PlenitradePlenitradeSeptember 2026. The note to share.
Plenitrade is a paper desk. It buys and sells only inside a simulation. The aim was a rule that beats the S&P 500 after trading costs over a full year the rule was not fit on. A rule here means a function: past prices go in, and the only outputs are buy, sell what we already hold, or stay in cash. We can clear a position. We do not sell short.
Stop searching. Own the S&P 500 through an ordinary index fund. Look once a day. If the close is above the average of the last 200 closes, hold it. If the close is below that average, sell and stay in cash until the close is back above the average. No shorts, no threshold hunt, and none of the 25 quieter stocks.
On $250,000 of paper, from 25 September 2025 through 25 September 2026, that rule finished near $289,000. Holding the index the whole time finished near $296,000. The rule was in cash for 12 sessions during a rising year. That gap is the cost of the off switch. In a falling market the same switch is what keeps you from riding the whole decline. It does not make money from the decline. Cash earns nothing in our test.
Real money stays last. If this is ever connected to a broker, the account cannot withdraw cash, each order needs a human approval, and there is a cap per order and a cap per day. The stake discussed was $5,000. That connection does not exist.
The Live trading button on the blotter opens the broker caps page. That page can store a broker name and a per-order and per-day cap after someone signs in. It does not accept an API key, and it cannot send an order. People are added by an owner at /users. There is no public registration.
The S&P 500 rose about 18% from 7 April 2026 through 25 September 2026, and about 18.5% over the year ending 25 September 2026. On $250,000 of paper, holding it for that year finished near $296,000 after a tiny spread. The 25 quieter stocks were not a shelter. Over the two years of daily prices we had, they fell about as much as the index in its worst months, and they jumped around more on ordinary days.
Each stock-picking idea was written down, then scored on a later period. R² is the share of later price moves the fit lined up with. Around 0.6% means almost all of the move was unexplained. “Rough size” around 2 was the bar for writing a measurement into its own if-statement. Finishing ahead of the index on paper was not enough to freeze a rule when the measurement was weak or the window had already been used.
| What we ran | Later result from $250,000 | Against the S&P |
|---|---|---|
| Price rule. Buy after a weak prior 20 sessions. The fit explained about 0.6%. | About $291,000. Behind holding the same quieter stocks too. | Behind. Not kept. |
| Contract and earnings filings from the company’s own 8-K. | No measurement was large enough to become an if-statement. | Not a rule. |
| Wording of those filings: opinion, agreement, certainty, numbers, and promotional adjectives. Fit explained about 0.4%. | About $490,000 for the combined wording book, and about $383,000 for promotional language alone, over the six months from 7 April 2026. Clearfield, Acme United, and J.Jill did most of the work. | Ahead on paper. Not frozen. That window had already been scored before this test was added. |
| Industry attention. Six words from each annual report filed before 25 September 2025. The count was Wikipedia reading interest, because the news archive refused the connection. Several pages were the wrong topic. Fit explained about 2.8%. | About $353,000 over the year ending 25 September 2026. 48 entries. The rule buys when that reading interest is cold. | Ahead of the index on paper. Not frozen. The measurement is not reliable. |
| Three leaders. Prior month of high-yield credit, long Treasuries, and the dollar, aimed at next month’s S&P. Ten months in the fit, which is a short sample. Only Treasuries cleared the size bar. | About $292,000. Invested in 5 of 11 later months. | Behind. Not kept. |
| Line and threshold. Fit a straight line to the last 63 closes. Buy below the line, sell above it. The distance was whichever of five choices made the most money in the earlier year. For the S&P that choice was 2.5. | About $277,000 for the S&P version. Across the 25 names, 12 of 25 beat holding that same name, and the average name finished behind holding it. | Behind. Not kept. |
| 200-session trend on the S&P. Hold when the close is above the average. Cash when it is below. No other fitted number. | About $289,000. Long on 240 of 252 sessions. Three switches. | Behind holding, by about three percentage points. This is the rule we are keeping anyway, as a way to step aside in a decline, not as a way to beat the index. |
The page at /desk is an older comparison, not the decision above. Cousin and Desk are two versions of the filing rule. Cousin buys the quieter stock and uses a larger order. Desk skips a trade it may not be able to get out of. The dotted lines are the benchmarks: “Same stocks” holds the 25 names, and “S&P 500” holds the index. The Live Trading button on that page is the notice described above.
The wording rule and the industry-attention rule both finished ahead of the S&P on paper. We did not settle on either one.
The wording rule was written after we had already graded other ideas on the same six months. That is taking the quiz, seeing the answer key, and then writing a new answer for that same quiz. Three stocks did most of the winning. More past years would be a better check, and they can be downloaded. We only stored about two years, so we cannot slice several separate years out of what is on disk. Extra past years would still be years we can inspect now. They would not be a year the rule has never seen.
The industry rule was supposed to count news stories. The news archive refused the connection, so the count was Wikipedia reading. The first page for each word was often the wrong page. More years of the same wrong pages would repeat the broken measurement.
A separate page names three trade publications for each of the 25 companies: /publications. That page is an outlet list only. It does not count articles, and it is not a rule.
We then reran the industry-attention equation with those titles. Where a site published a dated list of its articles, we counted an article only when the web address contained one of the frozen annual-report words, and only when that date was already in the past on the signal day. The news archive still refused the connection. Seventeen of the titles had no usable day, so those titles contributed nothing. On the same year, the slope was small, rough size about 0.7, which is below the bar of 2, and it pointed the other way from the Wikipedia version: buy when the trade press was busy, not when it was quiet. The paper book finished near $348,000, against about $296,000 for the S&P. The outlet list was written after that year had already been scored, and the measurement did not clear the bar, so the paper profit is not a rule. The rule in use is still the 200-session trend on the S&P.
The test that would count, for either idea, is this. Write the rule in advance, including the word lists and the pages, and do not change them. Then wait through one year of prices that have not happened yet. Score that year against the S&P after costs. Until that year is over, neither rule is in use. The rule we are using instead is the 200-session trend on the S&P, described above.
plenitrade.com points at its own server, 167.71.163.193, a $6-a-month machine with 1 GB of memory, separate from the classroom site and the WordPress site. The certificate renews on its own through 26 December 2026. The home page is the sign-in page. The equations in more detail are on the research page.