I Graded Four Years of My Own Public Calls. Here's the Scoreboard.

I graded 139 of my own market calls — every one I could find from February 2022 through July 2026 — against what price actually did afterward. 132 of them reach the published record; the other 7 are named and counted below rather than quietly dropped. Of those 132, 112 are gradeable, and 112 is the denominator for every rate on this page: 33 clean wins (33 of 112, about 29%), 40 more that were directionally right but imperfect (40 of 112, about 36%), 13 partial (13 of 112, about 12%), and 26 outright losses (26 of 112, about 23%). Split by direction, 49 of my 68 bearish-side calls were at least directionally right — 49 of 68, about 72% — while on the bullish side it was 20 of 39, about 51%.

That is the whole scoreboard, stated up front with its denominator attached, misses included. The rest of this page covers what got excluded and why, where the calls came from, how each one was graded, what the misses look like up close, and what a record like this can and cannot tell you.

What got excluded, and why it matters more than the rates

Two subtractions stand between the 139 calls I graded and the 132 on the published record, and both are stated here because a denominator you cannot audit is not a denominator.

• 6 calls are held because I cannot prove they were public. Every on-air call carries a venue tag. Only the ones spoken on a session an audience could actually have watched are published as proof — a call nobody outside the room could have followed is not public proof, whatever it scored. Six on-air rows are recorded but their public recording is unconfirmed, so they are held. Two of them scored well. They are still held. That rule fails closed on purpose, and it costs me more than it saves me.
• 1 call is a standing process rule with no instrument attached — a method statement, not a prediction on anything, so there is nothing for price to settle.

That leaves 132. Then, inside the 132, 20 are ungradeable and sit outside the denominator: 12 with no independent reference or no forward horizon, 4 whose window has not elapsed, and 4 too recently made to score. They are not published with a direction and no result, because a call with no settled outcome is not a retrospective — but the count travels with the record so the set is not quietly trimmed. 132 minus 20 leaves the 112.

Those are the same figures the record hub publishes, in the same arithmetic. The hub leads with the rows it actually prints — 112 published calls on 67 instruments, 2022-02-14 to 2026-07-13, of which 33 are clean wins and 26 are outright losses — and then states, separately, that a further 20 calls on the record are ungradeable and withheld. 112 plus 20 is the 132 on this page. The hub recomputes all of it from the database on every build; this page states it in prose. If the two ever disagree, the hub is right and this page is stale.

Why would anyone publish their own misses?

Because a record without misses is not a record. It is marketing.

Every trading educator claims a track record. Almost none of them show you one that is dated, unedited, and has losses printed in it. The reason is simple: an honest scoreboard is uncomfortable. Read honestly, parts of mine argue against me. The most recent full on-air session on the record, 2026-07-09, produced nine gradeable calls and not one clean win — five outright losses and four that were only directionally right. I am publishing that sentence anyway, because the alternative — a highlight reel — proves nothing to anyone who thinks for five seconds.

The scoreboard's job is not to prove the calls were great. Its job is to show you exactly who you are dealing with: someone who logs the call before the outcome exists, prints the result either way, and withholds the rows he cannot substantiate even when they scored.

There was a selfish reason too. Grading myself taught me things no journal ever did. The bearish-versus-bullish gap alone changed how I weight my own ideas.

Where did the 132 calls come from?

Three sources, two eras. Every call was pulled with its original date, direction, timeframe, and exact wording — nothing reworded, nothing added later. The counts below are the published record only; the 6 held on-air rows and the 1 process statement are already removed.

The 2026 on-air calls carry a third-party timestamp: they were spoken on recorded sessions, so the recording platform owns the timestamp, not me. I cannot move it, backdate it, or soften the wording after the fact. That is also why the venue tag exists — the same recording that proves the timestamp is what proves the call was public, and where it cannot, the row does not run.

Notice what the table tells you before you compute a single percentage: the oldest slice graded best and the newest, live slice graded worst. Of the 14 on-air calls that are gradeable, 7 — exactly half — were outright losses, against 10 of 62 in the 2022 archive. I am not burying that; it is the most honest shape in the dataset, and the pattern of my older material outperforming my newer material shows up elsewhere too — I wrote about it in my 2022 posts beat my 2026 posts.

How was each call graded?

Every call was scored against an independent reference: the instrument's actual price action after the call, inside the call's own stated timeframe, using publicly available daily price data. Not my memory of what happened. Not a screenshot I picked.

Two details matter more than the labels. First, the denominator, which is stated above and repeated next to every rate on this page: 20 of the 132 are ungradeable and excluded. Excluding them is legitimate, but it is also a judgment call, so the count is disclosed rather than hidden — and so is the fact that they are withheld from the record pages entirely instead of appearing there as a named instrument with a direction and no outcome.

Second, the conservative rule. Wherever a call's framing was genuinely ambiguous, it was graded to the less favorable outcome. One 2026-07-09 on-air call carried a bearish breakdown tag on a note that also read as a search for a bottom; the bottom-fish reading would have finished ahead, and the record grades it to the bearish tag, as a loss. Wherever a self-reported result could not be independently corroborated, it was downgraded to the verifiable move. The good numbers are only believable if the grading leans against me.

What does the full scoreboard say?

Of the 112 gradeable calls:

Add the first two rows and 73 of 112 — about 65% — were at least directionally right. That is the generous cut. The strict cut is the top row alone: 33 of 112, about 29%, were right the way I said it, when I said it.

The most useful split in the whole database is direction. Of the 68 gradeable calls made to the bearish side, 49 were at least directionally right — 49 of 68, about 72%. Of the 39 made to the bullish side, 20 were — 20 of 39, about 51%. (The remaining 5 of the 112 were neutral or range calls with no side to score.) The record says, in plain terms: this operator's read is strongest on weakness and on the broad regime, and weakest on picking individual longs. Both rates come from one operator over one stretch of market, and neither is a forecast about the next call. I did not know either number before I ran the grading. Now I do, and so do you.

What do the misses look like up close?

Aggregates hide texture. Here are three graded calls from a single day of the live record — 2026-07-09, all of them on the published record — exactly as they resolved.

2026-07-09 — AMD, long. Loss. On air: "AMD did give the signal today… a reason to go long… The sector supports it. The market is positive." It worked for exactly one day, up about 2%. Then it gave that back and kept falling — down about 9.3% from the call by July 17. Logged as a loss.

2026-07-09 — S&P 500 futures, long. Loss. On air: rising into a cycle peak, with the invalidation stated out loud as a close below that session's own low. Within a week the index closed through it twice. The call named the level that would prove it wrong, and price found it. Logged as a loss.

2026-07-09 — S&P 500, the peak read. Directionally right, not a clean win. On air: "It's now probably peaking… long-term buyers that are still trying to hold up the market, even though it's definitely not going to work out for them." The index made a marginal higher high over the next few sessions, then rolled over to about 1% below the call-day level within two weeks. The direction and the shape were right; the exact top came about a session later than "now," so it grades as directionally right rather than clean.

Same day, same desk, same process: two failed longs and a regime read that was right in shape but not in timing. That is what a real record looks like at close range — not a montage of wins, but a texture of partial credit and honest failure around a process that has a measurable shape.

What can this record tell you — and what can't it?

Here is the honest boundary of the whole exercise:

1. It is not a return. There are no dollar figures on this page, deliberately. A graded call is a public statement scored against price. It is not a filled trade with size, fees, slippage, or an account behind it, and I will not dress it up as one.
2. It predicts nothing. Past results do not indicate future results. 49 of 68 bearish-side calls landing directionally right across one period is a description of what happened, not a promise about the next call. Anyone who tells you otherwise is selling something with the honesty removed.
3. Grading involves judgment. I used a conservative rule and independent price data, and disclosed both exclusions — the 7 subtracted before the record and the 20 excluded inside it — but a different grader could score edge cases differently. Because every published call is dated and public, anyone can re-grade the whole set and check me.
4. Small slices are noise. A single session of nine graded calls means almost nothing on its own, and neither does any one instrument's handful of rows. The value is in the accumulated shape across 112, and even that is one operator, in specific market conditions.
5. What it can tell you is the shape of a read. Stronger on the bearish side and on regime, weaker on individual longs. That is genuinely useful information — arguably more useful than a win rate — because it tells you how to weight what I say, not just whether to listen.

If you want to run this kind of grading on your own trading, the full scoring model is written up in how a trading educator should score their own calls. If you are evaluating someone else's claimed record, start with how to verify a trading guru's credentials — and note that some days the correct call is no call at all, which is its own discipline: the null trade. The method that generates the calls in the first place — the cycle work behind the scans and the regime reads — is documented at the methodology hub.

Questions traders ask

Is 33 clean wins out of 112 any good?

On its own, the number tells you very little, and I would distrust anyone who leads with a bare rate in either direction — including me, which is why that 29% is written on this page as 33 of 112 every time it appears. Everything else depends on how strictly "clean win" is defined. My definition requires direction, the stated outcome, the stated window, and no serious adverse move first — and ambiguous cases were graded against me. A loosely scored 80% and a strictly scored 29% are not comparable numbers. Read the definition next to the number, check the denominator is stated, and check that misses sit in the same table with the same prominence. If any of the three is missing, you are reading marketing.

Why do the exclusions get their own section?

Because the exclusions are where a track record is normally faked. Anyone can hit a high rate by quietly dropping the rows that went badly or cannot be checked. So this page states all of them: 6 held for unconfirmed venue, 1 process statement with no instrument, and 20 ungradeable inside the published 132 — 12 with no independent reference, 4 whose window has not elapsed, 4 too recent to score. Two of the six held rows scored well and are still held. Subtract them all and you get the 112 that every percentage on this page divides by.

Why not just show the winners?

Because a highlight reel cannot be falsified, which means it cannot be believed. Anyone can assemble winners after the fact. The only record that counts as evidence is one that was logged before the outcomes were known, with the losses printed. Publishing the misses is not generosity — it is the entire mechanism that makes the hits mean anything.

Who did the grading, and why should I trust it?

The calls were scored against independent public price data, inside each call's own stated timeframe, with ambiguity resolved to the less favorable read and uncorroborated self-reports downgraded to what the data could verify. But you do not have to trust that: the published calls are dated and public, so the whole set can be re-graded by anyone with a price chart and patience. A self-graded record that is also dated, public, conservatively scored, and explicit about what it left out says "check me," not "trust me."

Does a good record mean your next call is more likely right?

No. Nothing here is a prediction, a promise, or advice to buy or sell anything. What the record gives you is weighting information — which kind of read has historically been stronger and which weaker, across a stated and countable number of calls — and proof of a process that logs before the outcome and publishes either way. That is information about an operator, not a forecast about the market.

Keep reading