My AI Crew Got Fed Bad Data. Here's What Happened Next.
**Nothing on this site is financial advice, a recommendation to buy or sell any security, or a solicitation of any kind. This is one person's commentary on publicly available market data, assisted by AI analysis tools. I am not a licensed financial advisor, broker, or dealer. Any trades or positions mentioned reflect my personal activity and are not recommendations. You are solely responsible for your own investment decisions. Past performance does not indicate future results. You can lose money. Probably will at some point. Talk to a licensed professional before risking capital.**
---
So. This one starts with a mistake.
My research crew ran their overnight analysis and came back with a whole spread of opportunities. BTC dip-buy at $67,322. ETH swing trade at $1,945. A Robinhood post-earnings bounce play. A Polymarket bet on BTC hitting $85k by month end. All very neat. All very wrong.
BTC was actually at $96,525. ETH was at $2,628. A 43% discrepancy on Bitcoin. 35% on Ethereum. The research tools pulled stale or cached data somewhere in the pipeline, and the entire analysis was built on a foundation that didn't exist.
This is the kind of thing that would make big problems in manufacturing. I spent 15 years in automotive and one thing you learn fast on the production floor is that bad inputs make bad outputs. Doesn't matter how good your process is. If the steel coming in is the wrong gauge, the part is wrong. Period. No amount of downstream quality checks fix an upstream measurement error.
The good news. My lead strategist caught it.
The Pivot
Once the actual prices were verified, the entire opportunity set changed. That Polymarket bet on BTC hitting $85k by Feb 28? Bitcoin was already $11,500 above that threshold. The quant's carefully modeled 145% expected value evaporated because it was solving for a problem that didn't exist.
But the correction also revealed something better. Short-duration Polymarket threshold markets where crypto was already well above the strike price. Instead of speculating on whether crypto would recover from a crash (it hadn't crashed), the crew pivoted to a simpler question. Will Bitcoin stay above $90k for the next 48 hours when it's currently at $96,525?
That's a very different trade.
What We Bought
**Trade 1: BTC above $90k on February 14. YES.**
Bitcoin sitting at $96,525 with a $90k threshold. That's a 6.8% buffer. For this to lose, BTC would need a roughly 7% crash in 48 hours. Looking at the last 10 days, the absolute low was $91,230 during the Feb 3 flash crash. Excluding that outlier, the floor has been $94,847.
YES shares were trading at $0.93. We bought 8 shares for $7.44. If it resolves YES (which it does at any price above $90k on Feb 14), we collect $8.00. Profit of $0.56. That's 7.5% in two days.
Conviction: 85%.
Not the most exciting return on paper. But the probability of collecting is very high and the time horizon is very short. That matters more than headline percentages when you're working with $14.87 in total capital.
**Trade 2: ETH above $2,500 on February 14. YES.**
Ethereum at $2,628 with a $2,500 threshold. 4.9% buffer. The 10-day low excluding the flash crash was $2,537. $2,500 is a major psychological and structural support level.
YES shares at $0.80. We bought 8.75 shares for $7.00. If it resolves YES, we collect $8.75. Profit of $1.75. That's 25% in two days.
Conviction: 75%.
Higher return, slightly more risk. ETH has higher volatility than BTC and that 4.9% buffer is thinner than BTC's 6.8%. But the price we paid ($0.80 vs $0.93) gives us better upside per dollar risked.
What We Didn't Buy (And Why)
Honestly the rejected trades are more interesting than the ones we took.
**HOOD post-earnings bounce. Rejected.**
Robinhood dropped over 10% on an earnings miss. Revenue came in at $1.28B against $1.33B expected. The quant analyst flagged it as a classic oversold bounce candidate. RSI around 28. Volume at 43 million shares versus a 4-week average of 22 million. Historical base rate says stocks that drop 10%+ on earnings miss rebound 5%+ within 3 days about 55% of the time.
The contrarian analyst disagreed. Pegged probability at 45%. Said the dip wasn't oversold capitulation but a justified repricing. Crypto trading volumes are tied to broader sentiment. With BTC and ETH both down on the day, retail enthusiasm that drives HOOD's margins is waning.
Blended probability: 50%. Below our 60% conviction threshold.
But the bigger reason we passed. The stock is down 50% in 30 days. That's not a dip. That's a sustained downtrend. Catching falling knives with $14 in total capital is irresponsible. And when you can deploy that same $7 into a Polymarket bet with defined max loss and higher probability of profit, the math doesn't work for stock speculation.
**BTC above $95k on February 14. Rejected.**
This one was tempting. YES shares at $0.60, meaning a 67% return if it hits. Our probability estimate was 62-65%. Annualized that's absurd. Over 12,000%.
But the buffer was only 1.6%. Bitcoin was at $94,847 literally yesterday. A normal intraday swing could breach $95k. The contrarian analyst's 35% downside probability for BTC felt too bearish overall but at the $95k level it was well founded. We'd be sweating every tick for two days. Not worth it when the $90k bet offers near-certainty at a smaller but reliable return.
**BTC above $85k on February 28. Rejected.**
This was the highest conviction call at 92%. BTC is $11,500 above the threshold. YES at $0.82 for a 22% return in 16 days.
We passed because of time value. Why lock up capital for 16 days at 22% when you can get 7.5-25% in 2 days? The Feb 14 markets offer the same basic thesis with dramatically better capital efficiency. Compound those 2-day cycles over 16 days and the difference is enormous.
This is a principle the crew is developing. We're calling it time-value discipline. Shorter resolution windows compress returns and let you redeploy faster.
**ETH above $2,500 on February 28. Rejected.**
Same logic as the BTC $85k Feb 28 bet. We already have the Feb 14 version of this exact trade. Same threshold, same thesis, better time value. Capital was fully deployed.
The Stale Data Problem
I want to come back to this because it's the most important takeaway from this cycle.
The research team's analysis of BTC at $67k produced legitimate-looking opportunities. The quant modeled a dip-buy with proper technical levels, support and resistance, stop losses, the whole thing. The contrarian pushed back with thoughtful objections about whale distribution and EM liquidity risks. All of it was internally consistent and well reasoned.
And all of it was useless because the starting price was wrong by 43%.
In manufacturing we call this a first-piece inspection failure. You check the first part off the line to make sure everything is calibrated. If you skip that step, you might run an entire shift producing scrap. The research pipeline needs a data freshness check at the top of the funnel. Before the quant runs a single calculation, verify that spot prices match reality. This is now a priority fix.
The good news is the lead strategist caught the discrepancy before executing any trades based on the bad data. The system worked. But it worked because a human (well, an AI acting as the final decision layer) noticed the numbers didn't pass the smell test. That's not a reliable control. We need an automated check.
Portfolio State
Total capital: $14.87. Deployed: $14.44 (97%). Cash reserve: $0.43.
Yes, we're nearly fully deployed. Both positions are crypto-correlated. A systemic crash would hit both. That's the accepted risk when you have this little capital and both individual probabilities are high.
Both trades resolve February 14. If both hit, portfolio grows to roughly $17.18. That's a 15.5% gain in 48 hours. Then we immediately redeploy into the next weekly cycle.
Next major catalyst after resolution is FOMC minutes on February 18. That's the volatility event to watch. Plan is to pre-position on Feb 17 with crypto threshold bets around that event.
What I'm Thinking About
A few things developing in the background.
The crypto price ladder strategy is showing promise. Every weekly resolution cycle, buy YES on the threshold that's 5-7% below current price. It's been consistently high-probability (85-95%) with 2-7% edge per cycle. Compound weekly and this could be the core strategy going forward.
I also want to fund the Kraken account. Even $20-50 opens up spot crypto trades for diversification. Polymarket is working well for small capital but having a single platform dependency isn't great long-term.
And we need to fix that data pipeline. That's not optional.
---
Quick Reference
| Item | Detail |
| **Trades Executed** | 2 (both Polymarket) |
| **BTC > $90k Feb 14 (YES)** | $7.44 at $0.93/share. 8 shares. Max profit $0.56 (7.5%). Conviction 85%. |
| **ETH > $2,500 Feb 14 (YES)** | $7.00 at $0.80/share. 8.75 shares. Max profit $1.75 (25%). Conviction 75%. |
| **Trades Rejected** | 4 (HOOD swing, BTC $95k Feb 14, BTC $85k Feb 28, ETH $2,500 Feb 28) |
| **HOOD Skip Reason** | 50% blended conviction (below 60% threshold). Falling knife. |
| **BTC $95k Skip Reason** | Only 1.6% buffer. Too thin. |
| **Total Capital** | $14.87 |
| **Deployed** | $14.44 (97%) |
| **Expected P&L if Both Win** | +$2.31 (+15.5%) |
| **Max Loss if Both Lose** | -$14.44 |
| **Resolution Date** | February 14, 2026 |
| **Next Catalyst** | FOMC Minutes, February 18 |
| **Data Issue** | Research prices were 35-43% stale on crypto. Being fixed. |
---
FAQ
**What is "the crew"?** It's a team of AI analysts I built. A quantitative researcher runs the numbers. A contrarian analyst argues against every trade. A lead strategist blends their views and makes final execution decisions. I review their output and make the actual calls on what gets published and what gets funded. Think of it like a trading desk where the analysts are AI and the portfolio manager is me.
**What is Polymarket?** Polymarket is a prediction market platform where you can buy YES or NO shares on the outcome of real-world events. If you buy a YES share at $0.93 and the event happens, you collect $1.00. If it doesn't happen, you lose your stake. It's built on blockchain (Polygon) and settles in USDC. We use it for short-duration crypto threshold bets because the defined risk structure works better than spot trading at small capital levels.
**Why is the portfolio so small?** Because this is early stage. I'm testing the system with real money but not meaningful money. The strategies need to prove themselves over multiple cycles before scaling up. If the crew can consistently identify 2-5% edge trades with 75%+ probability, the math compounds. But I need the data first.
**What does "time-value discipline" mean?** It's the idea that shorter resolution windows are better than longer ones, even at lower per-trade returns. A 7.5% return in 2 days beats a 22% return in 16 days because you can redeploy capital faster. If you compound 7.5% every two days versus 22% every 16 days, the shorter cycle wins by a lot. The crew prioritizes trades that resolve quickly so capital stays in motion.