Every trading morning the system reads the same frozen universe of 125 names and publishes what survives as a complete card. Entry, stop, target, size, and the exit fixed before the entry. Then it publishes every name the gates rejected, and why. Almost nobody publishes their rejects. Selection is where most of a decision happens, and it is the part nobody can reconstruct afterwards. It costs you nothing to read.
One email, every trading morning at 9:00 Eastern. It carries the setups the system flagged overnight, each written as a complete card: an entry, a stop, a target, a size expressed as a percentage, and the reasoning that got it through the gates. It also carries every name the gates rejected, with the reason for each one. The four counters directly above this paragraph are today's run. Five minutes, start to finish.
Underneath the email is a pre-registered experiment that runs the same way every morning. The universe is frozen at 125 names, and changing that list takes a signed amendment in a dated log. The constraint is the whole point. A universe that quietly widens when the results sag is re-rolling the dice and reporting the best roll, so freezing the pool is what makes any result comparable to every other result the system has ever produced.
Two further choices follow from the same idea. The rejects get published, because selection is where most of a decision actually happens, and rejection reasoning is the one part of a research process that cannot be reconstructed from a chart afterwards. Every card fixes its exit before its entry, so it states the condition that would prove it wrong before it starts, and a claim that names its own falsification in advance cannot be quietly rewritten once the outcome is known. That is pre-registration, applied one trade at a time.
Every figure the system reports arrives with a confidence interval, because a number without one is an opinion, and nothing here gets called an edge until that interval clears zero. The four conditions under which this shop would declare the whole thing a failure were published before the first subscriber arrived, on the principle that a threshold chosen after the number is known is not a threshold. Two of the four tripped on 21 August 2026. Profit factor came in at 0.571 against a published threshold of 1.2 over 307 gated picks, and win rate at 28.1% over the same 307. Both figures are under recomputation: they came from paper fills using stop levels we never intended to trade, disclosed in our correction of 31 August 2026. They stay tripped and stay published until a correct measurement clears them. The whole account sits at /legal#kill, next to a correction that runs the other way, where an earlier internal figure had overstated this shop's own losses by roughly a factor of two. A shop that only corrects errors in one direction is not correcting errors.
The email is identical for every subscriber, and it is free. It is free because no edge has been demonstrated here yet, and nothing gets sold that has not been demonstrated. What is on offer is a seat at an experiment that publishes its results against itself, including the mornings when the answer is that nothing qualified.
Complete cards, identical for every subscriber, with the exit fixed before the entry. Size is printed as a percentage rather than a dollar figure, because a percentage is the only unit that means the same thing in every edition and to every reader.
A frozen universe of 125 names. Four detectors. Three risk gates. Every card that survives is published exactly as the system produced it, with nothing added by hand and nothing removed to make the numbers look better.
Every Morning Cut card is auto-submitted to a dedicated Alpaca paper trading account the moment the market opens. Real broker. Real fills. Real slippage. Real stop and target triggers.
Not "we backtested this and it worked." Real machinery, real broker, real timestamps. Every position we would have taken is a position we did take. The account statement doesn't lie because it can't.
After the close each day, we reconcile the paper book's realized P&L against our OHLC-based classification. Sometimes they disagree: the tracker says STOPPED but the fill closed green, or the other way round. When that happens it tells us something about our own reporting, so we publish it.
The gates went live 2026-08-13. First honest performance numbers publish here after n ≥ 30 decided picks, which is roughly September 2026. Until then this section says "not enough data yet", because a win rate on a handful of picks is just noise with a percent sign on it.
The spread is the one claim you can audit yourself: if our filters are real, what we pass on must underperform what we publish. It's recomputed nightly from the journal, and the same data feeds this page and every email, so the two can never disagree. If it ever goes to zero you will read it here first.
How many liquid names showed institutional accumulation footprints today, meaning three of our four detectors agreed. One number, published every session, going back to the first one.
We are quants before we are publishers. We like numbers that either hold up or don't. The best part of this job is how fast we get to find out: a hypothesis goes into the replay rig in the morning and comes back with a verdict by the afternoon, so we know something at dinner that we did not know at breakfast. When something doesn't hold up, we write it down and we print it.
The list of things we have buried is longer than the list of things that survived. That ratio is what real research looks like from the inside. The original scanner. The defensive sleeve. The crowd thesis. Congress trades. Post-earnings drift, twice, under two different specifications. Earnings drift, on 8,616 announcements. And most recently our own risk gates, which we re-ran across our own published picks and found negative, with the confidence interval clear of zero. That last one was ours. We published it anyway, because a lab that only publishes its wins is just a brochure.
Then the most satisfying thing that has happened here all year: we audited our own bad news and found we had overstated it by a factor of two. Over half that sample was undecidable on daily bars, and our resolver had been pricing every ambiguous case at the stop. That is the cautious choice when you are guarding against a false positive. It quietly manufactures the answer when the conclusion is already negative. Re-resolved on five-minute bars, the real figure is −0.250% per pick, 95% CI [−0.358%, −0.143%], across 307 picks and 218 distinct publication dates. Still negative and still clear of zero, at half the size we first told you. So we are publishing a correction that makes our own bad news look better. That felt strange, and we did it anyway. We also caught it early, in a paper account, before anyone had a dollar in the market.
And one thing has survived everything we have thrown at it: a 12-1 momentum strategy that beat SPY by 10.5 percentage points a year across 25 years of history, at a Sharpe ratio of 0.99. It has been running in a paper account since 27 May. That is a backtest plus three months of live paper. It is not this newsletter's picks and we are not going to blur the two.
All of it drawn from the same frozen 125 described further up this page, which is what lets any of these results be compared with any other.
We write one letter a week explaining what we found and what we changed, and we keep every one of them. Read them here, including the first, sent to four people. We are going to be very glad we kept that one.
The Finance Wisdom team. Quants first, publishers second.
Four conditions under which this shop would call the whole thing a failure, published before the first subscriber arrived, because a threshold chosen after the number is known is not a threshold. Two of them tripped on 21 August 2026, and the numbers are at /legal#kill.
Two of the four have tripped. A stopping rule that only ever gets mentioned while it stays green is not a stopping rule, so the first real test of one is the day it goes off, and this is the section doing the job it was written for.
The tripped numbers say the product as shipped loses money. They do not say the detectors predict nothing. Run the same replay with trading friction set to zero and the result is −0.074%, with an interval of [−0.182%, +0.033%] that contains zero. What is established is that this geometry, a two to one target and stop that resolves against us about seven times in ten, does not survive the cost of trading it. What is not established is that the underlying signal is worthless, and neither of those gets blurred into the other. Nor is the answer to tune our way out: adjusting the gates, changing the volatility windows and improving the assumed fills have all been measured, all three are dead ends, and the one adjustment already made measurably made things worse.
Nobody has ever paid us anything. No payment processor has ever been connected to this site, so no refunds are owed, and the refund promise attached to these criteria stands for whoever comes later. All four criteria get reported here every quarter from now on, whether they trip or not, because a shop that speaks up only when something breaks has turned its own silence into a claim. The full account, including the correction where an earlier internal figure had overstated these same losses, is at /legal#kill.
Finance Wisdom is building toward its first investment fund. That is the whole ambition, said plainly. We want to be trusted with other people's money one day, and we want to have earned that instead of marketing our way to it.
We are early. The way we think you earn that trust is to do the work where people can watch: publish the research and publish the results, including the results that went against us. Let the record pile up in public over years, instead of turning up one day with a finished pitch deck.
Signal is that record being written, one morning at a time. Read it and you are watching a fund get built from the first honest measurement onward, with the false starts and the corrections in the order they actually happened. Almost nobody outside a firm ever gets to see this part of it. You may as well see it now, instead of hearing a tidied-up version in three years.
Nothing on this page is an offer to invest in anything. A subscription buys the newsletter and nothing else. It carries no ownership, no stake, no interest in any fund or entity, and no right to any return.
We have a replay rig that turns a hypothesis into a verdict in an afternoon. It uses a point-in-time universe, announcement dates pulled from EDGAR filings, split-adjusted bars, and a sanity gate that has already caught us being wrong. It has killed four of our own ideas so far, and honestly we hope it kills a fifth. Now we are opening it to you. Bring the idea you have been arguing about with a friend and we will run it properly and publish whatever comes back.
The morning email is free and stays free. What you'll be able to pay for, later, is a seat in the lab, never a better pick. Everyone gets the identical email, always. We are not going to sell you picks we have measured and found wanting. The workbench is the part worth paying for.
None of those are performance milestones, and none of them are promises about returns. We are not waiting for good numbers. We are waiting to be allowed to charge you properly.
And so it is on the record before a single dollar moves: we have no validated performance record, and we are not selling one. Our own year-long replay of this system, with every quality gate applied, returns negative expected value across 307 picks, and that range does not include zero. The smaller live sample on our record page is positive, but its confidence interval does include zero, which means it is indistinguishable from luck. The larger sample is the one we act on. You would be subscribing to a daily process and a public audit trail. You would not be subscribing to a demonstrated edge. If that isn't what you want, the free edition is genuinely the better deal and we'd rather you took it.
Membership pays for the lab: data, compute, the backtest harness, and the audits that keep telling us things we did not want to hear. It does not buy a position in anything, it is not pooled with anyone's capital, and it does not go into the market. When we can afford better data or a harder test, that is what it goes to.
Your first Morning Cut lands at 9:00 a.m. ET on the next trading day, with the plans that made it, the names that didn't, and the reason for each. If it isn't worth five minutes of your morning, one click and you never hear from us again. No hard feelings at all. Eight people read this today. We want the next eight to be people who check our arithmetic, and we are in no hurry to get to eight thousand who don't. Come and watch it get built from the inside.