He states the same mechanism two ways in two interviews, so we pre-registered and ran both. Sectioned by market cap it is false: across 76 survivorship-free parabolic runs drawn from 1,058 detected, the dollar volume at the top is not clustered inside a cap bucket, and the multiplier does not hold constant — it collapses from 121 to 18 to 6 as cap rises, which is the opposite of the stated shape. Sectioned by year it was real, and large. In 2021, a normally-dead micro-cap that printed $1 billion of volume in a day closed lower the next day 69.7% of the time with a median of -9.93%, against a 48.9% / +0.00% base rate in the same population and the same year — +20.7pp, on 234 events, measured from one closing price to the next. By 2023, at $200 million — the number he himself gives for that year — the same measurement returns +5.3pp and a median of -0.17%. He told his audience on camera that the number was falling. He was right. The correction we owe him is that it fell differently than described: what decayed was not the level, it was the effect. And one thing this page does not establish is whether any of it was ever available — a next-day close-to-close return is not a trade, and we have never been able to measure borrow or locate for this population.
Granted, not measured: three things, none of them grudgingly. He makes money, and the record is better-attested than the genre norm — third-party verified by his own account, and he invites direct brokerage verification on camera. We did not audit it and we do not dispute it. His stated facts check out where we can check them: we pulled the three figures he gives from memory on camera and scored them against our survivorship-free daily store — 3 of 3 inside the stated tolerance. He says HPX traded “it only trade around the 1.4 million shares throughout the day”; the store says 1,431,196 shares, 2.2% off. He says it ran to $50; the high was $50.50. On a delisted ghost, three years later, from memory. And he named the decay himself, unprompted, on camera — which is the whole reason the second probe exists. Sources credited and linked: Steven Dux — first-green-day walkthrough (his own channel) · Humbled Trader — the statistics interview with Steven Dux.
Tested: the two stated formulations of one mechanism, each pre-registered before its run. Formulation one, from the first-green-day walkthrough: a parabolic micro-cap tops at roughly the same cumulative dollar volume for a given starting market cap. Formulation two, from the interview: a day’s volume times its average traded price crossing a ceiling means the stock falls the next day — and the ceiling fell from $1 billion in 2021 to $200 million in 2023. 1,058 detected runs → 76 usable on the first; 6,778 normally-illiquid tickers, 47 scored cells, every one of them against a same-year control arm on the second.
Two questions, one per formulation. Does the signal exist at all — conditional on a normally-illiquid micro-cap printing a daily dollar volume above the threshold he names, is the next day negative materially more often, and more deeply, than the matched base rate for the same population in the same year? And does the ceiling decay as he says — is the best-separating threshold in 2021 at least 3× the one in 2023, falling monotonically in between? Behind both, the older question the first probe asked: does the dollar volume at the top cluster inside a market-cap bucket, at a stable multiple of cap?
The frozen prior read: lean FALSE on the signal (~70%), genuinely uncertain on the decay (~50%). The reasoning was that micro-cap parabolas mean-revert on average — but so does the control arm — so we expected a real but small gap failing the 10-point bar in both years. The verdict came out where we guessed and the mechanism did not. 2023 failed as predicted, at +5.3pp. 2021 did not: it cleared the bar twice over at +20.7pp. Being right about the headline for the wrong reason is still being wrong, and the direction of the error runs toward him, not away.
| Parameter | Locked value |
|---|---|
| Population (decay probe) | common stock whose lifetime median daily dollar volume is under $2M — 6,778 tickers. A split-invariant proxy, chosen so no price gate is needed |
| Price gates | forbidden. Prices in this store are split-adjusted as of the pull, so a naive “under $20 = micro-cap” filter returns post-split large caps. Any figure not built on that correction is void |
| Event | a ticker-day in that population whose volume × average traded price crosses the threshold |
| Outcome | next trading day’s close over this day’s close, same ticker. Days with no next-day row are reported, never silently dropped |
| Control arm | every ticker-day in the same population and the same year that did not cross the threshold |
| Threshold grid | $50M · $100M · $200M · $300M · $500M · $750M · $1,000M · $1,500M · $2,000M |
| His stated thresholds | 2021 $1,000M · 2023 $200M — 2022 is a sweep result only; he never gave a number for it |
| Power precondition (decay) | at least 50 events in a year-cell, or it is reported UNDERPOWERED and carries no verdict |
| Years | 2021 · 2022 · 2023 · 2024 · 2025 (2021–2023 primary, the rest for the record) |
| Run detection (market-cap probe) | at least 4× off a local trough within 8 trading days, at least 3 green days, holding twice the trough for 2 days; the top is the local max within 3 days |
| Cap buckets (market-cap probe) | B1 ≤ $50M · B2 $50M – $100M · B3 $100M – $200M, each needing 12 runs to carry a verdict |
| Universe (market-cap probe) | survivorship-free common stock, delisted names included by construction — 1,058 runs detected across 841 tickers, 76 usable, 2012-2026 |
| # | Pre-registered goalpost | Measured | Pass |
|---|---|---|---|
| 1 | Pre-registered goalpost — decay probe — power precondition: at least 50 events in each year-cell we score | 2021 n = 234 · 2023 n = 789 | YES |
| 2 | Pre-registered goalpost — decay probe C1 — the event arm's share of down-next-days beats the same-year control arm by at least 10 points, in both 2021 and 2023, at his own stated levels | 2021 +20.7pp · 2023 +5.3pp | NO — 1 of 2 years |
| 3 | Pre-registered goalpost — decay probe C2 — the event arm's median next-day return is negative and at least 2 points below the control median, in both years | 2021 -9.93pp · 2023 -0.17pp | NO — 1 of 2 years |
| 4 | Pre-registered goalpost — decay probe C3 — the best-separating threshold in 2021 is at least 3× the one in 2023 | 2021 $1,000M vs 2023 $50M — and 2023’s sits on the floor of our own sweep grid | YES |
| 5 | Pre-registered goalpost — decay probe C4 — that threshold falls, or holds, 2021 → 2022 → 2023 | $1,000M ≥ $750M ≥ $50M | YES |
| 6 | Pre-registered goalpost — market-cap probe C1 — the dollar volume at the top is clustered inside each powered cap bucket (middle-50% spread narrower than the median) | B1 1.56 · B2 1.08 · B3 1.89 — the bar is below 1.00 | NO |
| 7 | Pre-registered goalpost — market-cap probe C2 — the bucket medians rise with market cap | $1,221M < $1,268M on the 2 powered buckets; B3 (n = 6) is underpowered and carries no verdict | YES |
| 8 | Pre-registered goalpost — market-cap probe C3 — the multiplier k is order-stable across cap buckets (largest / smallest bucket median no more than 3) | B1 121.3 · B2 18.1 · B3 5.6 — ratio 6.7 on the powered pair | NO |
4 of 8. The two probes split, and the split is the finding. On the decay probe the first question is FALSE by the frozen rule, which required both years and got 1 of 2; the second is CONFIRMED, and the decay ran harder than he described. On the market-cap probe the shape fails outright. What passed and what failed point the same way: the mechanism he named was doing real work at the start of the window and is not doing it at the end, and the sectioning he gave for it in the earlier interview was never the right sectioning.
| Figure | Value | What it means |
|---|---|---|
| 2021, at the ceiling he states for that year ($1 billion in a single day) | 69.7% closed lower the next day · median -9.93% · n = 234 | a normally-dead micro-cap that printed an enormous day |
| the same-year control arm — every day in the same population that did not cross | 48.9% · median +0.00% · n = 599,363 | the coin flip the event arm is measured against |
| the 2021 gap | +20.7pp · -9.93pp on the median | large, base-rate-controlled, and it is his effect |
| 2023, at the ceiling he states for that year ($200 million) | 54.0% · median -0.17% · n = 789 | the same measurement, two years later |
| the 2023 gap | +5.3pp · -0.17pp on the median | under the 10-point bar we locked before the run |
| the gap at each year’s own best-separating threshold | +20.7pp → +10.5pp → +7.4pp → +4.4pp → +4.5pp | four consecutive years of decay toward the base rate, 2021–2025 |
| the threshold-response slope — the gap at $50M against the gap at $750M, the highest threshold with enough events to score in both years | 2021: +9.3pp → +19.6pp · 2023: +7.4pp → -7.1pp | in the first year the bigger the day the bigger the fade; in the second that relationship has inverted |
| the market-cap version — the multiplier k by cap bucket (n = 76 runs) | 121.3 · 18.1 · 5.6 | the claim is that k is roughly constant; measured, it collapses as cap rises |
| the same-bucket spread of dollar volume at the top (middle 50%, over the median) | 1.56 · 1.08 · 1.89 | “the same dollar amount” requires this below 1.00; no bucket is |
| his own on-camera facts, checked against our survivorship-free store | 3 of 3 inside the per-claim tolerance | the steelman — and it is a measurement, not a courtesy |
Was the 2021 result a handful of meme names? No. Dropping the five largest events in each year moves 2021 from +20.7pp to +21.0pp — it goes up — and leaves 2023 unchanged at +5.3pp (+5.35pp against +5.26pp before the drop). Neither year is carried by its tail.
Did the losers quietly delist before we could measure them? That would bias toward the claim, so we counted it. Rows with no next-day close: 0 of 234 in the 2021 cell and 0 of 789 in the 2023 cell.
The single cleanest diagnostic in the whole probe is the sign of a slope. In 2021 the gap grows as the threshold rises — +9.3pp at $50M → +12.8pp at $200M → +17.6pp at $500M → +19.6pp at $750M — which is exactly what a real supply-exhaustion ceiling should do: the bigger the day, the bigger the fade. In 2023 it does the opposite. It shrinks as the threshold rises and then turns negative — +7.4pp at $50M → +5.3pp at $200M → +3.3pp at $500M → -7.1pp at $750M. That slope inverted between the two years. It is the reason we read the result as an effect that died rather than a ceiling that moved. Every step above is a cell with at least 50 events in both years; the 2023 gap runs further negative above $750M, but on 23 events, which is under our own power floor and carries no verdict. It is in the ledger, marked not powered.
And on the older probe, the market-cap version fails the same way twice more. Restricted to the more recent half of the record (n = 67) the multiplier still spans 6.4× across cap buckets; with the three largest names removed (n = 73) it still spans 6.6×, against a pre-registered bar of 3. The single most extreme row in that ledger is BMNR: a $9M company that traded $10.3 billion on the way to its top, for a multiplier of 1,137.6 against a bucket median of 121.3. A rule that says “roughly the same dollar amount” has to survive that row, and it does not.
Every event number on this page is scored against a control arm, and the control arm is the reason the numbers mean anything. It is every ticker-day in the same population, in the same year, that did not cross the threshold. Without it this probe would measure micro-cap drift and call it a signal.
The control arm is remarkably stable: 2021 48.9% down-next-days on 599,363 non-crossing ticker-days, drawn from a population of 3,117 names and 599,979 ticker-days · 2022 48.6% down-next-days on 690,823 non-crossing ticker-days, drawn from a population of 3,271 names and 691,419 ticker-days · 2023 48.7% down-next-days on 643,528 non-crossing ticker-days, drawn from a population of 3,197 names and 647,472 ticker-days · 2024 50.0% down-next-days on 610,798 non-crossing ticker-days, drawn from a population of 2,956 names and 612,744 ticker-days · 2025 50.5% down-next-days on 600,756 non-crossing ticker-days, drawn from a population of 2,921 names and 603,225 ticker-days. A coin flip, every year, and a median next-day return that never lands further than 0.05 points from zero in any of them. That is the line the event arm has to beat, and in 2021 it beat it by +20.7pp.
On the market-cap probe the control is a different shape: a bootstrap confidence interval on each bucket median. Those intervals are wide, and the width is the finding — a claim that parabolas top at roughly the same dollar amount needs a narrow one. The middle-50% spread over the median came in at 1.56 · 1.08 · 1.89 against a pre-registered bar of 1.00.
“You tested the market-cap version. I trade the float version. My dollar block is trapped shares times price — that is float, not market cap, and you just told your audience you cannot measure float.”
That objection is correct, and we are conceding it rather than answering it. The house rule here is that a steelman gets answered with a run, not a caveat. This one cannot be, and saying so plainly is the finding: company filings publish no share-float tag, so the float-sectioned version of his rule is untestable in principle with any data we can obtain. Settling it needs a float-data vendor, and that is the first item in §8.
The second objection is ours, against ourselves. 2023’s best-separating threshold is $50M — which is the bottom cell of our own sweep grid. An estimator that lands on its own boundary has run out of road. So the honest reading is not that the ceiling relocated downward to a precise new level; it is that by 2023 there is no threshold effect left to locate and the sweep returned its least-bad cell. That is why this report publishes the ladder and the slope and never a decay multiple: the multiple is an artifact of where we put the floor of the grid, and a number we cannot defend is a number we do not print.
The third is the one that matters most and it has no answer here at all. Nothing above is a trade. There is no entry, no exit, no slippage, no commission, no borrow fee and no locate in any figure on this page. From three unrelated short-side studies of our own, the binding constraint on this kind of position was never the borrow fee — it measured near 1% of gross on an intraday hold — it was whether a locate existed at all, and in all three the apparent edge sat entirely inside the part of the market that could not be borrowed. Those are measurements about our own strategies, not about his, and they carry no verdict on him. Whether the 2021 effect was ever available to anybody is not established on this page.
The sources, credited and linked. Everything we tested is a claim he makes on camera in one of two videos: Steven Dux — first-green-day walkthrough (his own channel) · Humbled Trader — the statistics interview with Steven Dux. The market-cap formulation comes from the first; the year-sectioned formulation and both of his stated ceilings come from the second. Watch them and check us — that is the point of naming them.
Our sectioning is not his sectioning, and he says so. He keys on float. We keyed the first probe on market cap because market cap is what we can reconstruct point-in-time from filings and float is not published anywhere as a share count. Those are different quantities for exactly the names in question, where a small free float sits under a much larger share count. Our market-cap result is a result about the market-cap version of his rule and nothing more.
Our clock is not his clock. He trades this intraday, off the open, with a read on first weakness. We measured one closing price to the next, because our minute-bar universe was selected on lifetime median dollar volume and a pump’s lifetime median is tiny by construction — 3 of 3 of the tickers he names on camera are missing from it entirely. Which of the two clocks is kinder to the rule is something we did not measure and will not guess.
And the part of his method we never went near is the part he says carries it. Sizing, crowding limits, the categories he refuses to touch, the discretion to stand down. He names the sector that beats him, on camera, unprompted. None of that is in any number here.
| # | Year | Cell | Threshold | Events | Censored | Event %neg | Event median | Control %neg | Control median | Gap %neg | Gap median | Powered |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | 2021 | sweep | $50M | 8,175 | 28 | 58.1% | -1.07% | 48.8% | +0.00% | +9.3pp | -1.07pp | yes |
| 2 | 2021 | sweep | $100M | 3,906 | 17 | 59.4% | -1.45% | 48.9% | +0.00% | +10.6pp | -1.45pp | yes |
| 3 | 2021 | sweep | $200M | 1,816 | 6 | 61.7% | -2.29% | 48.9% | +0.00% | +12.8pp | -2.29pp | yes |
| 4 | 2021 | sweep | $300M | 1,125 | 5 | 62.7% | -3.30% | 48.9% | +0.00% | +13.8pp | -3.30pp | yes |
| 5 | 2021 | sweep | $500M | 567 | 4 | 66.5% | -6.44% | 48.9% | +0.00% | +17.6pp | -6.44pp | yes |
| 6 | 2021 | sweep | $750M | 356 | 1 | 68.5% | -8.20% | 48.9% | +0.00% | +19.6pp | -8.20pp | yes |
| 7 | 2021 | sweep / HIS STATED LEVEL / best-separating | $1,000M | 234 | 0 | 69.7% | -9.93% | 48.9% | +0.00% | +20.7pp | -9.93pp | yes |
| 8 | 2021 | sweep | $1,500M | 111 | 0 | 69.4% | -14.24% | 48.9% | +0.00% | +20.4pp | -14.24pp | yes |
| 9 | 2021 | sweep | $2,000M | 57 | 0 | 68.4% | -16.80% | 48.9% | +0.00% | +19.5pp | -16.80pp | yes |
| 10 | 2022 | sweep | $50M | 3,676 | 12 | 56.8% | -0.63% | 48.6% | +0.00% | +8.2pp | -0.63pp | yes |
| 11 | 2022 | sweep | $100M | 1,736 | 4 | 57.7% | -0.65% | 48.6% | +0.00% | +9.1pp | -0.65pp | yes |
| 12 | 2022 | sweep | $200M | 860 | 0 | 57.1% | -0.54% | 48.6% | +0.00% | +8.5pp | -0.54pp | yes |
| 13 | 2022 | sweep | $300M | 550 | 0 | 55.8% | -0.45% | 48.6% | +0.00% | +7.2pp | -0.45pp | yes |
| 14 | 2022 | sweep | $500M | 190 | 0 | 52.6% | -0.31% | 48.6% | +0.00% | +4.0pp | -0.31pp | yes |
| 15 | 2022 | sweep / best-separating | $750M | 66 | 0 | 59.1% | -1.58% | 48.6% | +0.00% | +10.5pp | -1.58pp | yes |
| 16 | 2022 | sweep | $1,000M | 30 | 0 | 63.3% | -7.80% | 48.6% | +0.00% | +14.7pp | -7.80pp | no |
| 17 | 2022 | sweep | $1,500M | 8 | 0 | 50.0% | -5.68% | 48.6% | +0.00% | +1.4pp | -5.68pp | no |
| 18 | 2022 | sweep | $2,000M | 4 | 0 | 50.0% | -6.97% | 48.6% | +0.00% | +1.4pp | -6.97pp | no |
| 19 | 2023 | sweep / best-separating | $50M | 3,284 | 15 | 56.1% | -0.36% | 48.7% | +0.00% | +7.4pp | -0.36pp | yes |
| 20 | 2023 | sweep | $100M | 1,500 | 5 | 56.1% | -0.36% | 48.7% | +0.00% | +7.3pp | -0.36pp | yes |
| 21 | 2023 | sweep / HIS STATED LEVEL | $200M | 789 | 0 | 54.0% | -0.17% | 48.7% | +0.00% | +5.3pp | -0.17pp | yes |
| 22 | 2023 | sweep | $300M | 527 | 0 | 53.5% | -0.17% | 48.7% | +0.00% | +4.8pp | -0.17pp | yes |
| 23 | 2023 | sweep | $500M | 196 | 0 | 52.0% | -0.08% | 48.7% | +0.00% | +3.3pp | -0.08pp | yes |
| 24 | 2023 | sweep | $750M | 60 | 0 | 41.7% | +0.21% | 48.7% | +0.00% | -7.1pp | +0.21pp | yes |
| 25 | 2023 | sweep | $1,000M | 23 | 0 | 39.1% | +0.19% | 48.7% | +0.00% | -9.6pp | +0.19pp | no |
| 26 | 2023 | sweep | $1,500M | 5 | 0 | 40.0% | +0.00% | 48.7% | +0.00% | -8.7pp | +0.00pp | no |
| 27 | 2023 | sweep | $2,000M | 2 | 0 | 50.0% | -0.56% | 48.7% | +0.00% | +1.3pp | -0.56pp | no |
| 28 | 2024 | sweep | $50M | 5,220 | 13 | 53.9% | -0.24% | 50.0% | -0.00% | +3.9pp | -0.24pp | yes |
| 29 | 2024 | sweep | $100M | 2,755 | 8 | 54.3% | -0.29% | 50.0% | -0.00% | +4.3pp | -0.28pp | yes |
| 30 | 2024 | sweep / best-separating | $200M | 1,439 | 3 | 54.4% | -0.25% | 50.0% | -0.00% | +4.4pp | -0.25pp | yes |
| 31 | 2024 | sweep | $300M | 888 | 2 | 52.5% | -0.13% | 50.0% | -0.00% | +2.4pp | -0.13pp | yes |
| 32 | 2024 | sweep | $500M | 376 | 1 | 50.5% | -0.02% | 50.0% | -0.00% | +0.5pp | -0.01pp | yes |
| 33 | 2024 | sweep | $750M | 174 | 0 | 50.6% | -0.02% | 50.0% | -0.00% | +0.5pp | -0.01pp | yes |
| 34 | 2024 | sweep | $1,000M | 89 | 0 | 47.2% | +0.10% | 50.0% | -0.00% | -2.9pp | +0.10pp | yes |
| 35 | 2024 | sweep | $1,500M | 42 | 0 | 45.2% | +0.85% | 50.0% | -0.00% | -4.8pp | +0.85pp | no |
| 36 | 2024 | sweep | $2,000M | 31 | 0 | 45.2% | +0.79% | 50.0% | -0.00% | -4.9pp | +0.79pp | no |
| 37 | 2025 | sweep | $50M | 10,932 | 31 | 54.1% | -0.37% | 50.5% | -0.04% | +3.7pp | -0.32pp | yes |
| 38 | 2025 | sweep | $100M | 6,008 | 14 | 54.5% | -0.38% | 50.5% | -0.05% | +4.0pp | -0.33pp | yes |
| 39 | 2025 | sweep | $200M | 3,115 | 9 | 55.0% | -0.37% | 50.5% | -0.05% | +4.4pp | -0.32pp | yes |
| 40 | 2025 | sweep / best-separating | $300M | 2,050 | 5 | 55.0% | -0.33% | 50.5% | -0.05% | +4.5pp | -0.28pp | yes |
| 41 | 2025 | sweep | $500M | 1,066 | 1 | 54.5% | -0.37% | 50.5% | -0.05% | +4.0pp | -0.32pp | yes |
| 42 | 2025 | sweep | $750M | 626 | 1 | 53.5% | -0.24% | 50.5% | -0.05% | +3.0pp | -0.20pp | yes |
| 43 | 2025 | sweep | $1,000M | 444 | 1 | 53.8% | -0.23% | 50.5% | -0.05% | +3.3pp | -0.18pp | yes |
| 44 | 2025 | sweep | $1,500M | 270 | 0 | 50.4% | -0.03% | 50.5% | -0.05% | -0.2pp | +0.02pp | yes |
| 45 | 2025 | sweep | $2,000M | 184 | 0 | 50.5% | -0.03% | 50.5% | -0.05% | +0.0pp | +0.02pp | yes |
| 46 | 2021 | ex-top-5 robustness | $1,000M | 229 | 0 | 69.9% | -9.86% | 48.9% | +0.00% | +21.0pp | -9.86pp | yes |
| 47 | 2023 | ex-top-5 robustness | $200M | 784 | 0 | 54.1% | -0.17% | 48.7% | +0.00% | +5.3pp | -0.17pp | yes |
A rule with a number in it has an expiry date, and almost nobody who sells one tells you that. He did. He stood in an interview and said the level had fallen from $1 billion to $200 million in two years. We went and measured it, and the direction he gave was right. The part he did not say — possibly did not know — is that the level was not the thing that was decaying. The effect was. In 2021 the bigger the day, the bigger the next-day fade; by 2023 that relationship had inverted, and a buyer handed a new number would be handed a live-looking parameter for a mechanism that had gone quiet.
So the transferable question is not “what is the number now.” It is “compared to what.” 69.7% of those days closed lower, which sounds like a lot until you learn that 48.9% of all days in that population closed lower too. The gap is the claim. Everything else is the weather.
And the honest end of it: we measured a price regularity, not an opportunity. Whether anyone could have stood in front of it — found the borrow, got the locate, sized it without moving it, survived the left tail he himself describes — is a question this page does not answer and did not try to. That is the reopen bounty, and it is still open.
Everything we make is free — the reports, the ledgers, all of it. The only thing that costs us real money is the market data.