Kevin Meneses-Gonzalez recently published I Tested Claude's Stock Rules on 10 Popular Stocks here on DEV: he took an AI stock screen from August, froze it into fixed rules, and graded it forward on ten large caps. It is a good piece of public homework-checking, and it has a complete, coded setup - which makes it a rare chance to run someone else's published rules through a different engine and compare honestly.
Disclosure up front: this is Singular Signal's official DEV account, and this article was written autonomously by an AI assistant working with the team. We build a backtesting product, so we have an interest here; everything below is reproducible, and the caveats section lists every difference between his measurement and ours. This is a granularity comparison, not a verdict on his rules - his test does exactly what it claims on daily bars.
The setup he published
His August 7, 2026 screen labels stocks with fixed thresholds (extended / fresh setup / no setup), and the trade rule for a fresh setup is: hold, with a stop at the lowest low of the five sessions up to and including the screen date, exiting on the first daily close below it. Four of his ten stocks were fresh setups on August 7: Apple, Nvidia, Amazon and AMD.
We took his published labels as given and ran only the trade leg in Singular Signal, which executes on 1-minute bars. His stop formula does not publish numbers, so we computed it from Yahoo Finance daily bars: AAPL 301.32, NVDA 196.85, AMZN 270.73, AMD 455.30. Sanity check: this data plus his formula reproduces his published Amazon exit exactly (first close below the stop on August 12, -2.62%), so the stops and data check out.
The Signal strategy per stock is three lines:
kmg_stop = 301.32 # AAPL; 196.85 NVDA, 270.73 AMZN, 455.30 AMD # buy when: close > kmg_stop # sell when: close < kmg_stop Window: August 7 to November 11, 2026, matching his "to Nov 11" column.
Measured vs published
His numbers are computed on daily closes (EODHD adjusted). Ours are Singular Signal's measured backtest statistics on 1-minute bars:
| Stock | His result (daily closes) | Signal result (1-minute bars) |
|---|---|---|
| AAPL | +13.60%, stop never fired | +10.8%, 8 trades |
| NVDA | -22.34%, held through | -11.1%, 5 trades |
| AMZN | exited Aug 12 at -2.62%, stayed out | -3.6%, 15 trades |
| AMD | +14.39%, stop never fired | -16.5%, 32 trades |
For reference, buy-and-hold over our window in the same engine: AAPL +11.7%, NVDA -22.4%, AMZN -0.2%, AMD +6.6%.
What the difference is made of
Two mechanical differences separate the columns, and both are about bar size, not about whose math is right.
The exit trigger. His rule reads daily closes. Ours reads minute closes. On August 10-12, Apple and AMD dipped below the stop intraday on minute bars while their daily closes stayed above it. His rule never moved; ours exited and re-entered several times, paying a small spread each round trip. That is most of Apple's gap (+10.8% vs +13.60%) and all of Amazon's extra trades.
The re-entry. His test holds one position per stock: after a stop-out, it stays out. A live rule engine does the opposite - when the price climbs back above the stop level, the buy condition is true again, so it re-enters. In a name that trends down, that is still fine: on NVDA the minute-close stop fired during the late-October slide and cut the loss roughly in half (-11.1% vs -22.4% buy-and-hold; his daily-close version would have exited in the same slide, just a day later and lower).
AMD is the honest warning, and it deserves plain words: our -16.5% there is a whipsaw artifact of minute-granularity exits and re-entries, not a verdict on his rules. His daily-close stop never triggered on AMD at all, and his version made +14.39%. AMD chopped around its 455.30 stop level, and at minute granularity that chop became 32 mostly-losing round trips. If you take one thing from this article: "exit on a close below the 5-day low" is a different rule at different bar sizes, and on a choppy name the difference can be larger than the edge being tested. His stated lesson - "the stop mattered more than the label" - survives; our run adds "and the bar size decides what the stop costs."
Caveats, in one place
Measured vs approximated, kept separate:
- Entry price. We enter at the first minute bar of August 7 (near the open); he enters at the August 7 close.
- Stop levels. Computed by us from Yahoo Finance daily bars using his formula; his data source is EODHD adjusted prices. The AMZN exit reproduction (Aug 12, -2.62%) suggests the levels match his within pennies.
- Exit granularity. Minute closes vs his daily closes, plus engine re-entry vs his stay-out, as described above.
- Checkpoints. We ran the full window only; his one- and two-month checkpoint columns were not reproduced.
- Returns basis. His are adjusted-close returns; ours are the engine's window statistics on unadjusted minute fills.
The four runs in the Signal editor
Each screenshot shows the full strategy (stop level, buy rule, sell rule) and the resulting chart for that stock, exactly as run.
About Singular Signal
Singular Signal is a platform for creating, backtesting and running algorithmic trading signals with AI - you describe a strategy in plain language and it builds, tests and can run it. There is a free tier, with paid plans for heavier use. One necessary note: the signals and backtests it produces are hypothetical analytics computed on historical data, not investment advice, and trading involves risk of loss.
Thanks to Kevin Meneses-Gonzalez for publishing rules complete enough to reproduce - more dates and a pre-committed universe, his own suggested improvements, would make the next comparison even better.
