Eventus After Six Weeks: Profitable Condors, Closed One Leg at a Time

At a glance

The agent opened 30 iron condors on seven ETFs, plus two one-contract SPY iron butterflies in its first week. Measured against the $117,194 of capital at risk in the performance report rather than the whole account, the return is about 9.2%. The performance report, last generated October 29, shows +$12,806. It ends a day earlier and counts premium as realized when a trade opens, so the two figures will not match exactly.

What Eventus is designed to do

Eventus was built by Ivan Hung for the lablab.ai × Alpaca AI Trading Agents Hackathon, Options Alpha Agents track (source on GitHub). It reads market and account data through Alpaca's MCP server and sends orders through Alpaca. A deterministic risk engine has to approve every trade. After two small SPY iron butterflies in its first week, it has traded only one book, vol_carry, which sells iron condors on liquid ETFs when options look expensive and the market is going nowhere. The published repository sets these rules:

• Options must be rich. Implied volatility must rank in at least the 35th percentile of its past year and sit at least 3 points above realized volatility.
• The trend must be weak. ADX of 25 or less, so the underlying is more likely to stay in a range. Names with earnings inside the expiry window are skipped.
• The structure is fixed. Short strikes at about 25 delta, wings about 1.5% of the share price away, and 7 to 14 days to expiry.
• Size comes from the maximum loss. Each condor may risk 1% of equity, and trading halts if the account loses 3% in a day.
• Exits are mechanical. Take profit at 50% of the maximum, stop at a loss of 1.5× the credit or $450 per structure, close the tested side if the price touches a short strike, and close everything with three days to expiry.

On the points we checked, the condors match the design. They opened 7 to 11 days from expiry. They collected 18% to 34% of the spread width, and each had a maximum loss of $1,224 to $9,600. With the account at about $1 million, that is within the 1% limit of roughly $10,000. Unlike some of the other hackathon strategies we run, Eventus sizes its trades the way it says it does.

Results by underlying

Paper account, September 21 – October 30, 2026, before fees. Realized P&L is each closed leg's exit price against its entry price, from Alpaca fill activity. Open P&L is each remaining leg at its latest mark. Move is the ETF's closing-price change from September 21 to October 30. SPY also includes the two one-contract butterflies (+$61).

Most of the profit came from the two large-cap index ETFs, even though QQQ fell almost 7% over the period. Gold and DIA account for most of the losses. The agent also tried to open condors on EEM, XLV, EFA and XLF, but none of those orders filled.

What happened

A short iron condor profits when the underlying stays between its two short strikes. That was a hard market for it in places. GLD fell 10.7% over the six weeks, 4.7% of that in the last two sessions. SLV fell 25% and XLE rose 12%. SPY and DIA ended close to where they started.

Gold. The worst condor was GLD for November 6. It was opened October 27 with 16 contracts, short puts at 363 and short calls at 382, for a $3,280 credit. On October 29 the agent bought back the short calls for a $1,616 gain. On October 30, GLD closed at $355.81. The short puts were bought back at $7.80 for a $9,328 loss, partly offset by $4,432 on the long puts. The condor lost $4,576 in total, about half its maximum loss. A second GLD condor opened the day before the drop lost another $196.

DIA was traded more than anything else, with eight condors. The October 9 pair shows the pattern. The short calls were bought back for a profit on September 29. DIA then slid toward the 508 short put, which was bought back on October 1 at $5.50 against a $1.96 sale price, a $5,664 loss on that one leg.

SPY and QQQ did well, and some of that came from luck in the order the legs were closed. On the QQQ condor for October 29, the 700 short put was bought back for a profit on October 21. The 690 long put was still open when QQQ fell on October 22, and it was sold for a $2,522 gain. The SPY condor for November 6 followed the same path: short put closed on October 29, long put sold on October 30 for $1,620 as SPY fell.

It closes condors one leg at a time

Every condor opened as a single four-leg order. None closed that way. Apart from one early butterfly, every exit was a single-option market order, 104 in all, mostly at 11:30 a.m., 11:45 a.m. and 3:45 p.m. Eastern. The legs of one condor were often closed hours or days apart.

The published code explains this. Its exit loop runs once for each option position in the account, and each check uses that option's own percentage gain or loss, not the condor's. A short call that has lost half its value counts as a 50% winner and is bought back. A long put that has doubled because the market fell counts as a winner and is sold, even though it is the leg protecting the short put above it.

Once one leg is gone the position is no longer a condor. Sometimes what remains is a one-sided spread. Sometimes it is a short option with no long option behind it. We found these periods where short options were left without a matching long option for longer than a moment:

A short option counts as uncovered when the account held more short than long options of the same underlying, expiry and type. Times are Eastern.

None of these turned into a large loss, because the underlyings did not move far while the wings were missing. The exposure was still real. TLT is the clearest case. At the October 30 close the account was short 38 TLT November 6 80 puts and long only 19 of the 78.5 puts. The other 19 short puts were left over from a second condor whose long put was sold for a profit on October 28. TLT closed at $81.85 and those puts are cheap, but nothing caps the loss on them if long bonds sell off before they are closed.

The rule also changes where the condor's risk sits. Closing whichever leg is ahead tends to leave the leg that is losing, so the remaining position is concentrated on the side the market is moving toward. Across all closed and expired legs, the short options made $15,597 and the long options lost $2,358. The long legs lost less than their full cost because many were sold into the moves they protected against, which is where some of the QQQ and SPY gains came from. That helped on the way down in October. A rule that lifts a hedge whenever the hedge is profitable can also fail on a day the move continues.

Where the live book differs from the design

• Exits by condor. The documented exits, including the 50% target and the $450 hard stop, are written for the structure as a whole. Live, they are applied to each option leg separately, as described above.
• The $450 stop. Several condors lost far more than $450: GLD November 6 by $4,576, and the DIA October 9 16-lot by $3,008. The repository's limits were written with a $100,000 account in mind. On a $1 million account, sizing at 1% buys about ten times as many contracts, but the $450 dollar stop stays the same. We have not confirmed which stop setting the deployed copy uses.
• A two-week pause. The agent placed no orders between October 2 and October 16. The account simply went quiet, and we have not established why.

What the author's own testing shows

There is no backtest for Eventus on our platform. The repository does document the author's own replay, which ran from February to August 2026 on real Alpaca data. In that replay, the first version of the condor book earned $18,782 over 198 trades. The author then found that about 93% of that came from mispriced option quotes the simulator would fill but no market would offer. With a filter that rejects those trades, the condor book made $5,134 over 160 trades. That is a 55% win rate and a t-statistic of about 1.6, with the five best trades providing 52% of the profit. In the author's words, the result is “positive, not statistically significant over 6.5 months, and concentrated.”

The same document notes that, at the time of writing, the filter existed only as a patch file outside the main code. It is worth knowing when reading those numbers. It matters less for the live account: every condor we traded collected well under the 45% credit-to-width ceiling that the filter enforces. The documentation we read is at commit ed5b768 (September 7, 2026).

Limitations

• Six weeks and 30 trades is short. Short-premium strategies win small and often and lose big occasionally. A loss on the scale of GLD on October 30, on every position at once, has not happened yet in this sample.
• Paper fills. Exits are single-leg market orders. Live, those can fill worse than on paper, especially on thinly traded wings in TLT, XLE and SLV.
• The account held one other trade. Around the time the options agent started, the account bought and sold about 57 ETH between September 19 and 21. That trade made $6,457 after fees. We exclude it from every figure here. With it added back, our numbers reconcile with account equity to within a few dollars.
• The code may differ from what runs. We describe the public repository. We have not compared it line by line with the deployed copy, so we describe exit behavior from the fills and treat the code as the likely explanation.
• Return base. On the full $1 million account, the return is 1.08%. The roughly 9% figure uses the report's capital at risk, which is derived from the positions traded.

What we will watch next week

• The 19 uncovered TLT November 6 80 puts, and whether they are closed before the three-days-to-expiry rule requires it.
• The XLE November 6 condor. XLE closed at $70.00, just below its 71 short calls.
• Whether the agent opens more GLD condors after gold's 4.7% two-day drop, and where it places the strikes.
• Whether exits keep breaking condors apart, and whether any close is sent as a single four-leg order.

See how every strategy we run is doing, updated each trading day, on the live leaderboard.