01
A proven edge
Not a backtest fantasy — a live, discretionary strategy with a real track record we can learn from.
An ML system is learning a real, proven options-income strategy — one trade at a time — toward a single goal: run it autonomously, with a discipline a human can't sustain. This is the honest log of how it's going.
* Early signal in research. Coin-flip = 0.50. We don't ship a model on "promising" — see the standard.
Behind BASED is a real, multi-year-proven options-income strategy run by a professional trader. It works — quietly and consistently. The problem with any human edge: it doesn't scale, and it gets tired. So we're doing the hard thing — capturing that edge as data, teaching a machine to recognize it, and stress-testing whether a model can run it with more discipline and zero emotion. If it works, BASED is attached to something that actually produces.
Not a backtest fantasy — a live, discretionary strategy with a real track record we can learn from.
Every trade becomes a labeled example: what was true at entry, and how it resolved.
ML finds the patterns that separate the wins from the rare, painful losses.
The goal: a system that sizes, enters, and — critically — exits with rules a human emotionally can't.
We don't ask the model to invent a strategy. We ask it to learn a good one — and to flag the danger a human sometimes rides too long.
Real trade history flows in automatically — clean and deduped.
Each trade described by what was knowable at entry.
The model maps entry conditions to outcomes.
It must survive a brutal honesty test to earn trust.
Only then does it touch a trade — first on paper.
We already built a first model. It found real signal — and it still wasn't good enough to deploy. So we didn't. That bar is the whole point: a model that can't reliably catch the rare big losers isn't useless, it's dangerous.
It scored below the "do-nothing" baseline on out-of-sample tests and caught fewer than half the danger trades. A flashier team ships it and shows you a pretty number. We logged it as a fail and went back for more data. That discipline is the product.
Every trade is predicted by a model that never saw it — the strictest honesty test for small data.
We grade ranking of risk, not hit-rate — accuracy lies when most trades win.
Rare losers are up-weighted so the model is forced to learn the danger, not ignore it.
Strict train / validation / holdout separation, time-ordered — judged on its future, not its past.
We've gone from a handful of hand-logged trades to a clean, automated pipeline ingesting a real book. The model sees genuine structure in the data; it just hasn't earned the keys yet. The gap between "real signal" and "trustworthy" is mostly closed by one thing: more data.
No promises — but it's worth being honest about the size of the ambition.
A disciplined system runs a real, proven options edge end-to-end — tireless, emotionless, around the clock.
BASED stops being "just a coin" and becomes the banner over something that actually does something.
Every milestone, every fail, every win — shared. Holders aren't spectators, they're early.
More data → a sharper model → more capability. The flywheel spins one direction: forward.
We treat the treasury like adults. No reckless pumps, no team hoarding the supply — steady, responsible growth that scales with the project.
Early buybacks are deliberately modest. We don't want to artificially inflate the price or have the team own too much of the supply.
As the market cap grows, buybacks grow with it — proportional and sustainable, not a one-time stunt.
Funds send the founder on the road to onboard everyday people into crypto — setting them up with a Phantom wallet and putting BASED in their hands.
A measured slice goes to getting BASED in front of the right people — growing the community all of this is built for.
Own little, build loud, grow real. Small buybacks now protect holders from a fake-pump dynamic; scaling them with the cap means the treasury's strength grows with the community, not at its expense.
Turn a real, proven strategy into a clean, growing dataset with automated ingestion.
Train models, test them honestly, refuse to ship anything that can't beat baseline and catch losers.
Let a validated model call trades in real time with zero money at risk, graded live.
Small, human-in-the-loop sizing — the model proposes, a human confirms, every decision logged.
A disciplined system that runs the strategy end-to-end — the engine BASED is built around.
A team that can't name its own risks hasn't thought hard enough. Here's how this could fail — stated by us, first.
A strategy that worked for years can stop working as markets change. We watch for it and will say so plainly.
Small data + a flexible model = learning noise. It's the exact reason we already rejected our first model.
A model great on history can underperform with real costs and slippage. That's why paper-trading is mandatory.
If a model never clears the bar, it never goes live. We'd rather ship nothing than ship something dangerous.
No promises — a dated record of real work. It only moves when something real happens.
A real trading book now ingests automatically — clean, deduped, growing weekly. No more hand-typing.
From a handful of examples to a 100+ trade corpus — the single biggest driver of model quality.
We set the "do-nothing" bar, trained a model, tested it honestly — and rejected it for not clearing it.
Folded in independent trade histories from additional accounts. The model's signal jumped (AUC 0.74 → 0.84) and it now flags 76% of losers, up from under 40%.
It's now a genuinely useful danger-detector — not yet a deployable trader. More loss examples keep sharpening it.
No "guaranteed." No "breakthrough." No cherry-picked stats. We report the bar and whether we cleared it — wins and fails alike. If we ever sound like we're hyping, hold us to this paragraph.
No. This is a research project built in public — not financial advice, not a promise of profit. We share the process and our honest results, wins and fails alike.
The edge is the asset. We'll show you the rigor, the milestones, and the honest scorecard — the specific mechanics stay in the lab. That protects holders, not hides from them.
BASED is the community and the banner over a team that actually builds. This is the flagship of what "DevBased" means: ship real things, in public, with receipts.
This page updates as the model and dataset grow. Watch the phase indicator and the dataset count — they only move when something real happens.