The Ruler We Measure With

How Every Call Gets Graded

An accuracy number means nothing without the ruler behind it. This page is the ruler: exactly how a sent alert becomes a win, a miss, or an exclusion on the public track record. Every threshold below is read from the same code that does the scoring, so this page cannot say one thing while the scorer does another.

The short version: we grade the direction we called, from the price when we sent it, over a window matched to the kind of event, and we publish every result.

What gets graded

Every alert that made a directional call, bullish or bearish, is graded. No sampling, no curation: if we sent it with a direction, it is on the record. Informational alerts that made no direction call have nothing to grade and are not counted either way.

Once graded, a call is locked. A miss that later turns around stays a miss. The one thing we never do is reopen old grades, because regrading history is how services in this category manufacture their numbers.

Two retention rules apply. Full alert detail is kept for two years and is browsable month by month in the member dashboard. The graded record itself is permanent: every scored call stays in the track record and in every published number, forever. Nothing ages out of the hit rate.

The entry price

Scoring starts from the live market price at the moment the alert was sent, not that day's close. A close-to-close measurement can hide the move that happened between our alert and the bell; measuring from the send price reflects what a subscriber acting on the alert could actually have captured. (Early records from before we stored the send price use the first close on or after the alert instead.)

The send price itself is sanity-checked before it is trusted. If the recorded quote sits implausibly far from every close the stock actually printed around the alert (more than 3 times the stock's own grading band, described below, from both the entry close and the prior close), we treat it as bad quote data, not a real move, and grade from the entry close instead. A corrupt quote must never manufacture a same-day win or loss.

The window

Different catalysts price in at different speeds, so each event class gets its own grading window, in trading days:

A tariff announcement is priced in within a session or two; insider buying is a weeks-long signal in the academic literature and gets a trading month. Grading both on the same next-day close would call working signals random.

Win, miss, or neither

Inside the window we use a first-touch read, the "triple barrier" method from quant research: a target, a stop, and a time limit. Each stock gets a price band scaled to its own daily volatility, floored at 2% and capped at 25%, so a decisive move means the same thing on a mega-cap and a micro-cap. Every alert carries the same trade plan we grade against: a profit target one band in the called direction and a stop one band against it - deliberately the same distance, because a target closer than the stop would hand a coin-flip call an inflated baseline. Insider-buy alerts are measured from the next session's close rather than the after-hours send price, because Form 4 filings pop in extended hours and fade the next day - the plan tells you to act at the close, so the grade measures that. Then, walking the daily closes:

• →If the profit target is reached before the stop, the call is a win.
• →If the stop is reached first, it is a miss.
• →If neither is touched, the call is graded on the window's final close: a win only if the stock still finished in the called direction by at least 0.5%.

Everything is measured on daily closes. No intraday peaks, no cherry-picked highs: the reported move is an exit a person could actually have taken. Until a band is touched or the window ends, the call shows as developing rather than being graded early.

When a call enters the hit rate

A call can be graded early, the day its target or stop is reached, and it shows up in the ledger right away. It only enters the published hit rate once its whole window has closed. Calls that go nowhere are graded last, on the final day of the window, and most of them are misses. Counting grades as they arrive would count a week's quick wins before that week's slow misses, so a young record would always read higher than it ends up. Waiting for the window means every week of calls is complete when it is counted. We added this rule on September 14, 2026, after finding that the rate we showed at the end of August read about nine points higher than those same calls finished.

Versus the S&P 500

The hit rate asks one question: was the direction right? It counts a 1% win the same as a 20% win, and a bullish call in a rising market gets credit for the market's own move. So the track record shows a second number next to it.

For every call in the hit rate, we take its graded move (entry to exit, as above) and subtract what the S&P 500 did over the exact same trading days. A bearish call is compared with betting against the index over those days. The page shows the average and the median of that difference and how many calls finished ahead of the index. Picking directions at random would score about zero, and a rising or falling market cannot lift the number. Like the hit rate, it ignores fees, slippage and taxes.

What gets excluded, and why

• →Moves smaller than 0.5% are recorded but not counted as a win or a miss. A stock that barely moved is not a clean read on whether the direction was right.
• →Stocks trading under $2 million a day in dollar volume are recorded but kept out of the hit rate. A move nobody could trade cleanly should not pad the number, in either direction.
• →A stock that gets halted or delisted right after an alert is marked excluded rather than silently dropped. Dropping those quietly would remove exactly the biggest wins and worst blowups from the record.
• →A price series with a split-sized jump inside the graded window (a close 4x its neighbor, or a quarter of it) is treated as a stock split or bad data, and the call is marked excluded. No real close-to-close move looks like that, and counting one produced a fake +2,000% win before this rule existed.

Exclusions are shown in the ledger with their reasons. They are not deletions.

When the rules change

We tighten our sending rules as the record teaches us what works. The published hit rate counts the calls our current rules make. A call sent under a rule we have since retired keeps its grade forever and stays visible in the ledger, but it is not counted in the rate, because the rate describes the alerts you would receive today. Retired calls are never re-graded and never hidden.

One exception, disclosed here because it matters: on August 27, 2026, while the product was still in beta with no paying subscribers, we changed the grading basis for insider buys (the next-close rule described above) and the set of alerts that count as calls and re-graded the entire record under it, so that one method covers every call rather than two methods being stitched together. That was a one-time reset made before launch, and the numbers under the old method are preserved in our audit notes. From that date forward, grades are locked as described above.

After the window closes

A window is a judgment about how long a catalyst takes to play out, and it can be wrong. So after a call is graded, we keep watching the stock and record where it stood about 30 days after the alert, in a separate internal field. If an event class keeps playing out after its window closes, that shows up in this data, and the fix is a longer window for future calls of that class.

This has already happened once. In August 2026 the data showed stock-offering calls kept moving in the called direction for weeks after their two-day window closed, so the offering window was widened to 10 trading days. The change applies only to calls sent after it was made; every call sent before it stays graded under the two-day window it was sent with.

What we never do with that data is flip old grades. A miss that came good in week three stays a miss on the record, because the version of us that sent the alert was still wrong about the timing, and the record grades the alert we sent, not the alert we wish we had sent.