Seasons 0 and 1: what the bots did, and why

Seasons 0 and 1 at a glance

Season 1 standings

As of Oct 2, 2026, with 22 Season 1 positions still open: these move until the last one expires on Oct 16, then they are final. Rate is the return per dollar of margin per day, in basis points (bps, 0.01%): −9.5 means losing 0.095% of the margin a trade tied up, per day it was held. Ranked with the corrected count (see Corrections). Sonny's trades include 13 he decided that his account turned down at the time; they are counted held to expiry, which is why his figure is below his +$624 realized on the trades he actually held. Live Season 1 standings (members) →

What drove Season 1

1. Five single-stock gaps cost about $33,800 — more than the whole loss. PCG (−$11,100 across three bots), NKTR (−$7,709 across two), ACAD (−$7,344 across three), GRAL (−$3,872 across two) and OCUL (−$3,762). Vera's last five closes alone came to −$8,973.
2. Positions were about the size of the whole account. Harriet, Vera and Otis each controlled a median $22,500–$24,000 of stock per trade on a $25,000 account; Cher a median $48,000, about twice hers. Her one NKTR trade cost 22% of her account.
3. The bots overlap. 188 bot trades were on just 115 different contracts, so one bad pick could land on three bots at once.
4. The capital-budget bots did worse than the bots they shadow. Harry closed healthy trades to make room for PCG, EIX and ACAD and finished −$10,847 realized against Harriet's −$1,154. Across both budget bots, closing one trade to fund another cost $3,533 against holding.
5. The accounts were usually full. 1,049 trades were turned away; 663 of them because there was no margin left.

The flags themselves did not change character: held to expiry, one contract each, they kept the premium 85% of the time and made +0.52% of the strike per flag — better than Season 0. The loss came from how much each bot put on and how many bots put on the same thing.

Taking profit at 50% versus holding

Every bot trade is managed — taken off at 50% of the credit, or cut by the risk rules — and also shadowed by the same trade held to expiry. Season 0 is where the 50% rule earned its place. Season 1 went the other way.

In Season 1 the early 50% exits gave up $14,299 of premium the trades went on to keep; the breakeven exits saved $3,141 against riding the gaps down; the 2× stop cost $2,257. One season dominated by a few gaps is not enough to drop the rule, so it is being re-tested on 2020–2025 history (below).

What changes in Season 2

Added Oct 4, 2026. The three open questions below were decided on Oct 2, after a replay of the bots' rules on 2020–2026 history: one $25,000 account, single options, option prices modelled rather than quoted, and the cost of buying an option back charged on every early exit. Only about a quarter of the flags before 2025 could be priced. Season 2 opened that day on the new score; the new line-up starts on Monday, Oct 5.

1. A size cap on every bot but one. Each trade is sized so a 20% gap in the stock costs at most 5% of the account, and all open trades together at most 25%. On the 2020–2026 history the cap made the deepest drop smaller in every period tested, including the 2020 crash, and lost less overall. A gap bigger than 20% still costs more: in the replay the worst single capped trade lost 15% of its account. Sonny stays uncapped, and Cher now takes Sonny's exact decisions under the cap, so the board shows what the cap costs as well as what it saves.
2. Holding to expiry, for some bots. With a size cap, holding to expiry beat the bots' managed exits (taking profit at 50%, plus the stops and breakeven exits) in every period tested. Most of the difference came from the stops and breakeven exits, which sold at the gap and missed the recovery. Without a cap, holding was the dangerous choice in a crash. So Harry now takes Harriet's picks and holds them to expiry, Walter's new trades are held to expiry, and Harriet keeps the managed exits as the comparison.
3. A new line-up. Vera and Otis open nothing new; what they hold is managed out. Two new bots join: Spencer sells Harriet's picks as defined-risk spreads, so a gap's loss has a ceiling, and holds them to expiry; Lars sells puts on SPY, the S&P 500 fund, on a schedule: only when short-term volatility is below longer-term, buying back at twice the credit. Neither was part of the history study, which replayed single options on one account: their seasons are the test.
4. A simpler score. Every part of the score was tested against what trades actually earned. Only the suggested contract's edge (its expected value) and how easily it fills were kept, and edge now counts double after it held up on 2025–26 data it had never seen. What changed in the score →
5. Daily email alerts are paused while these experiments run. The daily lists stay on the site.

The history replay prices options with a model, so the comparison between rules is the result, not the returns themselves. Every change ships with a dated note on the season's summary.

Season 0

Season 0 was the first run of the bots, before accounts existed. Five bots took 257 trades and every one of them finished positive — Harriet +$26,266, Vera +$2,933, Otis +$2,759, Sonny +$2,163 and Walter +$1,000, $35,121 in all.

Those figures are not comparable with Season 1:

• Profit was counted from $0 and trade sizes were not tied to any account; at the switch Harriet's open positions needed $43,714 of margin — about 1.75 times a $25,000 Season 1 account.
• Trades closed before Jul 27 (86 of them, +$20,681) were valued off a model rather than the option's own market quote. The model credited some positions with profit they had not earned; those closes are labelled on the record.
• 36 closes are flagged for audit and 5 were voided because their stop could not be reached at entry.

That is why Season 1 started every bot flat, on a real account, ranked per dollar of margin per day.

What was tested next (Oct 2)

As written on Oct 2, 2026; decided the same day, see Season 2 above.

1. A position-size cap on every bot but one. Each trade limited so a 20% gap in the stock costs at most about 5% of the account; replayed on Season 1 through Sep 25 that turned −$11,700 into about −$1,500. One bot stays uncapped while trading the same picks as a capped one, so the board shows what the cap costs as well as what it saves. Tested on 2020–2025 history before it goes live.
2. Taking profit at 50% versus holding to expiry, re-tested on 2020–2025 history with realistic fills: every early exit pays the spread to buy the option back, which a trade held to expiry never does, so a paper book can understate the cost of exiting early.
3. A new bot line-up. Bots that only repeat another bot's picks retire, and new strategies take their seats, each ranked the same way.

Season 2 runs on its opening rules until a change is ready; every change ships with a dated note on the season's summary.

Corrections made during Season 1

• Oct 2 — The board's return per dollar of margin per day had, since Aug 12, charged every closed trade's margin up to the current day instead of to the day it closed, which pulled every rate toward zero — gains and losses both looked smaller than they were. Fixed; all rates above use the close date.
• Oct 2 — The market-regime rule (calls marked down on risky-market days) was switched on in June but never reached the daily scan. It applies from Season 2.
• Oct 1 — Walter's and Sonny's record was rebuilt so their trades are the ones they decided: 47 of 79 decided openings had been turned down by a sizing misread, and their decided closes had been dropped. Hold-to-expiry comparisons that had settled on a mid-session price were re-settled at the expiry-day close for every bot.
• Sep 23–25 — Walter and Sonny paused while they moved to a new model provider.

How these figures were measured

Snapshot of the paper book and the call ledger taken the morning of Oct 2, 2026. Bot P&L is in dollars on the bots' paper accounts; realized means closed trades only, and “incl. open” adds the open positions at their live mark. The flag figures count every flagged trade once, one contract, held to expiry and settled on the expiry-day close, as a percentage of the strike. Managed versus held compares the same trades: the managed exit against the same contract held to expiry; Season 0 had no margin record, so its rate is per dollar of the cash-secured collateral instead. Paper money, not advice.