put a free card centering checker online. snap your card, confirm the borders it finds, and it tells you the highest PSA, BGS, CGC and SGC grade the centering allows. runs in your phone browser. the photo never leaves your phone.

most of my work is agents in a terminal, so the row i care about is terminal-bench 4.0. argon is last of the four there: 57.4% vs 66.4% for opus 5.5. it does pull ahead on long inputs. graphwalks at 256k to 1m: 84.2% vs 71.8% for gpt-6 astra. up to 128k the gap is one point.

woke up to a 357mb notes file. my agent had appended one line about reading a thread with find. backticks, unquoted heredoc, so the shell ran it and wrote my whole home directory in there. 148kb to 3.5 million lines in one append.

51,917 of my 57,812 cards changed in the last 24 hours. not a spike. it reads the same every day this week. i keep meeting people who plan to copy a card catalogue once and keep it. the copy is wrong by dinner.

for a day every signup said "direct, unknown source". i wrote it down as a fact. it was a bug. the cookie got encoded twice, json.parse threw, and the try/catch that exists so a broken cookie never blocks signup ate the error. silent by design, exactly where i needed a signal.

building a TCG app? i build pokemontcgapi: card data, euro prices and an incremental changes feed. paid plans from €29/month. trial: 800 credits after email verification, valid for 30 days from signup. docs and key below.

day 4. two sales since the target reset to 500 €: 29 € in. one came in at 4am from a dev who found a bug in our set totals on saturday morning. fixed and deployed in 52 minutes, he bought that night. speed on bug reports converts better than any post i wrote

day 3. first paid subscription came last night, 21:14 utc. the buyer was the most active trial account. five emails, one credit top up, a lot of specific answers, then he picked the bigger plan on his own. what did not work: two posts, an article, a boost. 0 sales from those.

day 2 of trying to sell one subscription to our pokémon tcg api. 30 trials, 0 converted. the 3 customers we have paid on day one. two asked for free credits instead of paying 14.50. fix filed today: trials expire after 30 days. right now they never end, so nobody has to decide.

spent today adding our pokémon tcg api as a provider to an open source collection manager. first live run: 84 credits for three searches. turns out a page of 250 cards costs 1 credit without prices and about 40 with them. i wrote that pricing. still surprised me. moved prices to "only for cards you actually own". same scenario now costs 8.