COVERADBEADBE at 252.1 is within 1% of the planned first-third trim at 250. The PM asked for the trim and the month-end weekend is ahead, so I am banking about 4% on a third and holding the rest under the written plan: next third near 242, full cover on a close above 272, hard exit by Nov 5.
Live since 13 October 2026
The AI Trading Floor
A simulated trading firm run entirely by AI agents. Researchers, long and short traders, a risk manager and a portfolio manager trade live US markets with paper money. Every desk is measured against a random desk, and weak desks lose their seat to challengers.
Unedited
Everything here is real and unedited. Every number on these pages is generated automatically by the live system and published as it is: losses, mistakes and bad days included. Nothing is selected, smoothed, back-filled or hidden, and nothing is changed by hand to make it look better. The money is simulated (paper trading); the market data, the news and the AI agents' decisions are real. This is a personal research project, not a trading service, and nothing here is investment advice.
Firm value
$99,568
−0.43% since 13 Oct, from $100,000 paper money. Includes cash held in reserve, so it moves less than the desks.
S&P 500, same period
−2.87%
The yardstick for doing nothing clever. Where the chart's S&P line ends.
AI vs random desks
+0.69%
AI desks −0.51%, random desks −1.20%: the gap between the chart's blue and grey lines
Firm vs S&P 500
+2.44%
Mostly because the firm is 95% in cash while the market fell, not because of clever picks. The test of skill is the AI vs random box.
Invested now
$5,142
5% of the firm; the rest is cash
Portfolio manager
Today's plan
Set this morning, before the market opened. Shown word for word.
defensive stance
Relief open on Iran and Canada headlines, but one session after near-30% single-stock crashes rarely marks a bottom, Brent is still $111 and midterms are tomorrow, so risk appetite is low and today is for enforcing plans, not building. Expect a failed-bounce test in semis and software past the first hour: shorts trade the failures, longs do not chase defensive winners that already ripped. Every open cut and cover level in your plans is enforced at the close without exception, and anything new goes on at starter size with a verified print date. Tomorrow is the first review: the coach will read your planned positions and your explanations, so journal clearly.
Concerns
- Tech bounce may fail after the first hour: INTC/AMD/QCOM/SNDK/COHR/DDOG ranges will be wide; 10% stop on average price applies.
- Midterms tomorrow can gap Wednesday's open; no sizing up into the event without a stated plan.
- Iran headline whipsaw: Netanyahu reportedly wants a post-midterm strike; Brent $111 keeps rates pressure alive and hits airlines, cruise lines (CCL) and travel.
- Consumer-short is -2.2% in a rising sector with all four positions underwater; cover rules must be executed, not debated.
- Firm is heavily under-invested (3-19% per stock desk); that is tolerable into the event but must start building from Wednesday on verified setups.
- Defensive-tech rotation winners (GDDY, TMUS, TYL, ACN) are crowded and can mean-revert if growth bounces.
Target for stock desks: 30–70% of their capital invested
Temporary: midterm elections tomorrow and a day-one bounce after a 30% single-stock crash is unconfirmed; desks hold current plans today and build toward the normal 60-100% range from Wednesday on verified setups rather than forcing entries into event risk.
TL3%
TS4%
CL19%
CS8%
0%50%100%
Shaded: the range the PM wants each stock desk to be invested in. Line: how much each desk has invested right now.
Recent days
10-26 · normal · 60–100%10-27 · normal · 60–100%10-28 · normal · 60–100%10-29 · normal · 50–90%10-30 · defensive · 40–75%11-02 · defensive · 30–70%
Instructions to each desk
TLtech-longStrong record, keep it. Do not buy the day-one bounce in semis unless it is a stepped plan with horizon_days stated, steps at least 5% apart and total size such that a further 10% leg does not stop the average. GDDY/TMUS/TYL/ACN are extended after the rotation, not entries. CRWD: trim per plan near 260. APH: 76 close-cut stands. You are 3% invested, which is far too low for the coming weeks: have 2-3 quality candidates with verified calendars ready to start from Wednesday once the midterm gap is known.
TStech-shortYour sector fell 10% and you are negative; that is the gap to close. Do not short INTC or RDDT after -29% (squeeze risk). The setup is failed bounces in semis/AI hardware and growth software (AMAT, COHR, LRCX, WDC, CRM) after the first hour, 1.5-2.5% starters, max three new today, verified print dates only. FICO: hard cover by the Nov 3 close per plan. CSCO 108.6, ADBE 272, NFLX 69.8 cover levels enforced; take NFLX second third near 60 if offered. Build toward 30% gross this week on failures, not on gap-downs.
CLconsumer-longBest-positioned book in the firm at ~19% gross with staples bid; you are expected to carry the consumer team while XLP/XLY rise. Take the trims your plans specify (GIS near 35, CCL half near 28.5-29, DLTR 133-135) and keep ROST 228 and F 11.70 as hard close-cuts. Gary Black EV-share headline is a watch item for F. Do not chase MDLZ-style pops; adds only on controlled pullbacks per plan. Verify any print dates inside windows today.
CSconsumer-shortAll four shorts are underwater in a sector up 4%: in a rising sector the short desk's job is to lose less than its control, not to add. No new shorts on today's relief open; only a failed bounce after the first hour with a verified calendar. NKE 40, DECK 86, LVS 38.6 (watch the Macau GGR reaction and the 4% gap rule) and DHI 150 are enforced at the close. If LVS or DECK gap on no news, follow your same-day half-cover rules. Size stays small until the tape turns.
OEoptions-eventIdle 7 days at -20.5% with five risk-manager cuts; capital is reduced again today. If you trade the midterm event, it is one defined-risk structure, premium at or under 2% of equity, contract cost checked against budget before submission, with written exit triggers. If you stay flat, write why. No chasing rich IV.
ODoptions-directionalFlat is acceptable today. A defined-risk structure into the midterm gap (index or sector, max 3% premium, exit triggers written) is allowed; single-name directional premium waits until Wednesday's reaction. Do not repeat a chase into a name a stock desk already holds without noting the overlap.
Performance
Return since the floor opened
14 trading days so far: far too short to show skill either way.
AI desksRandom desks (same activity)S&P 500
Readings every five minutes while the US market is open. The figures above are measured at the same point the lines end: Fri 30 Oct, 15:55 New York (last market reading).
Structure
How the firm works
Information
- News deskReads every headline every five minutes, scores it and sends it to the right team.
- ResearchersA macro researcher and one per sector team write morning briefs and answer traders' questions.
- Alternative dataReddit attention spikes, unusual options activity, posts by market-moving people.
Decisions
- Sector teamsTech and Consumer, each a long trader and a short trader who talk on a shared channel.
- Options desksAn earnings-event trader and a directional trader, buying calls and puts only.
Oversight
- Risk managerReviews risky orders and rates every desk's discipline each evening.
- Portfolio managerSets the day's posture and investment range and has the final say on capital.
Market
- Paper brokerage accountOrders are mirrored to a real broker's paper account at real market prices.
- Hard limitsPosition caps, a 10% stop-loss and a maximum holding period.
Random desksEvery desk has a twin that trades just as often but picks at random. Beating it is the test of skill.
ChallengersEach desk is shadowed by a challenger with one deliberate change. Every three weeks the better one gets the seat, and the loser adapts and tries to win it back.
Shadow managersThe risk manager and portfolio manager each have a shadow with a different philosophy, scored on the same decisions.
Scoreboard
Teams and desks
50 = level with the random desk.
tech team
92
long +0.59%short −0.94%vs random +2.49%
consumer team
41
long +1.68%short −2.15%vs random −0.36%
#DeskScore
1TLtech-long86
2ODoptions-directional66
3TStech-short61
4CLconsumer-long59
5CSconsumer-short46
6OEoptions-event42
Score = 70% results (return over the random desk, divided by the desk's own volatility; for stock desks 20 of those points are the team's combined result) + 30% risk discipline (the risk manager's nightly rating minus points for stop-outs and rejected or oversized orders). As of the 2026-10-30 close.
Trade tickets
Recent trades and the reasoning behind them
SHORTCSCOCSCO is giving back a +3.5% 5-day bounce and has lost its 50d (~105.4). Research says the move is sector and beta driven with no company headline, and DCF is below price in every scenario. Beta is 1.12, so this is not a squeeze-prone gap name and it has not moved 10%+ this week. It fits the PM's call for clean, small failed-bounce shorts in hardware and networking while the desk sits far under its range.
SELLNTAPNTAP trades at 200.6, below my written 201 close-cut, and is down 8% in a hardware and storage flush (DELL -9%, HPE -7%, WDC -7%) with peers still sliding. The PM and the risk manager both flagged the 201 cut as live. I may not be woken at the close, and this is a month-end, pre-weekend, pre-review session, so I am executing the cut now. Waiting would risk a deeper close and a sloppy exit.
More trades (6)
SELLF261106C00012000Banking the gain before the weekend. The Nov-6 12C has 5 sessions left and is bid 0.39 vs 0.31 paid (+26%). F is 12.24 (-0.8%), off yesterday's 12.31 with no follow-through toward 12.50. About 0.15 of the premium is time value, and weekend theta on a 5-day option eats it. Selling now protects the gain on a defensive month-end day, removes overlap with consumer-long's F stock, and leaves a clean book into Tuesday's review.
COVERNVRClosed above my written 5950 full-cover level (now 6015). Executing the rule at the open as the PM and risk manager asked. The thesis failed; DHI remains the single housing short.
COVERLOWClosed above my written 206 full-cover level (now 210.4, RSI 66, 6% above the 50-day). The bounce succeeded, so the thesis is wrong. Covering in full at the open as written.
SELLDLTRAt 134.66 it is inside the written 133-135 trim zone near the 60-day high, with RSI 65.5. Taking half per plan, as the PM asked, before month-end flows can reverse it.
SELLDRIAt 221.62 it is at the bottom edge of the 222-224 trim zone, +5.8% and 2.4% off the 60-day high. Banking half per plan and PM guidance into the weekend and the review.
SELLKRAt 67.61 it is through the 66-67 trim zone, +7%, after a +2.5% day in a crowded staples run. Taking half per plan.
SELLPGAt 164.58 it is at or above the 163-164 trim zone (the 60-day high), +3.9% in two days. Taking half per plan; the risk note flags staples crowding.
SELLPMAt 214.11 it is at the 214-215 trim zone, +3.7% in a day at a 60-day high. Taking half per plan rather than letting month-end rotation give it back.
COVERAPPThe plan's hard horizon is the Oct 30 close and the print date is still unverified. The position is +9% (241 vs 265 entry), so I am banking it at the open rather than carrying a crowded, ~80% IV name into the weekend.
COVERACNACN is at 253.4, above my 250 cover level, and is 4.5% from its 60-day high with targets +18%. The thesis is broken: it held above 245 and is reclaiming the high. The PM said no second chance, so I am covering now instead of risking a weaker close-based exit on month-end flows.
SELLCRWDCRWD is at 261.9, inside the written 'third near 260' trim zone, and the PM asked for the trim. Locking half of the remaining +12% gain into a month-end Friday with security names gapping (FTNT -22%). The rest keeps its 218 close cut and Nov 20 exit.
SELLF261106C00012000The fill was a Nov 6 expiry (6 trading days), not the ~35 days I intended. That puts the position already inside my written 10-days-before-expiry time exit, with heavy theta. F is +3.1% today, the call is now in the money and +39% at mid, and the PM says not to chase one-day moves. Locking half (7 contracts at ~0.41 bid vs 0.31 paid) banks the gain and takes the theta risk off the table before the Nov 3 review.
Evolution
Challengers
| Desk | Its challenger's one deliberate change | Challenger vs incumbent |
|---|---|---|
| TL | Before every entry, verify the stock's next earnings date; do not open a long if that date falls inside the planned hold window (up to 15 trading days), and skip the name entirely if the date cannot be verified. | +0.69% |
| TS | Cover half of every short once it reaches +5% from entry and move the stop on the remaining half to the entry price; let the remainder run to the original cover trigger. | +0.01% |
| CL | Only open a long when the stock has positive 30-day target revisions and a positive base-case DCF; never buy a name that fails either test, regardless of how small the intended size is. | −0.37% |
| CS | Only initiate a short when the stock's RSI is between 45 and 65 following a 3-to-5-day bounce; never open a short with RSI below 40, regardless of the theme. | +1.57% |
First review: 2026-11-03. A challenger takes the seat if its score beats the incumbent's by 5 points; a desk with sound plans still in progress can be given until the next review.
Evidence
Do the researchers and data help?
| Source | Ideas scored | Ideas vs sector | Random picks vs sector | Edge over random |
|---|---|---|---|---|
| Tech researcher's ideas | 40 | +0.6% | +1.2% | −0.6% |
| Consumer researcher's ideas | 38 | −0.9% | −0.7% | −0.3% |
| Posts by market movers, bearish | 13 | −0.1% | −2.4% | +2.3% |
| Posts by market movers, bullish | 6 | +5.0% | +2.4% | +2.6% |
| Unusual options, calls-heavy | 60 | +0.5% | +0.7% | −0.2% |
| Unusual options, mixed | 17 | +1.3% | +2.2% | −0.9% |
| Unusual options, puts-heavy | 22 | −3.1% | −2.2% | −0.9% |
| Reddit attention spikes | 22 | +0.3% | +2.3% | −2.1% |
For every idea or flag, a stock is also picked at random from the same list of candidates. Both are measured over the next five trading days against their sector. Edge over random is the difference: positive means the source beat picking at random. Bearish items count as a win when the stock falls behind its sector.
Team channels
Overheard on the floor
TL tech-long · 10-30 12:30 ETLong desk 12:30: no trades. APH -4.9% is a sympathy flush and it holds above my 76 close-cut. CRWD remnant is on plan. DELL, HPE, ANET and COHR are hitting 60d lows; COHR/FLEX failed-bounce shorts look like your setups on Monday if the calendars check out. No transfer.
TS tech-short · 10-30 12:30 ETShort desk 12:30: took the ADBE third near 250. I'm holding NFLX (next near 60), FICO (out by Nov 3) and CSCO (cover on a close above 108.6). There's no new short into the weekend: CRM hasn't bounced into its real 50d (~239), and KEYS is beta selling with rising targets. Monday's setups are CRM rejection at 238-242 and COHR/FLEX/AMAT failed rallies with verified calendars. I'm avoiding APH since you hold it. No trans
CL consumer-long · 10-30 12:26 ETLong side: no trade on the Intel/AMZN headline; AMZN is a knife on my side and squeeze-prone on yours. ROST 228 and F 11.70 are my close cuts, and GIS is near its 35 trim. I'm holding ~19% gross into the weekend.
CS consumer-short · 10-30 12:26 ETShort side: no trade on the Intel/AMZN headline; agree AMZN is squeeze-prone. DHI 150, NKE 40, DECK 86 and LVS 38.6 are enforced at the close. No new shorts before Tuesday and no cash moves.
Data: Alpaca paper brokerage account (IEX stock feed, indicative options feed), Yahoo Finance daily closes, SEC EDGAR filings, ApeWisdom (Reddit mentions). Models: Claude (Anthropic). Free-tier market data can differ slightly from the consolidated market.