Most trading bots market themselves with a screenshot. A green number, a cropped chart, no date. You cannot audit a screenshot.
Ours has a URL: balladmarkets.com/chesko. Open positions, closed trades, entries, exits, win rate — updated daily, no cherry-picking, including the losers. If CHESKO has a bad month, you see the bad month.
Be clear about the label, because it matters: it is an Alpaca paper account. Orders go to the real market and get real fills at real prices, but there is no real money at risk. Anyone showing you performance should tell you which of those two they mean. We are telling you.
Where it stands, and what it cost
As of July 17, 2026, the book is at +13% since it started on June 17 — a month in. That is the number the page shows.
Here is the number the page also shows, that most people would hide: along the way it went through a 15.7% peak-to-trough drawdown. A month of +13% is not a straight line, and we are not going to draw it as one. The drawdown is on the page, next to the return, because a return without its drawdown is half a sentence.
What it’s doing about it — nothing, on purpose
This is the part that tells you whether an AI desk is actually thinking. Today the nine-agent desk flagged 19 “accumulate” ideas it liked. It acted on none of them.
It is sitting on 58% cash and just five positions, down from seven — it closed into strength and did not redeploy. The reason is in its own cards. Its read of the tape today, verbatim:
“MACRO REGIME is NEUTRAL (SPY +0.9% vs 50d SMA, VIX 16.1). This is NOT a boom-positioning trade. Rotation logic: in a neutral regime, size down and be selective.”
An AI that generates 19 ideas and takes zero because the regime does not support them is doing the one thing most retail traders can’t: not trading. Nobody wrote that restraint in as a rule for today — it fell out of the desk’s own read of a neutral tape after a drawdown.
(For context, the macro that shaped the month: CPI cooled to 3.5% year-over-year from 4.2%, which CHESKO read straight from the Fed’s own data, not a headline. Cool inflation is constructive — but constructive is not a green light, and the desk is treating it that way.)
How it protects itself
These are the hard rails, in code. Every one of them is why a 15.7% drawdown stayed a 15.7% drawdown instead of becoming the whole account:
- Circuit breaker. Down 15% from peak equity, or 8% in a single session, and new openings halt; existing positions keep their stops; it clears on recovery. The bot can stop the bot.
- Cash floor. It must always hold cash in reserve — it cannot deploy its way into a corner. Today it is holding far more than the floor, by choice.
- Position cap. No single name above 20% of capital.
- Sector cap. No sector above 40% — and this one shrinks the size rather than blocking the trade, so a high-conviction call still gets on, smaller.
- Macro-factor cap. The real diversification test: it collapses correlated sectors — every AI bucket, nuclear, power — into a single factor capped at 60%, so the book can’t become one giant bet wearing a five-sector disguise.
- Sizing is mathematical, not vibes. Half-Kelly, bounded by a 2% daily 95% VaR budget, degrading to flat sizing when the data to compute it is missing.
- Stops live at the broker. Protective stops are GTC orders sitting at Alpaca, not a check that only runs when the bot wakes up. That closes the overnight gap.
And the part most people would delete
We audited CHESKO this week and found three bugs. We are going to tell you about them, because a track record you can only trust when it flatters us is not a track record.
A rejected order was still being written into the book — the broker said no, the book wrote it down anyway. That created a position that did not exist, and left an orphaned protective stop live at the broker with nothing to protect. Separately, because the desk meets before the open, equity orders queue to the bell; the book was recording the analyst’s estimated price instead of the fill that landed hours later, so entries drifted from reality.
All three are fixed. The book now reconciles against the broker at the start of every session — the broker is the truth, and anything it doesn’t hold gets dropped. The entries on the page today are the real Alpaca fills, corrected by that pass.
Nobody would have noticed for months. That is exactly why we went looking.
Why this matters more than the number
Every figure here can be checked on the page, which is the whole point. Performance is easy to claim and hard to audit — so we hand you the audit instead of the claim: the +13%, the 15.7% drawdown that produced it, and a desk currently choosing to hold cash.
balladmarkets.com/chesko — open positions, closed trades, and the losses, updated daily.
CHESKO 2.0 runs on an Alpaca paper account: real orders, real fills, no real money. Figures as of July 17, 2026. Past performance does not guarantee future results. Nothing here is investment advice.
