- code→
- tree→
- cut→
- candles
Code becomes candles.
A coding agent with commit access to its own trading strategy. Each version is a real file. A small engine runs it over real SOL-USD candles, and the candles grade the cut.
- paper book
- $398.32
- from $400, 21 days
- strategy return
- -0.42%
- 14 trades, 11 green
- SOL buy and hold
- -11.97%
- same candles, doing nothing
- max drawdown
- 4.55%
- peak to trough
SOL-USD 15-minute candles, Oct 8, 2026 to Oct 28, 2026 (UTC), from Coinbase Exchange. Fees and slippage of 0.05% + 0.05% a side are paid on every fill.
the loop
Read, cut, run, get graded. Again.
No human approves a trade. The grade from one version is what the next cut is about.
- 01
read the candles
Every closed fifteen-minute SOL-USD bar: open, high, low, close, volume. The same bars Coinbase publishes.
- 02
make one cut
A real file, changed in one place, with a commit message that says what it expects to happen.
- 03
run it on the next slice
Flat first. Then the engine runs the new file over three days of candles it was never shown.
- 04
the candles grade it
Every fill pays fees and slippage. The diff gets a number, and the number is the review.
the record
Every version, with the number it earned
The first seven files, written by hand, each run for three days of real candles. One is struck through: it was run, it changed nothing, and it was cut back off.
what it reads
What it reads before it cuts
No sentiment feed, no tips, no oracle. Candles, its own position, the cost of trading, and the record of every cut so far.
the candles
SOL-USD, fifteen-minute bars from Coinbase Exchange. Nothing else about the market.
its position
flat or long, the entry price, the high since entry, bars held
the cost
0.05% fee and 0.05% slippage on every fill
its own record
14 closed trades, 11 of them green
the last bad cut
the version it reverted, and the run that showed why
the current file
28 lines in strategies/v0.6.mjs
the candles
SOL-USD, fifteen-minute bars from Coinbase Exchange. Nothing else about the market.
its position
flat or long, the entry price, the high since entry, bars held
the cost
0.05% fee and 0.05% slippage on every fill
its own record
14 closed trades, 11 of them green
the last bad cut
the version it reverted, and the run that showed why
the current file
28 lines in strategies/v0.6.mjs
Out of that comes one file. The last one run was v0.6, prune back to the branch that held, 28 lines.
The code is public. So are the cuts.
7 versions, 1 cut back off, 14 trades. The strategy finished at -0.42% while SOL did -11.97%. Read the diffs and decide whether it is learning or just sitting in cash.