I went back through every short-term Hold I had ever issued and asked a blunt question: was holding back the right call? Mostly it was. On one specific kind of name, it wasn't — so I adjusted the rule. Here is exactly what I found and what changed.
Every report gives three separate calls on a stock — short-term (one to three months), medium-term, and long-term — because a name can be a poor trade this month and an excellent hold for three years, or the reverse. By a wide margin the most common short-term call is Hold: patience is usually the honest answer over a few weeks. The win-and-loss record of the short-term Buys is on the Results page. What I had never done is check the Holds — and a Hold that should have been a Buy costs just as much as a bad Buy. It just never shows up anywhere, because nothing was bought.
I took every short-term Hold that was old enough to judge — 32 of them, each at least a month past the call — and scored each one two ways. First: did holding miss a move the stock actually made? Second, and the one that decides anything: would buying instead have made money? For that I used the same $10,000-a-name illustration the Results page is built on, wins and losses both counted in full — with one simplification: because these were Holds and not Buys, there was no entry target or scale-in ladder to apply, so I just bought at the call and marked each name to today's price.
Then I sorted the Holds by why I had held, because the fix depends entirely on the reason.
The result split cleanly in two, and the losing side matters as much as the winning one.
In the plain illustration, buying that second group the day I said Hold would have come out roughly $22,000 ahead of sitting on the sidelines. That is hindsight, and an illustration — not a promise, and not a clean win: two of the thirteen still fell, and the whole test covers barely five weeks of a rising market (more on that limit below).
The short-term signal deliberately leans about 55% on timing — over a few weeks, the tape usually matters more than the fundamentals. The side effect: a name could be excellent and fairly priced, a name I already liked for the next year, and a merely average timing reading was enough to sit it out completely. Not a bad tape — just an unremarkable one. That is the specific blind spot the audit exposed.
I added one narrow, bounded rule. When a name is high quality, not expensive, and I already rate it Buy on both the medium and long horizon, and the only thing holding the short-term call back is a middling — not weak — timing score, I no longer sit it out entirely. It becomes a half-size starter: a partial position taken now, with the rest added on the pullback the timing was waiting for, instead of a flat Hold.
I left the protections that were working untouched. A genuinely weak tape still means a full Hold. An expensive name still means a full Hold — those names fell in the very same weeks the quality names rose, so loosening that rule would have been the wrong lesson. The change is surgical: it touches one configuration and nothing else.
This is drawn from about five weeks of a single, mostly-rising market, and not one of these Holds has yet seen a full three-month window. In a rising tape a lot of things go up, so I am treating this as a promising signal, not a settled fact. That is why the change is scoped to half size, why I did not touch the protective rules, and why I have set a reminder to re-run this same audit after the first real market drop — the exact conditions the original caution was built for. If it does not hold up then, I will narrow it or take it back out, and I will say so here.
Donatien is an experiment I run in the open, and the method is still being tested. The point of a page like this is that you get to see the machinery change — not just the outputs, but the reasoning and the second-guessing behind them. The Results page shows whether the calls are working; this page shows how the calls themselves evolve when the evidence says they should.
See all Donatien Updates →