SubjectiveBayes: Updated One Game at a Time
Some usernames are jokes. This one is a thesis statement. SubjectiveBayes started playing Blitz chess in August 2025 with a prior sitting near 569, then collected roughly 7,600 games of evidence and let the posterior do the arguing. By June 2026 that posterior had climbed all the way to 2225 (2026-06-24). The subjective part is only the confidence going into the next game. The math keeps saying the same thing: about half the games are winnable, and the fun is finding out which half.
The Blitz Chess Origin Story: From 569 to the 2200 Club
The debut rating turned out to be, by any Bayesian standard, badly miscalibrated in the humble direction. Two hundred four games later, within that very first month, the rating had tripled. By the end of September 2025 it had crossed 2100. What followed was less a climb than a siege: month after month of 250 to 1,000 games hammering the 2200 barrier, with the peak finally landing in June 2026, in the middle of a 674-game month, because of course it did.
September 2026 remains the busiest month on record at 1,160 games, which works out to nearly 39 a day. This is not a player who visits the Blitz pool. This is a player who moved in, pays rent there, and treats the falling flag as furniture. Speaking of flags: over 2,100 games have ended on time, split almost perfectly between 1,001 wins and 1,020 losses, a roughly 50/50 referendum on clock management.
The Opening Repertoire: 1.e4, the Alapin, and Sicilians All the Way Down
As White, SubjectiveBayes opens with 1.e4 in essentially every game. The four experimental 1.e3 games in 2026 have been sealed in the archive and are not discussed. Against the Sicilian comes the signature weapon, the Sicilian Defense: Alapin Variation, deployed more than 1,100 times with a solid 52% score and a fondness for the c3 pawn the way other people are fond of pets.
As Black, the answer is the Sicilian Defense in open, Closed, and occasionally Najdorf Variation flavors. The model does have documented soft spots. The Scandinavian Defense has been a recurring headache across 372 meetings, and the classical King's Gambit Declined setups with 3.Bc4 have not gone much better. The data never lies, and this player, of all people, would never accuse the data of lying.
Playing Style: The Endgame Grinder Who Never Resigns
Most blitz players want fireworks. SubjectiveBayes wants a siege that outlasts your battery. The average win runs past move 70, endgames appear in about 78% of games, and the comeback rate sits at an absurd 85.6%. Even after dropping a whole piece, the win rate only sags to 47%, which is statistics-speak for refusing to read the room. Early resignation happens in under 1% of games, and truly one-sided losses account for just 1.5%. In online chess circles this is known as cockroach chess, and SubjectiveBayes has turned it into an identity.
The queen is the preferred checking piece, though at that stage the checks are less threats and more formalities. Land the first check and the win rate is 58%. Eat the first check and it drops to 41%. The lesson, learned 7,600 times: initiative matters, and SubjectiveBayes would like theirs back now, please.
The Nemesis Ledger: Opponents Who Broke the Model
Every honest prior needs counter-evidence, and the following accounts have supplied it in bulk:
- p003876: 1 win, 8 losses. A walking falsification of the entire model.
- Matt Matthew: 2-7. The king does, in fact, run.
- ron mag: 2-7. Numerically dominant, historically dominant.
- melvin1234567889: 2-7, softened by one merciful draw.
- kaylazeline: 2-6, proof that usernames starting with K are dangerous.
Balance does exist. con-passant stands at 4-3-1 across eight meetings, which feels correct for a rivalry named after a surprise pawn capture, while ahmad-alkhatib-7 and alaskapalla hover near even at 5-4 and 6-4. mathieuchess06 remains the most frequent sparring partner at twelve games and counting.
Body Clock, Streaks, and Other Rituals
The optimal game begins at 11 AM, where the win rate peaks at 59%. Noon is a statistical cliff at 38.4%, and the 10 AM slot shows 0% on a sample so small that even a devoted Bayesian files it under anecdote. Fridays are the roughest day of the week at 46%, while Mondays and Tuesdays flirt with 51%. The model's official recommendation: sleep in, then strike before lunch.
The streaks tell the rest of the story. There was a 36-game winning streak, the crown jewel of the profile. There was a 10-game losing streak, tuition paid in full. The current five-game skid is filed under rounding error, because the tilt factor reads a placid 10. The prior was 569. The evidence is 7,600 games and counting. The posterior says: see you at the board.