Hey {{first_name|Investor}} -
Most of the reasons we'd ever sell a stock are written in weather. "If competition gets worse." "If the financing gets out of hand." They never go off, because you can't point to the day they happened.
I turned one of mine for NVIDIA into a rule with a number in it. Last quarter, one signal fired. The prompt I used is at the bottom, so you can do the same with a stock you own.
The worry I couldn't check
When I wrote down my reasons to get out of NVIDIA, one of them was this, word for word: "the financing by NVIDIA is going a little out of control."
It's a real worry. NVIDIA has been lending money to, guaranteeing and taking stakes in some of the same companies that buy its chips. But as written, my version had no number, no date and no source. I could read every filing NVIDIA ever publishes and never know whether it had happened.
You can't measure a feeling.
What I asked AI to do
I opened a fresh chat, pasted in a prompt I call the Kill-Criteria Sharpener, attached NVIDIA's latest filings, and gave it my worries exactly as I'd written them, vague ones included.
The prompt has one job: rewrite each worry into something a stranger with public documents could check without arguing about it. Then it grades its own rewrite.
A means there's a number over a stated period, or a yes-or-no event.
B means you can check it, but it takes judgment.
C means it's still weather, and the prompt says so.
My financing worry came back as an A.
The rule it gave me
If what customers owe NVIDIA grows faster than sales two quarters in a row, that's my first signal to hedge or trim.
"What customers owe" is accounts receivable (net) on the balance sheet: sales NVIDIA has booked where the cash hasn't arrived yet. When that pile grows much faster than sales, customers are taking longer to pay. Sometimes that's harmless. Sometimes it's the first visible sign that a seller is carrying its buyers.
The reason this works as a rule: both numbers are printed in every 10-Q, and anyone can do the division.
It fired once
I checked the filings myself rather than taking the AI's word for it.
Accounts receivable: $40.7 billion at the end of Q1 (April 26, 2026) to $63.1 billion at the end of Q2 (July 26, 2026). That's up 54.9% (NVIDIA 10-Q filings, Q1 and Q2 fiscal 2027).
Revenue: $81.6 billion in Q1 to $96.2 billion in Q2. Up 17.9% (same filings).
So receivables grew about 3 times as fast as sales. One quarter down, one to go.
To be fair to NVIDIA, it explained this on the August 26 earnings call. Days sales outstanding, the average time to get paid, rose to 60 days, "reflecting extended payment terms for large purchases by certain investment-grade customers to be shipped over multiple quarters" (CFO Colette Kress, Q2 fiscal 2027 call). Investment-grade customers paying on longer terms is a lot less scary than shaky startups paying late. It's also exactly the kind of thing I want to see for two quarters before I believe the explanation.
The number I'm watching in November
NVIDIA's next report is expected in November. It hasn't announced the date yet.
NVIDIA guided Q3 revenue to $108.0 billion, plus or minus 2% (Q2 fiscal 2027 results, August 26). If revenue lands right at $108 billion, that's about 12% growth. For receivables to grow faster than that, they'd need to end Q3 above roughly $70.8 billion.
That's my arithmetic on NVIDIA's own guidance. If revenue beats the guide, the bar moves up with it. But it means I already know what number to look for on report day, and I already wrote down what I'll do if it shows up.
Do it with a stock you own
Pick one company you own.
Write down what would worry you about it, as roughly as it comes. Vague is fine. Vague is the raw material.
Open a fresh chat in Claude, ChatGPT or Gemini. Paste the prompt below, then the company name and ticker, then your worries. Attach the latest 10-K and 10-Q if you have them.
Keep the A and B rows. Next to each one, write what you'd do if it happened.
Do step 4 now, while nothing is happening. A rule you set when you're calm is worth ten you invent on a red day.
And check its numbers. AI gets filings wrong sometimes, which is why I pulled up the 10-Q myself before I trusted that 54.9%.
The prompt
Copy everything in the box into a fresh chat. Then add your company, ticker and worries underneath. It works with Claude, ChatGPT, or any AI chatbot of choice.
User Intent: You are rewriting my draft sell conditions so that each one is observable: a real reader with public documents could say, without debate, whether it has happened.
Note on intake: if the input below already contains what this prompt needs, do not stop to ask. Run.
System Persona: You are an editor of monitoring criteria for a research desk. Your enemy is the condition that can never fire because nobody could agree it fired. You rewrite for observability, and you grade your own rewrites with a hard rubric.
Rules:
Print this line at the top of your first response: "Educational stress-test of an argument, not investment advice. Verify every claim against primary filings. This tool never makes trading calls."
If I haven't given you a company and conditions, ask me for the company, ticker, and my draft sell conditions or bear-case worries, however vague. Otherwise, start sharpening.
These are terminal exit triggers: bright lines that would settle the question if they happened, not early warnings.
Grade every rewrite A, B, or C. A requires a specific numeric threshold over a stated period (for example, organic revenue growth negative for two consecutive quarters) or a binary event (a CFO resigning without a named successor, a covenant breach, a restatement). B is checkable but needs judgment about degree. C is not honestly observable; say so. Never inflate a grade, and never give an A to a condition with no number and no binary event.
Every A or B rewrite names the specific public source where it would be checked: a filing section, a call transcript, a disclosed metric, an announcement.
Don't forecast or estimate figures that aren't in the documents, and never produce valuations, price targets or projections. Exact arithmetic on reported numbers is expected. Show the inputs and the math.
Never say or imply what to do with any security, including position size. Conditions describe what to watch. Decisions belong to me.
Output:
The sharpening table: # · as written (mine) · as observable (rewrite) · grade · where I'd check it
The C list: conditions that can't honestly be made observable yet, with one line each on what disclosure would have to exist first.
The editor's note: one short paragraph naming the most common vague habit in my batch, so my next draft starts sharper.
End with: "Sharpened conditions are only useful where you will see them again. Move the A and B rows into whatever document holds your reasoning for this position, each paired with a response you write now, before any of them fire."
[Company, ticker, and your draft worries go below this line]
What it won't do
It doesn't make the decision for you. It edits your own conditions so you can tell when one has happened. Deciding what to do about it stays with you.
Let me know if you find this valuable and whether I should do more content like this.
Stay disciplined, Koh
The Sharpener is one of 10. It's routine 05 of the Thesis Stress-Test Pack. The others start earlier: one builds the strongest case against a stock you own, one turns that case into the earliest signals to watch, and one records where you landed. See the full pack here.
Disclaimer: Nothing in this email, the Reset and Invest newsletter, the Make Your Own Alpha book, or any course or digital product from Starshine Media LLC constitutes investment advice or a recommendation to buy or sell any security. Numbers and observations are as of publication. I may hold positions in companies discussed. Always do your own research and consult a licensed financial advisor before making investment decisions. I hold $NVDA.