A bot reading a free court feed could have made 55% on Liquidia in one afternoon... or could it?
On September 30, 2026 at 11:15 a.m., a federal court in Delaware posted an order saying Liquidia had lost a patent trial over its lung drug. The court puts new filings on a free public feed. Liquidia's press release didn't come out until about 1:30 p.m. A program that saw the order and shorted the stock one minute later, at $67.97, would have been up 55% when the market closed at $30.26.
Someone is already doing this. In the minute the order appeared, Liquidia traded 318,713 shares. In the minute before, it traded 267.
That looks like a trading strategy: point a bot at court dockets, and trade each ruling before the company announces it. Before building one I wanted to know if Liquidia was a fluke. So I checked 763 other rulings from the last two years, asking the same thing of each: did the stock start moving in the minute the ruling was posted?
Mostly, no. Rulings like Liquidia's are rare to begin with, about four a year by my estimate. Patent office decisions are the easiest to monitor, and they don't move stocks. Court rulings do, but I found only one more where trading started the minute it was posted: the ruling that blocked Kroger from buying Albertsons. Most of the biggest movers come from courts that don't publish a posting time, so for those I still don't know.
Was the docket really first on Liquidia?
I started by lining up the docket against Liquidia's trades, minute by minute. At 11:14 a.m. the stock traded at $71.60. In the next minute, the one the order was filed in, volume ran about 232 times the stock's typical minute that morning, and the price fell as low as $64.65. The full opinion went up at 11:21, Nasdaq paused trading at 11:24, and the stock closed down 57%.
So the docket was first, but a bot would have had about four minutes to get short. After a 10% drop, a rule restricting short sales kicked in at 11:20. The 55% also ignores the cost of borrowing the shares, and I didn't check whether any were available to borrow. There was a second trade that needed no borrowing: United Therapeutics, the company that won, rose 11% from 11:16 to the close.
The free feed may also be too slow. I polled Delaware's feed every two minutes one evening after the close, and it rebuilt only about once an hour. New filings reached it a median of 55 minutes after their docket time. I haven't measured it during market hours. If it's the same then, a bot on the free feed would have seen Liquidia's order after the window closed, and it would have to check each case's docket on PACER, the courts' paid system.
CourtListener, a free archive of federal dockets, logged the order at 11:15:59. With one-minute price bars I can't tell whether the first trades came just before or just after the filing within that minute. CourtListener
That gave me a test to run on every other ruling. If the docket is where the news first comes out, trading volume should jump in the minute the ruling is posted, not before. I counted a jump as any minute with at least 10 times the stock's usual per-minute volume earlier that day, within two minutes of the posting.
How often does a ruling like this happen?
Before testing any feed, I counted how often there's anything to trade. I built lists of seven kinds of rulings from 2021 on: drug-patent trials, government merger challenges, Delaware and other deal suits, antitrust cases, jury verdicts, trade commission import bans and patent office trials. That came to 7,444 company rows. Counting the same stock on the same day once, 1,624 ruling days have a stock price.
A ruling like Liquidia's has three things. Nobody knew the date in advance, it came out while the market was open, and the stock moved more than 10% against its benchmark. My estimate is about 4 of those a year. It's an estimate because most rulings have no posting time, so I assumed the untimed ones come out during market hours as often as the timed ones.
| Kind of ruling | Estimated per year |
|---|---|
| Patent office trials | 2.2 |
| Drug-patent trials | 1.4 |
| Merger challenges | 0.2 |
| Delaware and other deal suits | 0.2 |
| Antitrust cases | 0.1 |
| Jury verdicts (the dates were all known in advance) | 0 |
| Trade commission import bans (no posting times) | unknown |
Counting only rulings with a known time, I found 9. One was Pacira, which fell 48% more than the biotech index on August 9, 2024, the day a drug-patent opinion against it was filed under seal. The other 8 were patent office decisions. Four of those were at Ford, Micron, Ericsson and Roku, companies big enough that a 10% move more likely came from other news.
Company size predicted a big move better than anything else I had. Among companies worth under $1 billion, 17% of rulings moved the stock more than 10%. Over $10 billion, 2% did.
Could a bot trade patent office decisions?
The Patent Trial and Appeal Board is the patent office panel that decides whether a challenged patent survives. I checked it first, because its system timestamps every decision to the second and anyone can pull them. I had minute prices for 625 of its decisions.
Only 19 of them had a volume jump when they were posted. Stocks jump like that at random moments too, so I measured how often each stock had a 10x minute in an ordinary three-minute stretch that day. Added up over every decision, chance alone predicts about 18.5.
A backtest that bought the winners still showed a small profit, +0.78% a trade over 23 trades. That's too few trades to tell from luck, and I added a rule that removes earnings days after looking at the losing trades, which flatters the result. With no volume reaction at posting, I don't think the rulings caused that profit.
Could it trade court rulings?
The rulings that look like Liquidia's come from courts and the International Trade Commission: drug patents, merger challenges, antitrust cases, jury verdicts and import bans. I had two years of minute prices covering 138 of them, counting a ruling once for each listed company it hit. Of those, 24 moved a stock 5% or more, which is roughly one a month for a bot to catch.
To catch one, the bot needs a source that shows the ruling the minute it's public. I looked each ruling up on CourtListener to find that minute, and got one for 18 of the 138.
| What the docket showed | Rulings |
|---|---|
| Timed to the minute | 18 |
| International Trade Commission (its document site blocked my requests) | 35 |
| No docket number in my data | 23 |
| State court, not on CourtListener | 18 |
| Delaware Chancery or Supreme Court, not on CourtListener | 15 |
| No ruling entry on the decision day | 12 |
| Federal Circuit (opinions carry no posting time) | 11 |
| Entry has no filing time | 4 |
| Filed under seal | 2 |
Among the 18, one traded like Liquidia: the ruling that blocked Kroger from buying Albertsons. The opinion hit the Oregon federal docket at 3:21 p.m. Eastern on December 10, 2024. In that minute Kroger traded 71 times its normal volume and Albertsons 225 times.
The others gave a bot nothing to trade. Rulings involving Apple, BlackRock, Novartis and Meta were filed during market hours, and volume stayed under 4 times normal. Alphabet's search remedies ruling came out at 4:05 p.m., after the close. Jury verdicts are read aloud in court before the form reaches the docket, so a bot watching the docket would be late to those.
Is the docket always where the news comes out first?
No. Heron Therapeutics was waiting on a drug-patent trial decision in December 2024. The judge's findings went on the docket under seal at 11:36 a.m. on December 3, so the public could see that something was filed but couldn't read it. Heron had been trading at about 10 times its usual pace since 9:54 that morning. The next day it opened 68% higher.
A docket bot would have missed that one. The market learned the result from something else, and I haven't found what.
Where else could a bot look?
Courts aren't the only place official news can appear before a press release, so I checked four others. These results come from a separate project, and its data isn't in the public repo.
- SEC filings. Across 109 announcements, the filing reached the SEC's site a median of 9.2 minutes after the press release went out. It was more than a minute early in 11.
- The Federal Register. Its public-inspection page showed 44,031 documents a median of 21.2 hours before official publication, but the ones about a single company almost never got news coverage.
- State layoff notices. Of 40 large California notices, 27 had no news story at all.
- Nasdaq's trading-halt feed. On Liquidia, every halt came after the price had already moved.
So could it?
On the rulings I could test, a bot watching patent office postings and federal trial dockets for two years would have found one ruling worth trading besides Liquidia. I didn't find a profitable strategy in what I tested. But I only tested the easy part. To answer the question properly, the bot would need:
- Feeds for the sources I couldn't time. Of the 24 rulings that moved a stock 5% or more, I could time only 4. The other 20 came from the trade commission, the Federal Circuit, state and Delaware courts, and sealed opinions.
- A list of cases worth watching. The bot has to know in advance which docket belongs to which stock, and which outcome is good for it.
- Speed. I measured whether the market reacts in the minute a ruling is posted, not whether I could be the one to react first. On Liquidia the volume arrived inside the first minute.
- Older prices. Liquidia's ruling came after my case lists end, and earlier drug-patent trials such as Pacira's and Exelixis's are older than the two years of minute prices I had.
- Real trading costs. Mine are estimated from minute prices, and nothing here accounts for how much stock you can actually trade in one minute.
I'm not building it. The code and case lists are on GitHub for anyone who wants to.
At a glance
- Question
- Could a program watching public court and agency feeds trade rulings before the company announces them
- Approach
- Posting times for 625 patent office decisions and 138 court and agency rulings, 2024 to 2026, checked against minute-by-minute volume in the posting minute
- Result
- About 4 Liquidia-style rulings a year (estimate). Patent office postings draw a volume spike no more often than chance; of 18 court rulings timed to the minute, only Kroger/Albertsons reacted the way Liquidia did
- Not shown
- The first public source for the 20 large movers the docket could not time