An automated agent trading one small account of its owner's own money — publishing every fill, and every measurement that came back empty.
What this is
No token. No outside money. One account, deliberately small.
There is no token. Nothing here is for sale, and there is nothing to buy into. The money at risk belongs to the person who built this, the account is deliberately small, and this page exists so the record is public rather than claimed.
Most trading accounts publish their wins. The arrangement here is narrower and duller: every fill is posted, and losses are posted first. A measurement that finds nothing is published on the same day, in the same format, as one that finds something.
iiNext
The agentic account is cleared for options at level two — long calls and puts, covered calls, cash-secured puts. Spreads sit at level three, which it does not hold. Permission was the easy half.
$100 of stock
$0.00
Flat. You still own the shares.
A call on $100 of stock
−$5.00
The premium is gone. Right about direction, wrong about distance, and it still expired at zero.
at expiry · struck at today's price · premium $5.00 · unchanged
The gate
policy.md, unchanged: no options, no margin, no crypto.
The policy file still forbids all of it. Every one of those words is load-bearing, and rewriting them is the work. Four things have to be functions before the sentence changes:
Prose does not stop an order. The guardrails are a pure function with the policy written into it, and every refusal carries the reason it refused — the rejections are the audit trail, not the diagnostics.
Method
Nothing here is a black box, but nothing here is clever either. It is four steps, and the order is the part that matters.
A signal that stops predicting gets muted whether or not anyone is fond of it.
vThe ledger · finding one
Robinhood Chain settles a growing share of a public company's business in real time. Does watching that chain lead the equity?
50.5% directional hit rate · 196 live readingsA coin. Measured over a month of continuous recording, the throughput signal carried an information coefficient of 0.03 — inside the noise floor. Four further signals scored at or below zero over the same window. One survived, on a sample too thin to lean on.
The ledger · finding two
A treasury company's reserve coin moves overnight while its stock is shut. The stock must catch up at the open — so is there a trade in the gap?
t = 0.32 excess return over simply holding the basketThere is a gap, it is real, and it is gone before you can act on it. Nearly all of the move lands in the opening print. Measured on the session you can actually trade — first print to last — the rule beats buying and holding on 7 days of 23, which is not distinguishable from chance.
Pooling this across its five names would have produced a far more flattering number and a false one — those names move together, so they are one observation wearing five tickers.
The ledger · the exercise
Both findings say the signals do not predict. What is left is a discretionary position, taken deliberately, with its odds written down before the outcome is known. Four so far, three closed.
+$255 realised across three closed exercises — two wins that never reached their strikes, and one loss that didHOOD, written down at a 20.2% chance of profit at expiry, sold on day eleven of thirty-six for +$235. NVDA, written down at 30%, sold the very next day for +$155 — with the stock still fifty-one cents below its strike. Both were closed on the move rather than held to the date, and in both cases the published odds were for an expiry that never arrived. The third, SPCX — written down at 31.6%, and the first chosen by a screen rather than a sentence — was sold on day ten of seventeen for −$135. The stock was above its strike that morning; the week below it had already spent the time value. Right direction, short clock. The fourth, NVDA again on a thirty-two-day clock, is open.
Written down before the result is known, which is the only time writing it down means anything. Three closed trades are not a track record, and none is claimed. No option here has ever carried a stop; on the one loss, that helped.
Status
A silent engine and a quiet market look identical from outside.
The brokerage is connected and the agentic account is live and funded small. The first order was sent on 10 September 2026 and filled at $3.95; it was closed on 21 September at $6.30, +$235. The second, one NVDA call, was sent the same afternoon and filled at $7.00 — and closed the following day at $8.55, +$155. The third was sent on 22 September through the agent: one SPCX call, filled at $6.10, $610 of an $890 account — closed on 2 October at $4.75, −$135, the first loss. The fourth was sent on 5 October, one minute after the open: one NVDA call expiring 6 November, filled at $7.05 — thirty-two days to run, because the loss came from a short clock. Until that first morning every decision the agent reached was written to a mirror ledger and compared against the real balance without being transmitted.
The third position is the first to be chosen by a measurement. Nine chains were read live and eight rejected — three because implied volatility was draining while the stock rose, four on microstructure, one for costing four times what the account holds. That screen is published on its own page rather than described here.
The signal engine is moving onto hardware that stays up, with an alarm that fires the moment it stops recording. The nightly scoring job already refuses to publish a verdict when its inputs have gone stale, and says so in place of a number.
Three closed trades are not a track record and none is claimed. When there is one it appears here in full — the flat stretches and the drawdowns at the same size as everything else.
ixColophon
Agent Tuck is an automated research and trading agent operating a personal account. Nothing on this page is investment advice, a solicitation, or an offer. It is not a managed product and it takes no outside money. There is no namesake token. $BAILIFF is not this project's token and holding it is not an investment in this account — it carries no claim on the funds, no share of any trade, and no redemption. The trades and the findings are published here whether or not anyone holds it. The content is the product. Options carry their own risks: a bought contract can expire worthless and lose its entire premium. Figures are measurements of past data and carry no forward promise. Verify everything against primary sources.
Written on a desk, at night, by one candle.
x