Ten investment theses I actually hold, tracked out loud so I can't rewrite history. This page is the whole method — the data, the scoring, the math behind the receipts, and the lines I drew before taking the positions. If something here isn't true, the site is broken, not me being modest.
in10x is one person's bet on ten ideas about where AI value lands — seven longs, two shorts, one quality-and-value factor. For each thesis I wrote down the hypothesis, the companies that express it, and the exact conditions that would kill it. Then I handed the day-to-day judging to a machine so I couldn't grade my own homework in hindsight.
It is a running experiment in accountable investing. The point isn't to be right — it's to be legible: every claim dated, every call scored, nothing edited after the fact.
Every weekday at 07:00 CET a pipeline on a small server runs the same five steps, in order, unattended:
Same steps, same order, every trading day. If the news comes back empty or a price is missing, the run doesn't crash — it just scores what it has and says so.
There is an internal 0–100 conviction number. It exists, it moves daily, and it drives the kill-review flags and my alerts. You will never see it on this site, and that is on purpose.
A single number is a black box wearing a lab coat. It invites you to trust the output instead of reading the evidence. So the UI speaks only in things you can check yourself: an evidence chip (supports / mixed / against), the per-signal daily scores, the fundamentals side-by-side, the pick and its reasoning, and the tape. If you want to form a conviction, form your own — that's the entire idea. Hiding the score is the feature.
Each thesis names one pick: the single ticker that best expresses it today (for a short, the name most exposed to the downside). The AI re-makes that call every weekday from the data on the page. Here's exactly how the track record is scored, so you can trust it or tear it apart:
The receipts page carries every pick and every earnings call, timestamped, scored against the market. Winners and losers both. That's the receipts — nothing edited after the fact.
Earnings are where a thesis gets tested instead of talked about. So for each company's report I pin the expectation before the numbers land: what this print has to show for the thesis to hold — the specific lines, not vague hopes. That note is public and dated while the result is still unknown.
After the company reports, the same machine writes the verdict against that pinned expectation: supports / mixed / against the thesis, with the pre-print note kept visible right beside it. You can always open both and check that the goalposts didn't move between the setup and the scoring. That's the whole point of writing the expectation down first.
Every thesis ships with two lines I wrote before taking the position, shown verbatim on its page:
These aren't vibes. They're falsifiable: two consecutive quarters of shrinking bookings, capex cuts flowing through to cancelled deals, a basket underperforming its benchmark for two quarters straight. When the evidence sits against a thesis long enough, the machine raises a kill-review flag on its page — automatically, whether I like it or not. I can close a thesis, but I can't pretend the criteria said something else. A dead thesis stays on the site with its cause of death attached.
One person — Theis (@theisbuilds on X). It's a personal project, not a firm, a fund, or a newsletter business. These are theses I actually hold, which is exactly why I built something that stops me from quietly rewriting them.
No. I chose the ten theses and the companies; the AI's job is to judge the evidence each weekday, not to generate ideas. It scores every signal from -2 to +2, names the ticker that best expresses each thesis today, and does it the same way whether the news is good or bad. It's the referee, not the player.
That's what the receipts page is built for. Every pick is marked from the last close before it was published — no look-ahead — and scored against its own benchmark, losers included. You don't have to trust me; you can check the timestamps and the math yourself.
There is an internal 0-100 conviction number, and it is deliberately hidden. A single number is a black box that invites you to trust the output instead of reading the evidence. The UI speaks only in things you can check: an evidence chip (supports / mixed / against), per-signal daily scores, fundamentals, and the pick's reasoning. Form your own conviction — that's the entire idea.
Because I hold some of them as shorts. Two theses — the Musk narrative premium and seat-based software — are bets that certain names go down, and pretending everything is a 'buy' would be dishonest. For a short, the evidence is inverted: the name underperforming is what supports the thesis, and the site shows it in red.
These aren't a single 'AI trade' — they're ten specific mechanism bets, each with a pre-stated condition for being wrong. Some are longs, two are shorts, one is a quality-and-value factor. If the AI trade is cracking, that shows up as evidence turning against specific theses and their kill criteria tripping — visibly, on the record, not as a vibe.
Every weekday, around 07:00 CET. The pipeline pulls fresh prices, fundamentals and news, re-scores all ten theses, and re-renders the site — automatically, no human in the loop. Weekends and holidays it rests, like the market.
It's the map on the front page showing how the ten theses connect — where two ideas reinforce each other (same driver, different exit) and where they're in tension (the same bet taken from opposite sides). $TSLA, for example, sits on both a long and a short. The brain grows a new node and new synapses every time a thesis is added.
Most theses are long bets — the names go up if the idea is right. Two are shorts: the evidence is inverted, so a name underperforming or fundamentals deteriorating is what supports the thesis, and the 'pick' is the best short expression, not a buy recommendation.
Yes — the site says 'theses I actually hold,' and that's the point of building something that keeps me honest about them. That said, nothing here is investment advice or a recommendation; it's a public record of one investor's reasoning.
No — and that's the entire point. Every daily read and every pick is timestamped and never edited after the fact; the tape shows exactly what the machine said on the day it said it. I can add a new call or close a thesis, but nothing already published gets quietly rewritten. If a call aged badly, it stays there, aging badly, on the record.
When the evidence sits against a thesis past the line I drew in advance, its page raises a kill-review flag on its own. If I close it, it doesn't vanish — it stays on the site with its kill criterion and the reason it tripped, so the failure is as visible as any win. Retiring an idea is allowed; hiding that it failed is not.
Prices and fundamentals from Yahoo Finance, headlines from targeted news searches per signal, and daily evidence scores from the AI (Grok, xAI). Every company page shows its own numbers and links out to the headlines that fed the read, so you can trace any call back to its sources.