The Day My AI Scanner Got It Wrong — and What I Learned From It

Everyone who publishes trading content shows you the wins.

Here is one I got wrong.

The Setup

Two weeks ago my options flow scanner flagged RXRX with a put/call ratio of 0.38. That's in extreme bullish territory — heavy call buying relative to puts. The scanner classified it [EXTREME BULLISH] and it hit my Telegram before market open.

I had been watching RXRX anyway. It's a biotech name I track because it's volatile, options-active, and the P/C signal tends to be cleaner than large caps with heavy retail flow.

The signal looked right. The momentum was there. I sized into a small position.

What Actually Happened

The stock dropped 11% over the next four days.

Not a catastrophic loss — I had a stop set and honored it. But wrong is wrong.

When I went back to audit the signal, I found two things I had missed:

1. The lottery call problem

I run a lottery call filter on most tickers — it strips out OTM contracts below 0.15 delta that are almost certainly retail speculation. I had not run it on RXRX that morning. When I ran it after the fact, the adjusted P/C came out at 2.14, not 0.38.

The raw signal said extreme bullish. The filtered signal said bearish. I had traded on noise.

2. Volume context

The call volume that drove the 0.38 reading was concentrated in two strikes: the $12 and $15 calls. RXRX was trading around $8.40. Those contracts needed a 43% and 79% move respectively to expire in the money. They were lottery tickets, not institutional accumulation.

I knew about the lottery filter. I just didn't run it that morning because I was in a hurry.

The audit output

RXRX Post-Trade Audit — 2026-05-13

Signal at entry:
  Raw P/C:          0.38  [EXTREME BULLISH]
  Lottery-adj P/C:  2.14  [BEARISH]
  Delta < 0.15 calls: 91.4% of total call volume

Price action:
  Entry:   $8.47
  Stop:    $7.60 (triggered day 4)
  Exit:    $7.63
  Loss:    -9.9% on position

Was the signal wrong? No.
Did I use the signal correctly? No.

This distinction matters. The tool worked. The workflow had a gap.

What I changed

Three things:

1. Lottery filter is now mandatory, not optional

I updated the skill so the adjusted P/C always runs alongside the raw P/C. If they diverge by more than 0.5, the output flags it as [SIGNAL UNRELIABLE — HIGH LOTTERY DISTORTION] and I don't act on it without manual review.

2. Added a volume concentration check

If more than 60% of call volume is concentrated in two or fewer strikes, the scanner now notes it. Concentrated volume in far-OTM strikes is a retail pile-on pattern, not institutional.

3. Slowed down the entry decision on biotech

Biotech names have more retail options activity than most sectors. The lottery distortion problem is worse there. I now require the adjusted P/C to confirm before acting on any biotech signal.

Why I'm publishing this

Most trading content is survivorship bias in article form. The writer shows you the XLI signal that worked, not the RXRX signal they misread.

I build tools. Tools have bugs and workflow gaps. Publishing the failure is more useful than publishing only the wins — it shows what the system actually catches and what it misses, and it forced me to fix the gap I had been ignoring.

The scanner is better now than it was two weeks ago. That's the honest version of this story.

The updated Options Flow Analyzer with mandatory lottery filtering: github.com/tellmefrankie/ai-investment-skills

Full suite with the audit trail and Telegram alerts: jaehyunpark.gumroad.com/l/tcyahy

Not financial advice. This is a post-trade audit of personal tooling decisions, not investment recommendations.