LLM as an analyst

Saturday April 25, 2026

Apr 26, 2026

I have at my disposal a full database of SEC records and press releases and price history for the US public sector. All nicely organized and sanitized.

I run Claude locally for prototyping and kick stuff. Presentation graphics, etc. I asked Claude to look at stock for me. Just for fun while I ate a hamburger. It recombobulated which I think means to waste money.

The company is Envirotech Vehicles, Inc. (NASDAQ:$EVTV). I don’t have any sort of position in it I just sometimes go down interesting warrens. And there are a lot of bunny rabbits around.

Here is what Claude said about the company (after costing me a lot of tokens.) I have high-lighted a lot of fun words. Text is below the screen shot.

The Core Mechanism

EVTV is a shell-pivot operation run through a specific repeating playbook:

1. Dilution engine first — Before any pump, they file S-1 registration statements to activate a Standby Equity Purchase Agreement (SEPA) with YA II PN, Ltd., a Cayman Islands hedge fund that specializes in toxic convertible financing. YA II gets convertible notes it can convert to shares at a discount and immediately sell. This is the exit infrastructure.

2. The pivot narrative — The company was an EV bus maker with minimal revenue. When the EV story stopped moving the stock, they pivoted sequentially: drones → PPE/medical gowns → AZIO AI infrastructure → “AI Compute Platform.” Each pivot generates a wave of press releases.

3. AZIO AI is not a real company — AZIO AI was spun out of Azio Corp (a peripheral accessories brand founded 21 years ago in Taiwan). It had no audited revenue, no government contracts verifiable through public records, and claims of “$100M in government purchase orders” were issued via PR Newswire with no SEC backing. The merger LOI was signed twice at the exact same price ($3.00/share) one year apart, producing identical price spikes.

4. Identical pump, one year later — The most striking detail: Jan 6 and Jan 12, 2025 produced +382M and +761M share volume days respectively. Jan 6 and Jan 12, 2026 produced the exact same volumes on identical claims. The second pump is a replay of the first.

5. The reverse split — After Pump #1 collapsed the stock to $0.17, they did a 1:10 reverse split in August 2025 to avoid Nasdaq delisting. This reset the price to $1.70 and restored the share count capacity for another round of dilution.

6. Financials confirm the fraud — FY2025 revenue was $5.9M (mostly medical supplies, not EV or AI). Net loss was $39M. Stockholders’ equity was $1.8M. The company filed a going concern warning. The stock is currently trading at $1.68 with 761M-share volume days behind it.

I’ve never heard of a shell-pivot scheme. I had to look it up… and I think Claude just made that term up.

But, I want to focus on #4 above. Do you see anything wrong with that claim? Don’t read ahead here. Actually think out this and figure it out. Shameless dots to prevent cheating.

Obviously, the stock volume data is completely wrong for 2025. Which should be the one thing Claude should get write since it just has to read the numbers and because it did correctly chart this information. You can see in the chart there is nine figure volume event in January 2025.

Now I pointed this out to Claude and it sycophantically responded “You’re right!" like it always does and I asked it to reanalyze. And here is what it said in follow-up.

That is pretty interesting, right?

Now I asked Claude the following.

Claude insists on trying to make fucking tables everytime which are hard to cut-and-paste.

That does not paint a very rosy picture of this company. Now, I personally haven’t gone through this company, yet. But, this does beg a LOT of questions.

First, what Claude completely screwed up the most basic thing. The one bit of analysis that required absolutely no thought. What is going to happen with LLMs are used in a legal scenario? This is sort of like Claude reading a medical report that says “Amputate Left Leg” and it diagnosing cancer of the right arm.

Second, how is it that Claude understands fraud (it’s word. Not mine.) so well that it would accuse a company (or individual) of it? What blogs has it been reading?

Third, and most interesting to me, why isn’t the SEC just asking Claude to find every potentially fraudulent company in the US to investigate if the LLM can so easily analyze an entire company in a few minutes?

That is probably going to be a new business for someone. Let LLMs fish out businesses and then build a case to submit to the SEC to collect a reward.

Now, to be fair I did ask Claude more questions.

I am not sure I would ever use the word “Honest” in regards to an LLM but there you go. Is this analysis wrong?

I am not going to answer that question and will leave it to you to decide.

Now low-fi, I pulled the data and made a spreadsheet. I could have had Claude do this but since it appears to be bad with numbers and dates I thought I should give it to Google Sheets instead. Here is every recorded event we have and price movement around that event.

The company does seem to cluster PRs after tax season is due and it did just consolidate… (see gold diamond above) but Yorkshire Advisors is also on my personal naughty list. So I tend to avoid anything they go near.

I wonder if Claude feels the same?

So I asked it another question while getting some ice cream.