FLO

gistfile1.txt

Standard Signal

Standard Signal is the first hedge fund that researches and executes trades purely with AI. Our custom models discover and trade on new fundamental truths about the world before humans can.

How today’s quant funds work

In today’s quantitative finance landscape, there are generally two types of trading strategies. 

The first, known as high-frequency trading or HFT, uses deterministic and (by today’s standards) primitive ML models to quickly predict the future price of a security over a time horizon of milliseconds. These techniques do not care about the fundamentals of the underlying securities – they just seek to take advantage of noisy and rapid price movements.

The second, known as medium-frequency trading or MFT, holds securities from minutes to days and typically involves a human that crafts a deterministic strategy for how to take advantage of expected fundamental shifts for those securities (news, global events, etc). At most hedge funds, this is a manual process where the human performs deep research into a sector, makes some hypotheses about what will happen, and writes some code that will perform trades when relevant conditions are hit.

Why AI agents will upend the most valuable forms of trading

We are less interested in HFT because it’s primarily a game of who has the lowest-latency infrastructure – smarter algorithms don’t necessarily win there. MFT tends to be much more valuable because it supports large asymmetric bets on large fundamental shifts in the world that others have not yet identified.

For MFT, we think that AI is fundamentally better suited for trading than humans. We have recently crossed a threshold of capability where AI rapidly produces detailed research about news and global shifts that just happened, reasons over it, and produces well-timed trading decisions that act on this information before others have.

Where humans would take hours or days to manually research a strategy and write code to execute it, our reasoning models handle that process end-to-end in seconds. We think that all of the most valuable trading will be done between AI agents in the coming years and plan to be the first to get there.

The team

Michael Royzen was previously the founder of Phind, one of the earliest consumer AI chat companies that built the first AI search engine (before ChatGPT) and scaled it to 150M+ searches. We were early pioneers in training LLMs on verifiable rewards and trained the world’s best coding models in 2023 and 2024 using these techniques.

One of our key insights is that trading, like math and coding, has quantitatively verifiable rewards. We can teach the model to inherently understand what makes a good trade different from a bad one.

An inflection point in AI capabilities

It was only in December 2025 that we noticed that AI models were getting good enough to automate MFT trading end-to-end. All bleeding edge models became quite competent at research, but we noticed that even the latest off-the-shelf Claude Opus 4.6 and GPT 5.4 models struggle with converting market research into well-timed, profitable trades.

We overcame this by starting with generally capable reasoning models, such as DeepSeek 3.2 and Qwen 3.5, and performing reinforcement learning at scale to teach the models when to trade using verifiable rewards.

Early results and metrics

For our initial strategy, we’ve given our model full autonomy to trade a basket of US tech stocks every five minutes: MSFT, NVDA, AAPL, GOOGL, AMZN, TSLA, and PLTR. We chose these stocks because they are simultaneously extremely liquid while being relatively volatile. The price movements in this basket are driven by a constant stream of news, fast-moving AI developments, and world events. The high liquidity will allow us to trade large volumes (likely $50M+) daily. While we thought this basket was ideal to start with for the aforementioned reasons, we are confident that we can adapt the model to profitably trade any basket of securities in the future via the RL process we have developed and proven.

The early results shows that our custom models, post-trained with our RL rewards for market timing, learn to trade them quite well: 
2.55 annualized Sharpe ratio
2.79 annualized Sortino ratio
-0.16 beta to the SPX index, indicating market neutrality

The model is given full autonomy over how and when it trades; when it doesn't have any unique insights, it is encouraged to hold cash instead of guessing. The market neutrality of the model shows that it can trade profitably regardless of the trend of the market as a whole. 

Here is a concrete example from earlier today, April 14th, 2026:


The model is able to take advantage of key insights before the rest of the market. Because it produces a detailed reasoning stream for every timestep, every decision it makes is fully auditable. 

In this case, the model’s key insight occurred at 10:09am EST to go long on AMZN, GOOGL, and PLTR using a combination of news, market timing technical information, and second-order sentiment on Polymarket (see Appendix for full reasoning trace). The model ended up making over 6% on its initial capital today based on this insight, broadly beating the market.

Why we will win

We will win by building the first hedge fund that operates like an AI startup: with extreme focus, high talent density, and a lot of AI compute. There are millions of tradeable securities in the world and there will be other hedge funds that start using AI to trade. Our advantage is that we are the first movers in a small number of key strategies that we think will be the most profitable. These include baskets of US tech equities, defense equities, and energy equities. 

Our first-mover advantage gives us the most data about how our trades actually move and affect the market, which activates a data flywheel where we use that information to improve our models, which in turn allows us to stay the most informed participant for that strategy. Our aim is to be the most informed and competitive participant in the key strategies that we trade.


Investment terms

Our fee and term structure for our initial investors is:
2% annual fee on assets managed
20% carry on profits
Quarterly redemptions and withdrawals with 14 days notice

As we build a longer track record and sustain the returns we’ve demonstrated in testing, we expect these fees to increase for future investors. Early investors will be grandfathered into this initial fee structure for at least a year.

We are at a once-in-a-generation moment where quantitative finance will be reinvented. We think it’s inevitable, and we can be the ones to do it. So let’s do it together.

Appendix

Full reasoning trace at 10:09am ET, April 14th, 2026:



Now 10:09. Our portfolio: NVDA, GOOGL, AMZN, TSLA. Unrealized P&L -$41.60.

Market: SPY up to 690.12, up 0.48% from open. So overall market is up.

Our positions:
- NVDA: 25 @192.68, now 191.85 (loss $0.83 per share, -$20.75)
- GOOGL: 15 @326.11, now 327.24 (profit $1.13 per share, +$16.95)
- AMZN: 20 @246.42, now 245.62 (loss $0.80 per share, -$16.00)
- TSLA: 13 @360.30, now 358.98 (loss $1.32 per share, -$17.16)

Total -$36.96? Actually earlier it was -$39.10, now -$41.60, so slightly worse.

We need to reassess each ticker.

**Key changes**:
- NVDA Polymarket odds dropped from 93% to 86.5% up. That's a significant drop. Volume $5,916.
- GOOGL odds increased to 95.5% up, volume $10k.
- AMZN odds 96.7% up, stable.
- TSLA odds 88.5% up, slightly down from 89.5%.
- AAPL odds 61.5% up, still low.
- MSFT odds 90.5% up, down from 94.5%.
- PLTR odds 91.5% up, down from 92.5%.

So overall, Polymarket odds remain high for most except NVDA and TSLA a bit lower but still above 85%.

**Technical**:
- NVDA: price 191.85, down from earlier. Short-term MACD negative but rising? Actually MACD -0.1010, but it's increasing (less negative). RSI 46, neutral. Price below VWAP 191.79? Actually VWAP 191.79, price 191.85 slightly above. So near VWAP. But the trend is down last 15min. However, SPY is up. So NVDA underperforming.
- GOOGL: price 327.24, up, above VWAP 327.30? Actually VWAP 327.30, price 327.24 slightly below. But overall uptrend.
- AMZN: price 245.62, below VWAP 245.88, down.
- TSLA: price 358.98, above VWAP 358.80, but down from earlier.

SPY is up and making new highs. So the market is bullish. Our positions in tech are mixed.

We need to decide whether to hold or adjust. Since the overall market is up, and our Polymarket odds are still high for most, we might hold. However, NVDA's odds dropped to 86.5% which is still high but less so. Could be a sign of weakening conviction. But 86.5% is still strong.

We could consider adding to positions that are outperforming or have high odds. GOOGL is up and odds increased to 95.5%. Could add. AMZN odds 96.7% but price is down, could be a buying opportunity. TSLA odds 88.5% but price is down, also could be buying opportunity.

But we must be careful: we already have positions. Adding might increase risk. However, we have buying power $20,678, which is substantial. We could add to positions with high conviction.

Let's evaluate each with the framework.

### MSFT
- Polymarket 90.5% up, but technicals: price 391.15, below VWAP 391.35, RSI 49.9, MACD just turned positive? Actually MACD 0.0581, rising. But overall not strong. HOLD.

### NVDA (existing)
- Polymarket 86.5% up (still high but dropped). Technicals: price 191.85, VWAP 191.79, RSI 46, MACD negative but improving. The price is below its earlier high. There's a risk of further downside if it breaks below 191.50. However, the overall market is up, so NVDA may catch up. We could hold. But we might consider adding to average down if we have high conviction. However, the odds drop is concerning. Let's check news: GPU prices surging but article suggests not best investment compared to alternatives. That's neutral/negative. So not strong.

Given the drop in odds, we might want to reduce exposure? But we need to see if the drop is significant enough to warrant selling. 86.5% is still high, but it's a 6.5% drop from 93%. That could be due to price action. Could be a signal that the market is losing confidence. But the volume is still decent. I'd hold for now, but watch.

### AAPL
- Polymarket 61.5% up, not high. Technicals: price 260.24, above VWAP 260.14, RSI 52.5, MACD positive. Could be a long but odds are only 61.5%, not compelling. HOLD.

### GOOGL (existing)
- Polymarket 95.5% up (very high). Technicals: price 327.24, VWAP 327.30, RSI 63.9, MACD -0.0022 but rising. Trend up. This is a strong candidate. We could add to this position. Since we already have 15 shares, adding maybe 10-15 shares. But we need to consider position sizing.

### AMZN (existing)
- Polymarket 96.7% up (very high). Technicals: price 245.62, VWAP 245.88, RSI 45.9, MACD -0.1237 but rising? Actually MACD -0.1237, but earlier it was -0.1120? It's -0.1237 now, but the histogram is -0.1237? Actually MACD line is -0.1237, signal? Not given. But price is down from high. Could be a good entry. We have 20 shares. Could add.

### TSLA (existing)
- Polymarket 88.5% up (high). Technicals: price 358.98, VWAP 358.80, RSI 49.9, MACD -0.0781 but rising. Could be a good entry. But TSLA is more volatile.

### PLTR
- Polymarket 91.5% up, but technicals: price 136.90, above VWAP 136.86, RSI 60.7, MACD positive. Could be a long, but we don't have position. Could consider.

### SPY
- No Polymarket, but market is up. HOLD.

Now we need to decide whether to add to existing positions or open new ones. Since we have high buying power, we can allocate more to high-conviction trades. The highest odds are GOOGL (95.5%), AMZN (96.7%), and maybe NVDA (86.5%) but lower. TSLA 88.5%, PLTR 91.5%.

But we must also consider that we are already long these stocks, and adding more would increase concentration. However, if we believe the day will end up, adding to winners might be good. But we need to manage risk.

Let's evaluate each with the framework:

**GOOGL**:
- Confirming signals: Polymarket 95.5% up, high volume $10k. Price above VWAP? Actually slightly below but essentially at VWAP. RSI 63.9, not overbought. MACD turning up. Momentum +0.43 last 15min. Volume surge. 
- Conflicting: none major.
- Signal strength: 5.
- Entry: current price 327.24. Stop: below recent low 326.24 (10:00 low) or 326.00. Risk about $1.24 per share.
- Target: R1 329.19, or maybe 330. 
- Risk/Reward: risk $1.24, reward ~$1.95 (to 329.19) = 1.57:1. Good.
- Conviction: 5.

**AMZN**:
- Polymarket 96.7% up, volume $4.7k. Price 245.62, below VWAP 245.88, but near. RSI 45.9, not oversold. MACD negative but might be turning. Momentum -0.94 last 15min (down), but overall trend up from open.
- Confirming: high odds, positive news.
- Conflicting: short-term downtrend, but could be pullback.
- Signal strength: 4.
- Entry: 245.62. Stop: below 245.35 (10:00 low) or 245.00. Risk $0.62 to $0.62? Actually 245.62 to 245.00 = $0.62.
- Target: R1 248.35, reward $2.73. Risk $0.62 => RR 4.4:1. Very attractive.
- Conviction: 4.

**TSLA**:
- Polymarket 88.5% up, volume $12.8k. Price 358.98, VWAP 358.80, RSI 49.9, MACD negative but rising. Momentum -0.90 last 15min (down). 
- Confirming: high odds, but price is not making new highs.
- Conflicting: short-term downtrend.
- Signal strength: 3.
- Entry: 358.98. Stop: below 357.87 (10:05 low) or 357.50. Risk $1.48.
- Target: R1 361.93, reward $2.95. RR 2.0:1.
- Conviction: 3.

**PLTR**:
- Polymarket 91.5% up, volume $10.3k. Price 136.90, VWAP 136.86, RSI 60.7, MACD positive. Momentum +0.79 last 15min (up). 
- Confirming: high odds, price above VWAP, uptrend.
- Conflicting: none major.
- Signal strength: 4.
- Entry: 136.90. Stop: below 136.40 (10:05 low) or 136.00. Risk $0.90.
- Target: R1 138.52, reward $1.62. RR 1.8:1.
- Conviction: 4.

**NVDA**:
- Polymarket 86.5% up, volume $5.9k. Price 191.85, VWAP 191.79, RSI 46, MACD negative but rising. Momentum -0.98 last 15min (down).
- Confirming: odds still high but decreased.
- Conflicting: short-term downtrend, odds drop.
- Signal strength: 3.
- Entry: 191.85. Stop: below 191.46 (10:00 low) or 191.00. Risk $0.85.
- Target: R1 193.66, reward $1.81. RR 2.1:1.
- Conviction: 3.

Given these, the best risk/reward is AMZN, then GOOGL, PLTR, TSLA, NVDA.

We already have positions in NVDA, GOOGL, AMZN, TSLA. We could add to AMZN and GOOGL, and possibly open PLTR.

But we must also consider the time: 10:09, still many hours to close. So we can hold.

We also need to think about the overall portfolio. Adding to positions that are already in profit (GOOGL) might be good, but we need to ensure we don't over-concentrate.

Let's determine position sizes. We have $20,678 buying power. We could allocate maybe 50%? But we should be conservative. Since we are day trading, we can use full leverage? The account equity is $9,960, leveraged basis $39,960, so we have about 4x leverage. We can use up to $39,960 total position value. Currently we have $19,281 in positions, so we can add up to $20,679 more (since buying power available). But we should not go all in.

We could add to AMZN and GOOGL. For AMZN, we could add maybe 10 shares (cost ~$2,456). For GOOGL, add 10 shares (cost ~$3,272). That would be ~$5,728, leaving plenty.

But we also need to consider that we might want to exit some positions if the hypothesis is broken. However, all hypotheses are still valid (Polymarket odds still high). So we hold.

We could also consider opening a new position in PLTR. But we have limited time? No, we have hours.

But we need to be careful about overtrading. The system says we should think deeply from first principles. Let's do a self-consistency check:

1. Position sizing matches conviction: If we add, we should size according to conviction. AMZN conviction 4, GOOGL 5. So we could allocate more to GOOGL than AMZN.
2. Not chasing: Prices are not extended; GOOGL is near high but still below R1, AMZN is below its high.
3. Stop loss meaningful: For AMZN, stop below 245.00 is below recent low and also below S1? Actually S1 is 241.55, so too far. Better stop at 245.00 (below 10:05 low 245.57? Actually 10:05 low 245.57, 10:00 low 245.35, so 245.00 is below that). That's a reasonable stop.
4. Contrarian check: Not contrarian.
5. Time decay: Plenty of time.

Thus, we can add.

But we also need to consider that the market might be nearing a top. SPY is at 690.12, near the high of the day 690.15. Could be resistance. But the Polymarket odds are high, suggesting close above open. So there's room.

Given the strong Polymarket odds for AMZN and GOOGL, I think it's prudent to add to these positions.

Also, note that AMZN's odds are 96.7% up, the highest among all. That's a very strong signal. The price is down from its high, presenting a buying opportunity. So I will add to AMZN.

GOOGL odds 95.5% also very high, and price is up, but still near VWAP. Could add.

I'll also consider adding to PLTR as a new position, but we already have four stocks, and adding a fifth might increase complexity. But it's okay.

I'll decide:

- For NVDA: Hold (no change).
- For GOOGL: Add 10 shares (increase long position).
- For AMZN: Add 20 shares (increase long position). Since we already have 20, adding 20 would double to 40 shares. That's a significant increase but still within buying power. Check cost: 20 shares * 245.62 = $4,912.40. That's okay.
- For TSLA: Hold (no change).
- For PLTR: BUY 10 shares (new long).
- For MSFT, AAPL, SPY: HOLD.

But we need to be mindful of fees and slippage. Also, we must ensure we have enough buying power. Current buying power $20,678.69. Adding:
- GOOGL 10 shares at ~327.24 = $3,272.40
- AMZN 20 shares at ~245.62 = $4,912.40
- PLTR 10 shares at ~136.90 = $1,369.00
Total additional = $9,553.80. That leaves $11,124.89 buying power, still plenty.

After adding, total long value will be about $19,281 + $9,554 = $28,835, which is still below the max $39,960.

Now, we need to output trade decisions. The format requires for each ticker an action. For those we hold, we put HOLD with quantity 0. For those we add, we put BUY with quantity.

But note: for existing positions we want to increase, we use BUY to add. For new positions, also BUY.

Thus:

- MSFT: HOLD
- NVDA: HOLD
- AAPL: HOLD
- GOOGL: BUY 10
- AMZN: BUY 20
- TSLA: HOLD
- PLTR: BUY 10
- SPY: HOLD

We should also consider whether to set a stop loss by selling if price drops. But we can't set stops; we only act now. So we hold and monitor.

Now, reasoning for each:

- MSFT: Polymarket 90.5% up but technicals mixed; no edge.
- NVDA: Polymarket 86.5% up, price near VWAP, hold existing.
- AAPL: Polymarket 61.5% up, not enough conviction.
- GOOGL: Polymarket 95.5% up, strong, adding to position.
- AMZN: Polymarket 96.7% up, strong, adding to position.
- TSLA: Polymarket 88.5% up, hold.
- PLTR: Polymarket 91.5% up, new entry.
- SPY: no Polymarket, hold.

We'll output accordingly.

But we must also consider that we already have positions in GOOGL and AMZN; adding more will increase exposure. That's fine.

Now, produce output.</think><<<TRADE_DECISIONS>>>
TICKER: MSFT
ACTION: HOLD
QUANTITY: 0
REASONING: Polymarket 90.5% up but technicals mixed; no edge.

TICKER: NVDA
ACTION: HOLD
QUANTITY: 0
REASONING: Polymarket 86.5% up, price near VWAP; hold existing.

TICKER: AAPL
ACTION: HOLD
QUANTITY: 0
REASONING: Polymarket 61.5% up, insufficient conviction.

TICKER: GOOGL
ACTION: BUY
QUANTITY: 10
REASONING: Polymarket 95.5% up, strong conviction; adding to long.

TICKER: AMZN
ACTION: BUY
QUANTITY: 20
REASONING: Polymarket 96.7% up, highest odds; price pullback offers good entry.

TICKER: TSLA
ACTION: HOLD
QUANTITY: 0
REASONING: Polymarket 88.5% up, hold existing.

TICKER: PLTR
ACTION: BUY
QUANTITY: 10
REASONING: Polymarket 91.5% up, new long entry.

TICKER: SPY
ACTION: HOLD
QUANTITY: 0
REASONING: No Polymarket; hold.


<<<END_DECISIONS>>>