due to finan­cial reg­u­la­tions and is no longer use­ful for dona­tions. It had a good run but I will prob­a­bly be switch­ing to Patre­on.

Promotion: Early 2015

As before, I did min­i­mal active pro­mo­tion of my con­tent.

  1. 27,783
  2. 11,538
  3. 3,264
  4. Twit­ter ( 2,700
  5. Slate Star Codex: 2,357
  6. Longecity: 1,871
  7. 1,403
  8. Deep Dot Web: 1,226
  9. 1,197
  10. Face­book: 1,148

My monthly mail­ing list has reached 606 sub­scribers.

July 2015–January 2016

“We will never run out of things to pro­gram as long as there is a sin­gle pro­gram around.”

( 1982)

Activ­ity graph for this period (2015-07-03–2016-01-03):

Plot of patch cre­ations (y-ax­is) ver­sus date (x-ax­is): July 2015 to Jan­u­ary 2016

Traffic: Late 2015

Ana­lyt­ics traf­fic records 472,244 page-views by 400,305 unique vis­i­tors over the 185 days, or 2552 page-views per day for a daily increase of +461 com­pared to the pre­vi­ous six months. (Life­time total: 4,742,795 page-views by 2,191,138 unique vis­i­tors.)

Plot of page-views (y-ax­is) ver­sus date (x-ax­is), late 2015

Page rank­ings:

  1. Modafinil: 67,152
  2. home/main page: 50708
  3. : 44,164
  4. : 23,879
  5. LSD micro­dos­ing: 22,078
  6. : 20,581
  7. DNM arrests: 18,524
  8. Nootrop­ics: 15,709
  9. Ter­ror­ism is not about Ter­ror: 15,223
  10. Spaced rep­e­ti­tion: 14,903
  11. DNB FAQ: 13,954
  12. Silk Road: 11,606

The new entry here, “DNM archives”, is the cul­mi­na­tion of my DNM work: mirrors/scrapes of dozens of mar­kets, along with all sorts of archives and datasets related to the DNMs, cov­er­ing years of mar­ket activ­ity (1.2tb in total, to give an idea of scope). The other 2 sur­prises were a tuto­r­ial on how to set up char-rnn on an Ama­zon EC2 instance and train it to gen­er­ate remark­ably valid CSS (sug­gest­ing that it may be pos­si­ble to train a rein­force­men­t-learn­ing deep neural net­work to design the visual appear­ance of web­sites and opti­mize the appear­ance based on vis­i­tor exper­i­ments), and an old lit­tle com­pi­la­tion I made of things exhibit­ing Tur­ing-­com­plete­ness which I found bizarre or sur­pris­ing or dis­turb­ing (which nat­u­rally led to a lot of nit­pick­ing and debate over def­i­n­i­tion­s).

My Patreon pro­file, set up after I was forced out of Grati­pay, has been a smash­ing suc­cess, with gen­er­ally ~$300/month in dona­tions.

Promotion: Late 2015

As before, I did min­i­mal active pro­mo­tion of my con­tent.

  1. 51,256
  2. 25,479
  3. Twit­ter ( 7,748
  4. Slate Star Codex: 3,166
  5. Face­book: 3,044
  6. 3,035
  7. Dis­qus: 2,042
  8. Bul­let Proof Exec: 1,931
  9. Longecity: 1,887
  10. Wired: 1,854

(I have no idea why Face­book traf­fic increased so much, since the refer­rers are use­less.)

My monthly mail­ing list has reached 857 sub­scribers.


January 2016–July 2016

“It is easy enough to say that man is immor­tal sim­ply because he will endure: that when the last ding-­dong of doom has clanged and faded from the last worth­less rock hang­ing tide­less in the last red and dying evening, that even then there will still be one more sound: that of his puny inex­haustible voice, still talk­ing. I refuse to accept this. I believe that man will not merely endure: he will pre­vail.”

William Faulkner (1950 Nobel Prize speech)

Activ­ity graph for this period (2016-01-04–2016-07-03):

Plot of patch cre­ations (y-ax­is) ver­sus date (x-ax­is): Jan­u­ary 2016 to July 2016

Traffic: Early 2016

Ana­lyt­ics traf­fic records 294,009 page-views by 141,795 unique vis­i­tors over the 182 days, or 1615 page-views per day for a daily decrease of 937 com­pared to the pre­vi­ous six months. (Life­time total: 5,036,804 page-views by 2,329,713 unique vis­i­tors.)

Plot of page-views (y-ax­is) ver­sus date (x-ax­is), early 2016

Page rank­ings:

  1. homepage/index: 35,673
  2. Modafinil: 33,928
  3. LSD micro­dos­ing: 21,712
  4. DNM arrests: 14,322
  5. Spaced rep­e­ti­tion: 14,044
  6. Nootrop­ics: 12,117
  7. DNB FAQ: 10,983
  8. Silk Road: 7,654
  9. About: 6,642
  10. Links: 6,410

My Patreon pro­file con­tin­ues to be a suc­cess.

Promotion: Early 2016

As before, I did min­i­mal active pro­mo­tion of my con­tent.

  1. Red­dit: 17,792
  2. Face­book: 4119
  3. Twit­ter: 2,572
  4. HN: 2,518
  5. Slate Star Codex: 2,467
  6. Less­Wrong: 2,135
  7. Longecity: 1,555
  8. 1,440
  9. 1,035
  10. MetaFil­ter: 1,018

My monthly mail­ing list has reached 1053 sub­scribers.

July 2016–January 2017

“With the peo­ple of the past / How I wish I could share the beauty / Of these cherry blos­soms. // Have each of us across the years / Left behind this very thought?”

(“Shogaku Hyakushu”, 1181 AD, trans. Bundy 1990)

Activ­ity graph for this period (2016-07-04–2017-01-03):

Plot of patch cre­ations (y-ax­is) ver­sus date (x-ax­is): July 2016–­Jan­u­ary 2017

Traffic: Late 2016

Ana­lyt­ics traf­fic records 339,529 page-views by 172,203 unique vis­i­tors over the 184 days, or 1845 page-views per day for a daily increase of 230 com­pared to the pre­vi­ous six months. (Life­time total: 5,376,333 page-views by 2,499,949 unique vis­i­tors.)

Plot of page-views (y-ax­is) ver­sus date (x-ax­is), late 2016

Page rank­ings:

  1. Nootrop­ics: 43,266
  2. index/main page: 39,161
  3. Spaced rep­e­ti­tion: 26,965
  4. Modafinil: 26,849
  5. DNM arrests: 16,279
  6. LSD micro­dos­ing: 12,943
  7. DNB FAQ: 10,501
  8. : 8,026
  9. In Defense of Inclu­sion­ism: 7,037
  10. Links: 6,677

My Patreon pro­file has largely lev­eled off at a con­stant $310.

Promotion: Late 2016

As before, I did min­i­mal active pro­mo­tion of my con­tent.

  1. HN: 42,341

    Traf­fic was dri­ven pri­mar­ily by a few big HN sub­mis­sions by other peo­ple.

  2. Red­dit: 16,151

  3. Twit­ter: 4,915

  4. Face­book: 4,702

  5. 1,585

  6. Longecity: 1,175

  7. 1,171

  8. Duck­DuckGo search: 990

  9. 912

  10. Wikipedia: 794

My monthly mail­ing list has reached 1212 sub­scribers, up ~200.


January 2017–July 2017

“Authors write things down so as to have to think of them less.”

Activ­ity graph for this period (2017-01-04–2017-07-03):

Plot of patch cre­ations (y-ax­is) ver­sus date (x-ax­is): Jan­u­ary 2017–July 2017

Traffic: Early 2017

Ana­lyt­ics traf­fic records 228,652 page-views by 112,814 unique vis­i­tors over the 181 days, or 1263 page-views per day for a daily decrease of 582 vis­its com­pared to the pre­vi­ous six months. (Life­time total: 5,604,985 page-views by 2,609,657 unique vis­i­tors.)

Plot of page-views (y-ax­is) ver­sus date (x-ax­is), early 2017

Page rank­ings:

  1. Modafinil: 16,980
  2. Spaced rep­e­ti­tion: 14,484
  3. LSD micro­dos­ing: 13,394
  4. DNB FAQ: 10,145
  5. : 9,760
  6. Nootrop­ics: 9,309
  7. DNM arrests: 8,148
  8. Silk Road: 6,456
  9. Links: 5,668
  10. About: 5,176

Aside from a crit­i­cal exe­ge­sis of “Story Of Your Life” (prompted by its suc­cess­ful movie adap­ta­tion), I spent much of early 2017 tweak­ing and work­ing on unpop­u­lar sta­tis­tics projects (par­tic­u­larly , which though thor­ough and highly edu­ca­tional for myself, like , will never be pop­u­lar read­ing). In late 2017, I hope to fin­ish up many draft or in-progress works, which may help main­tain traf­fic.

After rewrit­ing & includ­ing social proof & tweet­ing, my Patreon pro­file has increased to $600.

Promotion: Early 2017

As before, I did min­i­mal active pro­mo­tion of my con­tent. Refer­rals:

  1. HN: 10,881 (“Story Of Your Life” was an unex­pected hit)
  2. Red­dit: 9,538
  3. Twit­ter: 3,702
  4. Face­book: 2,274
  5. Less­Wrong: 1,568
  6. Slate Star Codex: 1,462
  7. 1,315
  8. Longecity: 1,071
  9. Brain Work­shop: 861
  10. Duck­Duck­Go: 820

My monthly mail­ing list has reached 1419 sub­scribers, up ~207.

July 2017–January 2018

“It is only the attempt to write down your ideas that enables them to devel­op.”

Wittgen­stein to Drury (Rec­ol­lec­tions of Wittgen­stein, Rhees 1984)

Activ­ity graph for this period (2017-07-04–2018-01-03):

Plot of patch cre­ations (y-ax­is) ver­sus date (x-ax­is): July 2017–­Jan­u­ary 2018

Traffic: Late 2017

Ana­lyt­ics traf­fic records 326,852 page-views by 155,532 unique vis­i­tors over the 184 days, or 1776 page-views per day for a daily increase of 513 vis­its com­pared to the pre­vi­ous six months. (Life­time total: 5,931,837 page-views by 2,763,402 unique vis­i­tors.)

Plot of page-views (y-ax­is) ver­sus date (x-ax­is), late 2017

Page rank­ings (some­what scram­bled due to the big renam­ing and GA not merg­ing ana­lyt­ics across URLs):

  1. index: 43,863
  2. : 20,013
  3. Modafinil: 14,340
  4. DNM arrests: 14,115
  5. : 13,808
  6. DNB FAQ: 13,650
  7. : 11,401
  8. Spaced rep­e­ti­tion: 12,428
  9. : 11,150
  10. Nootrop­ics: 7,615

My A/B test of ban­ner ads turned in an unex­pected result of sub­stan­tial harm, which was of con­sid­er­able gen­eral inter­est (espe­cially to any­one who runs an ad-­sup­ported web­site), so no sur­prise that had a lot of traf­fic. I wrote off the obscure neural net/tank story years ago after some cur­sory inves­ti­ga­tion showed it had all the clas­sic traits of an urban leg­end; but it has kept com­ing up, and at Anders Sand­berg’s request, I wrote it up in con­sid­er­ably more detail, and was sur­prised that peo­ple took such an inter­est in it and even defended it as prob­a­bly gen­uine, in the absence of all evi­dence—I guess peo­ple like debunk­ings? The final odd­ity was the resur­gence of my old essay on an extremely obscure bit of US his­tor­i­cal triv­ia, that it had tried to acquire Green­land in the early Cold War, where I noted that Den­mark’s refusal was a costly one; I don’t know why any­one was read­ing it, but I was annoyed at the idi­otic objec­tions peo­ple made, typ­i­cally either pure fan­ta­sy, already refuted like the many sug­ges­tions that Green­land oil would yet pay off, or illog­i­cal virtue sig­nal­ing—iron­i­cally demon­strat­ing my point. (Con­sid­er­able effort invested in defin­ing hun­dreds of redi­rects for 404 errors did­n’t pay off as peo­ple kept dis­cov­er­ing ever new ways to mis­spell URLs, but it did allow me to rename pages with­out destroy­ing traf­fic by break­ing most incom­ing links, switch­ing from spaces to hyphens in order to idiot-proof them.)

My Patreon pro­file has increased to $700.

Promotion: Late 2017

As before, I did min­i­mal active pro­mo­tion of my con­tent. Refer­rals:

  1. HN: 26,112
  2. Red­dit: 10,715
  3. Mar­ginal Rev­o­lu­tion: 8,064
  4. Twit­ter: 7,111
  5. Slate Star Codex: 5,906
  6. Face­book: 2228
  7. Less­Wrong: 1,547
  8. The Browser: 1,057 (this would have been higher if The Browser had not care­lessly bro­ken its URL due to the space-escap­ing issue, which prompted my site-wide renam­ing)
  9. Duck Duck Go: 782
  10. Longecity: 777

My monthly mail­ing list has reached 1786 sub­scribers, up ~367.


January 2018–July 2018

“Peo­ple will have their mir­a­cles, their sto­ries, their heroes and hero­ines and saints and mar­tyrs and divini­ties to exer­cise their gifts of affec­tion, admi­ra­tion, won­der, and wor­ship, and their Judases and dev­ils to enable them to be angry and yet feel that they do well to be angry. Every one of these leg­ends is the com­mon her­itage of the human race; and there is only one inex­orable con­di­tion attached to their healthy enjoy­ment, which is that no one shall believe them lit­er­al­ly. The read­ing of sto­ries and delight­ing in them made Don Quixote a gen­tle­man: the believ­ing them lit­er­ally made him a mad­man who slew lambs instead of feed­ing them.”

(A Touch­stone For Dogma)

Activ­ity graph for this period (2018-01-04–2018-07-03):

Plot of patch cre­ations (y-ax­is) ver­sus date (x-ax­is): Jan­u­ary 2018–July 2018

Traffic: Early 2018

Ana­lyt­ics traf­fic records 329,057 page-views by 148,481 unique vis­i­tors over the 181 days, or 1818 page-views per day for a daily increase of 42 pageviews a day com­pared to the pre­vi­ous six months. (Life­time total: 6,236,345 page-views by 2,945,757 unique vis­i­tors.)

Plot of page-views (y-ax­is) ver­sus date (x-ax­is), early 2018

Page rank­ings:

  1. index: 43,535
  2. : 29,755
  3. Modafinil: 16,339
  4. DNB FAQ: 15,922
  5. Spaced rep­e­ti­tion: 15,599
  6. : 10,531
  7. Nootrop­ics: 8,020
  8. Links: 7,105
  9. DNM arrests: 6,707
  10. Death Note Anonymi­ty: 6,464

“Laws of Tech: Com­modi­tize Your Com­ple­ment” is based on an old Joel Spol­sky essay I found insight­ful and to explain a lot of puz­zling behav­iors in Sil­i­con Val­ley, so I was sur­prised when, dur­ing my SF trip, no one who I men­tioned it to in dis­cus­sions of tech­nol­ogy had heard of it; I returned resolved to try to pro­mote it some more by re-em­pha­siz­ing it and col­lat­ing a list of exam­ples so no one could fail to see the pat­tern, and I believe I suc­ceed­ed. The launch of Dan­booru2017 was less sat­is­fy­ing. I kept run­ning into dif­fi­cul­ties and delays, most asso­ci­ated with how to for­mat the meta­data (pro­vided in a not-­too-use­ful JSON for­mat)—in­deed, I had expected to launch it in April 2017 and build my deep learn­ing work­sta­tion in Decem­ber 2017. Instead, it was­n’t until Feb­ru­ary 2018 that I gave up on the meta­data, rea­son­ing that I was let­ting the per­fect be the enemy of bet­ter, and until July 2018 that I fin­ished build­ing my work­sta­tion with 2 GPUs (after 2 dead moth­er­boards, among other prob­lem­s). There have been some­where around 200 down­loads of the dataset as far as I can extrap­o­late from the tor­rent client upload & rsync logs, but thus far I only know of 1 seri­ous use of it (a CycleGAN paper). I remain hope­ful that uses will grad­u­ally increase over the next few years, the way they did for my dark­net mar­ket dataset; in any case, I’ve begun work­ing on some uses of Dan­booru2017 myself.

My Patreon pro­file decreased to $670.

Promotion: Early 2018


  1. HN: 13,533
  2. Red­dit: 6,686
  3. Twit­ter: 6,529
  4. Slate Star Codex: 6,246
  5. Face­book: 2,830
  6. 2,414
  7. Android Google Search: 788
  8. The Browser: 562
  9. 4chan: 523
  10. Duck Duck Go: 521

My monthly mail­ing list has reached 2183 sub­scribers, up ~307. By cross­ing the 2000 sub­scriber mark, MailChimp dis­abled my account and began charg­ing ~$30 a month to send out a sin­gle newslet­ter; as it’s a lit­tle hard for me to jus­tify spend­ing any­where up to $400/year on my newslet­ter, which I don’t directly make any money on, I had to stop using MailChimp. after look­ing around some, I switched to TinyLet­ter, which oddly enough is also run by MailChimp but does­n’t have that arbi­trary sub­scriber limit (though it lacks many fea­tures of MailChimp prop­er).

July 2018–January 2019

“Is not a patron, my lord, one who looks with uncon­cern on a man strug­gling for life in the water, and when he has reached ground, encum­bers him with help?”

Samuel John­son (, 1755)

Activ­ity graph for this period (2018-07-04–2019-01-03):

Plot of patch cre­ations (y-ax­is) ver­sus date (x-ax­is): July 2018–­Jan­u­ary 2019

Traffic: Late 2018

Ana­lyt­ics traf­fic records 407,395 page-views by 185,141 unique vis­i­tors over the 184 days, or 2214 page-views per day for an increase of 396 pageviews a day com­pared to the pre­vi­ous six months. (Life­time total: 6,638,722 page-views by 3,113,460 unique vis­i­tors.)

Plot of page-views (y-ax­is) ver­sus date (x-ax­is), late 2018

Page rank­ings:

  1. index: 50,849
  2. : 27,241
  3. Tur­ing-­com­plete: 26,662
  4. Notes: (for /): 17,706
  5. Spaced rep­e­ti­tion: 14,471
  6. Modafinil: 13,958
  7. “Bit­coin Is Worse Is Bet­ter”: 13,169
  8. DNB FAQ: 10,344
  9. Links: 9,729
  10. Drug heuristics/Algernon’s Law: 9,724
  11. And hon­or­able men­tions for the new pages which almost made it to the top 10 (#13) despite being released mid-way Decem­ber 2018, and (on cat psy­chol­o­gy, genet­ics, domes­ti­ca­tion, & dys­gen­ics) at #21.

I was dis­ap­pointed that two of my new ones had lit­tle or no inter­est ( & ), and that Dan­booru2017, despite its use in the impres­sive , saw lit­tle uptake. Maybe Dan­booru2018 will do bet­ter.

My Patreon pro­file increased to $835.

Promotion: Late 2018

Refer­rals saw some unusual changes:

  1. HN: 37,603
  2. Twit­ter: 10,909
  3. Red­dit: 6,046
  4. Slate Star Codex: 4,479
  5. Face­book: 1,389
  6. Github: 1,209
  7. Coin­Mar­ket­Cap: 1,115
  8. TinyLet­ter: 1,089
  9. : 795
  10. Less­Wrong: 782

My monthly mail­ing list has reached 2753 sub­scribers, up ~570.


January 2019–July 2019

“You think you know when you learn, are more sure when you can write, even more when you can teach, but cer­tain when you can pro­gram.”

( 1982)

Activ­ity graph for this period (2019-01-04–2019-07-02):

Plot of patch cre­ations (y-ax­is) ver­sus date (x-ax­is): Jan­u­ary–July 2019

Traffic: Early 2019

Ana­lyt­ics traf­fic records 653,876 page-views by 325,842 unique vis­i­tors over the 181 days, or 3612 page-views per day for an increase of 1398 pageviews a day com­pared to the pre­vi­ous six months. (Life­time total: 7,286,760 page-views by 3,431,759 unique vis­i­tors.)

Plot of page-views (y-ax­is) ver­sus date (x-ax­is), early 2019

Page rank­ings:

  1. site write­up: 95,635
  2. index: 97,767
  3. Dan­booru2018: 30,936
  4. Com­ple­ment: 28,165
  5. Spaced-rep­e­ti­tion: 26,846
  6. : 20,368
  7. tuto­ri­al: 17,616
  8. Links: 14,594
  9. Modafinil: 14,488
  10. About: 14,057
  11. : 11,738

The main event in early 2019 was the release of the StyleGAN code­base, which I imme­di­ately began using to gen­er­ate anime faces; it worked even bet­ter than I hoped, cre­at­ing amus­ing 512px anime faces in a frac­tion of the time of ProGAN. For a lark, in Feb­ru­ary I cre­ated a sim­ple lit­tle sta­tic web­site, to show off ran­dom face sam­ples, which then went viral, quickly receiv­ing over a mil­lion vis­its (mostly from Chi­na). I enhanced TWDNE with addi­tional StyleGAN improve­ments and anime plot sum­maries gen­er­ated by , and wrote up and the as tuto­ri­als, which ben­e­fit from the stil­l-­con­sid­er­able traf­fic and have been use­ful to many peo­ple train­ing their own StyleGANs.

My Patreon pro­file was steady at $838.

Promotion: Early 2019

Refer­rals changed dra­mat­i­cal­ly, dri­ven mostly by TWDNE:

  1. 75,707
  2. 34,761
  3. 12,744
  4. 7,993
  5. 4,754
  6. 2,476
  7. 1,886
  8. 1,746
  9. 1,342
  10. 1,164

My monthly mail­ing list has reached 3559 sub­scribers, up ~806.

July 2019–January 2020

“Edit­ing is a reword­ing activ­i­ty.”

( 1982)

Activ­ity graph for this period (2019-07-03–2020-01-02):

Plot of patch cre­ations (y-ax­is) ver­sus date (x-ax­is): July 2019–­Jan 2020

Traffic: Late 2019

Ana­lyt­ics traf­fic records 715,601 page-views by 349,472 unique vis­i­tors over the 184 days, or 3889 page-views per day for an increase of 277 page-views a day com­pared to the pre­vi­ous 6 months. This is inflated by host­ing var­i­ous HTML archives, par­tic­u­larly the which I need for my —tak­ing that out, net traf­fic appears to have decreased. (Life­time total: 8,004,312 page-views by 3,814,522 unique vis­i­tors.)

Plot of page-views (y-ax­is) ver­sus date (x-ax­is), late 2019

Page rank­ings (ex­clud­ing hosted pages):

  1. index: 50,757 + 28,276
  2. TWDNE: 44,335
  3. Death Note Anonymi­ty: 27,006
  4. Green­land: 22,887
  5. Faces: 21,124
  6. About: 15,678
  7. Spaced rep­e­ti­tion: 14,918
  8. Links: 14,231
  9. Modafinil: 14,229
  10. Dan­booru2018: 14,121
  11. DNB FAQ: 10,368

None of the 7 new pages or 5 new long sec­tions were major traf­fic suc­cesses (as much as I was fond of some like or ); I con­tin­ued sev­eral -re­lated pro­jects, but they did­n’t fin­ish before 2020. Site design was a major focus of my efforts dur­ing this peri­od, and I’m pleased with the suite of new popup tools & new mar­gin notes fea­tures;

My Patreon pro­file was up to $884.

Promotion: Late 2019

Refer­rals remained sim­i­lar:

  1. TWDNE: 29,075
  2. HN: 25,630
  3. Twit­ter: 10,911
  4. Red­dit: 4,516
  5. Slat­eStar­Codex: 4,040
  6. Face­book: 3,841
  7. 3,400
  8. Github: 1,905
  9. Alexey Guzey: 1,381
  10. 1,083

My monthly mail­ing list has reached 4347 sub­scribers, up 788.


January 2020–July 2020

“No, no, let us work just one more hour; let us com­pose just one more page. Of course we shall not get the book done, but we must keep on try­ing to fin­ish a lit­tle more…When you have a great and dif­fi­cult task, some­thing per­haps almost impos­si­ble, if you only work a lit­tle at a time, every day a lit­tle, with­out faith and with­out hope, sud­denly the work will fin­ish itself.”


Activ­ity graph for this period (2020-01-03–2020-07-04):

Plot of patch cre­ations (y-ax­is) ver­sus date (x-ax­is): Jan 2020–July 2020

Traffic: Early 2020

Ana­lyt­ics traf­fic records 641,127 page-views by 326,662 unique vis­i­tors over the 184 days, or 3,484 page-views per day for a decrease of 404 page-views a day com­pared to the pre­vi­ous 6 months. Because Gawker linked the Rot­ten.­com archive (send­ing >400,000 vis­i­tors & seri­ously dis­tort­ing my ana­lyt­ic­s), I’ve fil­tered out the HTML archives entirely & removed the GA code from all HTML pages to obvi­ate the issue, which appears to be respon­si­ble for most of the decrease.

(Life­time total: 8,354,905 page-views by 3,882,482 unique vis­i­tors.)

Plot of page-views (y-ax­is) ver­sus date (x-ax­is), early 2020

Page rank­ings (ex­clud­ing hosted pages):

  1. TWDNE: 64,906
  2. : 48,178
  3. index: 41624
  4. : 35,679
  5. Spaced rep­e­ti­tion: 24,974
  6. Faces: 23,618
  7. About: 19,712
  8. Links: 18,847
  9. May newslet­ter (): 16,877
  10. : 15,739

The rank­ings changed up a bit due to the remark­able longevity of TWDNE—people appar­ently never get tired of look­ing at the faces, that, or peo­ple keep dis­cov­er­ing it—and due to the suc­cess of some shorter pages I wrote to take a break from my deep learn­ing projects like the GPT-2 pref­er­ence learning/MIDI music-­gen­er­a­tion (which few cared about) or my GPT-3 work (a lit­tle more atten­tion there, but much less than it war­rant­ed) site design has­n’t quite fin­ished; the biggest change in this period was a rad­i­cal exper­i­ment in auto­mat­i­cally archiv­ing & mir­ror­ing most exter­nal links to elim­i­nate linkrot entire­ly.

My Patreon pro­file was up to $1,123 (large­ly, I sus­pect, on the strength of the deep learn­ing pro­ject­s).

Promotion: Early 2020

Refer­rals changed mod­est­ly:

  1. HN: 49,090
  2. TWDNE: 47,996
  3. Twit­ter: 19,214
  4. Slat­eStar­Codex: 9,193
  5. Red­dit: 7,426
  6. Face­book: 6,714
  7. Less­Wrong: 1,556
  8. Newslet­ter: 1,411
  9. GitHub: 1,321
  10. Medi­um: 1,030

My monthly mail­ing list has reached 5,849 sub­scribers, up 1502; this forced me to leave TinyLet­ter due to its 5k sub­scriber limit and move to Sub­stack (de­spite atten­dant dif­fi­cul­ties).

  1. Anec­do­tal­ly, the rank­ings seem cor­rect. When I went to a Less­Wrong meetup in Cal­i­for­nia, many knew of or had read the DNB FAQ, some had read or used my modafinil price-chart, and few remem­bered read­ing any­thing else.↩︎

  2. I am some­times reminded of another waka, by Ikkyu:

    To write some­thing and leave it behind us,
    Is but a dream.
    When we awake we know
    There is not even any­one to read it.

  3. They spend an aver­age of 27 sec­onds; in com­par­ison, my sec­ond largest source of traf­fic, Less­Wrongers, aver­age 3 min­utes and 29 sec­onds; even my third largest traf­fic source, Red­diters, man­age almost 2 min­utes. Even ran­dom peo­ple com­ing from Google man­age to spend 44 sec­onds on their vis­it!↩︎

  4. It ulti­mately fell through as O’Reilly decided the QS moment had passed, and killed the series.↩︎

  5. Since I con­verted the repo to use git, darcs-graph was no longer avail­able, so I wrote my own script in R using splines rather than a 7-day mov­ing aver­age:

    start <- as.Date("2014-01-02")
    end <- as.Date("2014-07-02")
    patches <- as.Date(system(paste0("git log --after=", start, " --before=", end,
                               " --format='%ad' --date=short master"),
    patchFrame <- aggregate( patches , by = list(patches) , length )
    days <- seq(start, end, "day")
    dayFrame <- aggregate( days , by = list(days) , length )
    bar <- rbind(dayFrame, patchFrame)
    all <- aggregate(bar$x, by=list(bar$Group.1), FUN=function (x) { sum(as.vector(x));})
    plot(x ~ Group.1, data=all, xlab="Date (Jan--July 2014)", ylab=" daily patches")
    library(splines); lines(predict(interpSpline(days, all$x)))
  6. I rewrote the script a lit­tle to use the nicer plots of ggplot2 and switches to LOESS for the smoothed curve:

    start <- as.Date("2014-07-03")
    end <- as.Date("2015-01-02")
    patches <- as.Date(system(paste0("git --git-dir=/home/gwern/wiki/.git/ log --after=",
                              start, " --before=", end,
                              " --format='%ad' --date=short master"), intern=TRUE))
    patchFrame <- aggregate( patches , by = list(patches) , length )
    days <- seq(start, end, "day")
    dayFrame <- aggregate( days , by = list(days) , length )
    bar <- rbind(dayFrame, patchFrame)
    all <- aggregate(bar$x, by=list(bar$Group.1), FUN=function (x) { sum(as.vector(x));})
    qplot(Group.1, x, data=all) +
     theme_bw() +
     xlab("Date (2014)") + ylab("Patch count") +
     theme(legend.title=element_blank()) +
  7. Of writ­ing after 3 spinal surg­eries (pg161–162 of Images Of Truth: Remem­brances And Crit­i­cism, Wescott 1962)↩︎