about
fifteen, in barcelona. i build things and then try to prove they don't work. execution got cheap; knowing whether the thing works didn't, so the scarce good was never building, it's verification.
i don't kill things cheaply. the ones i've called dead were called after i'd spent everything trying to save them, and the verdict is in the repo with the code. english, spanish, catalan, russian and french.
i'm fifteen.
the worst call i've made was ignoring someone who had earned the right to tell me i was wrong. i knew he was right while he was saying it. i'd spent a year telling anyone who'd listen that i'd rather be corrected than agreed with, and when it finally cost something to mean it, i didn't. fixing it cost more than the argument was worth.
so i stopped trusting the instinct and started trusting the check.
execution got cheap. knowing whether the thing works didn't. anyone can produce the thing now, and nobody made it cheaper to find out whether it's right, so the scarce good was never building. it's verification. most people are still pricing the wrong half.
the obvious fix fails. stack three models to check each other and you don't get three opinions, you get one opinion in three costumes. they read the same internet, so they're wrong in the same places. more auditors makes confidence climb faster than accuracy, which is the wrong direction to move. three systems agreeing is weaker evidence than one test that runs.
so i only trust ground truth from outside. a number, a market, a person, a relationship. with something to lose. when something agrees with me easily i file it as unmeasured, not confirmed.
i'm good at catching a system telling me what i want to hear. models, rooms, coaches. i also know exactly what that skill is: a detector, not a corrector. it tells me an answer is unreliable. it never tells me what's true, and it goes quiet precisely when someone is confidently wrong about something i like the sound of. that's why i build instead of only arguing.
what i want to be right about is markets, machines, defence, and most of all, people. and even more where those meet, because that's where being confidently wrong is expensive and somebody has to answer for it.
i don't kill things cheaply. the ones i've called dead were called after i'd spent everything trying to save them. that's what makes the verdict worth reading.
i read rooms fast. that isn't a gift, it's a repair. i've been the new one in enough of them, in enough languages, that i learned to model a room before it could decide about me.
everything i build removes my need to ask anyone for anything. what i actually want is to be inside something and be the reason it works. i'm aware those point in opposite directions.
i started now because the experience curve is flat. nobody has ten years with these tools. this is the only stretch of my life when being early and being fifteen are the same fact.
there's a phrase i stole and never gave back. an appeal to heaven.
it's what's left when two people disagree and there is nobody standing above either of them to settle it. no ref, no teacher, no adult. you just play it out, and whatever happens is the answer.
i think about that more than is normal. almost everything i actually care about works like that. nobody is coming to save you. that reads bleak for about a day and then it becomes the most freeing sentence you'll ever get.
i shoot alone more than i actually play. it's quieter, nobody argues with you, and it's the closest i get to the same idea in a gym.
i ask myself what would have to happen for me to be wrong. sounds like nothing. it's deleted about half my opinions, and the half still standing i'd actually fight for. and when i commit to one, i finish it.
i'm the easiest person in the world to lie to, and i'm always in the room. i'd tell you i'd notice. everyone says that. that's the tell. i've made my peace with it.
the clever version loses to the boring version with more work behind it. every time. i've hated this since i worked it out and i still can't find the hole in it.
i learn things fast and i'm bad at what comes after. getting good in a month is easy and it's not the interesting part. month seven is where people find out who you are.
i'm bad at chess and i keep playing it. two hundred games, no better than when i started. it's the only thing i do purely because i like it, which probably makes it the most honest thing on this page.
most of what i build dies. nine public repos, a couple of them actually useful. that's the ratio and i'm fine with it. i try to finish things. i don't always.
i speak three of my five languages badly enough to be funny in them.
i talk too fast. people ask me to start again roughly once a day.
i'm fifteen, so none of this is settled. it's a direction.
2018 · no home base until florida. travelling around asia and the middle east, which is the leg this map cannot hold.
skills
- reverse engineeringtaking apart software that shipped without its source, then building something better out of what's inside it. the writeup for one of them is public and you can check it against the binary yourself. it's not only software. most things worth understanding are something somebody else already built.
- social engineeringpeople are the attack surface nobody hardens. what a room rewards, who actually decides, what it wants to hear. it's the same move i run on software.
- resolutionreading a transaction graph until two addresses resolve to one person. about an hour's work, a decent hit rate, admissible nowhere. the same read works on rooms and on systems. that part i actually enjoy.
- measurement designbuilding the test that can embarrass me before i've spent anything on the idea. two of my projects died that way and both write-ups are up.
- languages i write in · node · python · javascript · java · rustenough to read what an agent hands back and notice when the numbers don't add up. that's the bar i need rather than fluency, and i'd rather say so than imply otherwise.
- running work in parallelseveral attempts at one problem, then cutting the branches that lose. the cutting is the skill. launching is free.
- management and coordinationon most projects there's something i can't do myself. on sts i found the people who could, then ran the work through to a verdict. i hadn't met any of them before. building something with people who are as into it as you are is the part i want more of.
- time managementbasketball, school, my own projects and a social life. all four want the same evenings. the skill is saying out loud which one loses this week, instead of letting all four go badly.
- five spoken languagesthree studied, one off a town, one off my friends.
now
- halt · the browser that links ai agents and humansthe wrapper failure rate is the pitch, not the engine. measuring it properly, on more than one draw, is the work i'm on.building
- barcelona high school · grade 10american curriculum, class of 2029. two aps this year, computer science a and cybersecurity, both sat in may.2026/27
- english · spanish · catalan · russian · frenchthe russian is sketchy but it works, picked up from friends rather than studied.
projects
@cadaquesenc- Halta browser built for agents. lightpanda speed with live visibility. pixels are for humans, never the agent's input.live · private · 2026
- Quiretypora is closed source, so i took it apart. the teardown is the public writeup of how it actually works, and quire is the open replacement i built from what i found. it replaced typora on this machine on 7 september and hasn't been switched back. both repos public.live · 2026
- WeLda wallet manager for solana. point a set of wallets at one contract address and trade them as a single instrument. encrypted keystore, every leg built and simulated before anything is sent, and it tells you which wallets an operation links before it fires. zero runtime dependencies.retired
- Backstopfading liquidations on hyperliquid. i learnt this by doing it: measured the reaction window on live cascades instead of assuming one, and it's real, 1.771 seconds minimum. then priced the seat as a book and it still doesn't pay. the measurement stands. the trade doesn't.killed
- StSwallet graph and ev engine for solana. graded no-go: −7.76% a trade, −0.86% even at zero fees. thirty agents verified it across four waves, every entry signal negative. the signals detect volatility, not direction.killed
the dead ones are here on purpose. sts shipped as a public repo with its own no-go verdict inside it. a dead end you can read beats a demo you can't.
basketball
FCBQ profile- jac sants, barcelonaguard. around sixth in catalonia. i came off a higher-ranked club with offers and picked this one.2026/27
- princeton · harvard · yaleexposure camps, plus the program in greenpoint, taught by ex-point guards out of houston.summer 26
- ncaa division igetting recruited to it out of europe is hard, which is the entire reason for the summers.the target
want
- software engineeringnot a quant, and not the guy in the chair. i want the version of this job where the person doesn't disappear into the machine.wrong if: a year of doing it where the interesting decisions all turn out to be the machine's, and mine are paperwork.
- the overlap, not one fieldthe markets work is what's public now. the defence side is one of my biggest ambitions and i haven't built it yet. security is the way in to it, not a separate thing i want.wrong if: shipping one defence thing and finding the markets half taught me nothing that transferred.
- my own companymy father is an entrepreneur and i grew up watching it. i want the job first, so i learn the industry properly, then the company.wrong if: two years inside a good one where i learn faster than i would have alone, and the company stops being the point.