Overhead Supply at 78.7%: Why Retail Loses Short-Term

This is a record of a decision, not an analysis. For the past several months I wrote a weekly plan for every position I held. How much volume sat above my entry, what the reward-to-risk ratio was, where the stop belonged: I calculated all of it per stock and filed it as a document every week. I am shutting that process down this week.

What triggered it was not a loss. It was a win. In a setup my own weekly plan explicitly flagged as do not buy, the stock ran 23.6% in seven sessions. It is not quite right to say the tools were wrong. It is more accurate to say I was asking them a question they cannot answer, every single week. I will start with what the weekly plan actually was and how it ran, then walk through what I saw and why I bought.

First, what my weekly plan actually was

One document per stock. The Circle document I wrote last week runs 639 lines, and I wrote that for 20 US names and 11 Korean names in a single week. I lead with the page count because, as you will see, the page count itself turned out to be one of the reasons this failed.

The order of questions was fixed

The core of it was asking three questions in a set order, where an earlier question overrules a later one.

Volatility was deliberately kept out of those three. It was not there to predict direction but to serve purely as a filter for position size and stop width.

Four operating principles enforced that order, and all four ran every week.

I was particularly fond of that last one. "I still have 8% before my stop" is not a reason to own something. If you would not buy it here today, you do not have a reason to keep holding it either.

How overhead supply got calculated

Question two, breakout capacity, was the largest block in every document. I ran two windows and cross-checked them.

For each bin I estimated the share of volume on a VWAP-weighted basis, then attached turnover days and a digestion rate, meaning the last 20 days of volume divided by the total volume resting in that bin. Every remaining resistance level above the current price got labeled a gate, split into cleared and uncleared.

Here is what the August 17 Circle document printed. Volume above the current price was 63.7% on the primary window and 78.7% on the long window, and of the six remaining gates, not one was cleared. My threshold was 20%, so the long window came in at four times the limit.

A week ran like this

To use the same Circle document as the example, here is what was registered for the week of August 17. The stop was a close below $65.94 on the full 21 shares. The exit ladder was 10 shares at $75.80, 5 at $80.00 and 6 at $87.30. If a rung filled, the stop was to be re-registered immediately on the remainder, written out in advance as 21 to 11 to 6 to 0 shares. Adding to the position was locked behind three conditions that all had to clear, and that week all three were unmet, so it stayed locked.

I still do not think the design itself was bad. It was built to stop me from rewriting the plan mid-session because my nerve went, and it achieved that. The problem was somewhere else.

So why call it a failure

Four reasons. The first three get taken apart with data later in this post. The fourth only appears here.

First, 20 documents produced 0.02 percentage points of excess return.

The cost-weighted return on 18 US names for the third week of August was +1.59%. Strip out three names, Circle, Robinhood and IAU, and it becomes -2.39%. QQQ over the same stretch was -2.41%. Identical to the decimal. The alpha from the other 15 names was zero. And those three did not rise because of stock research either. On August 19 the Treasury expanded long-end buybacks, the dollar weakened, and crypto and gold rose together. One macro event produced the entire week's return.

Second, the stops were sitting inside the noise.

Of 31 positions, 13 touched their stop during the week, and 10 of those closed back above the line by Friday. 42% got hit and most came back. That is less about a rough tape than about where I drew the line, and I will get to the sigma math later.

Third, the buy gate was structurally shut.

For reward-to-risk to clear 2, the gates above have to be far away or absent. My account was neither, because 22 of 26 positions were in a downtrend. Reviewing 19 names, 8 could not produce a target at all or had a target below my cost basis. That is why the strategy review I wrote on August 21 concluded, literally, "do nothing."

Fourth, the documents did not actually run themselves.

This one differs from the other three in kind. It is an operations failure, not a tool failure. Among the 11 scored items for the week were these.

The expensive one was Doosan Robotics. The stop was either 67,900 won or 70,000 won and I never picked that week. The weekly low came in at 68,500 won, precisely between them. At 67,900 it never triggers. At 70,000 it does. Deferring a decision does not leave you holding. It leaves you unable to grade yourself.

The two items carried five times each say the same thing. Once you are writing 600-line documents for more than 20 names, you no longer have the capacity to follow through on the one-line action you assigned yourself last week. I assumed page count would buy accuracy. In practice page count ate execution.

And then the decisive one

You could read all four of those as fixable. Then this week produced the last one.

If I had followed my own rules, the trade this post is about would never have happened. On August 11 Circle had a reward-to-risk of 0.98, which fails question three, and 78.7% overhead supply with zero cleared gates, which fails question two. The rules said do not buy. I bought. Seven sessions later it was up 23.6%.

Put plainly, the weekly plan lost twice. When I followed it I could not buy anything, and when I ignored it I made money, which exposed the rule as wrong.

So the rest of this post covers two things. Why the quantitative tools failed in that particular setup, and if that is true, which horizon an individual should actually be fighting on. Starting with what I saw and why I bought.

The absence of news was itself the signal

From July 24 to August 6, Circle spent 10 sessions boxed between $57.84 and $68.40. Then on August 7 it cleared the box high of $65.67 on a green candle. I went looking for the news. I found nothing.

The read I made there is the whole trade. Not I bought despite no news, but I treated the absence of news as the signal itself.

Counting the volume shows the observation was accurate. Below are multiples against the 20-day average.

There is a clean line through that table. Every advance before the news came on below-average volume, from 0.57x to 0.93x, and every advance after the news came on twice the average. A 5% gain on 0.9x volume is not rotation, it is the absence of sellers. Almost nobody wanted out, so modest buying moved the price.

One correction for accuracy. The day I actually bought, August 11, printed 14.3M against a 20-day average of 11.8M, which is 1.21x. Above average. So the low-volume advance was not a property of August 11 specifically but of the box that preceded it, on July 27, July 30 and August 7. That said, next to the July spikes of 36.8M on 7/10 and 38.1M on 6/30, August 11's 14.3M is about a third. The reading that it broke out of a range where volume had died still holds for the stretch as a whole.

A low-volume green candle alone predicts nothing

This part of the data does not support me, so here it is as-is. Splitting every up day in Circle's 2026 by volume multiple:

There is no difference. Buying purely on a low-volume green candle is a coin flip. What paid this time was not one condition but four of them stacking.

Pull any one out and there is no case left. That is exactly what the table above is saying.

And this is not an exception to my own rule that buying without a reason to rise is chasing. I think it is a precise application of it. Not knowing the reason and there being no reason are different things. Price and volume were the evidence that somebody held a reason, and eight to ten sessions later the news confirmed it.

Laid out session by session from the box through August 21, with the news marked, it looks like this.

The stop was the 20-day, the target was the 60-day

The exit design mattered as much as the entry. I set the stop at the 20-day moving average, for a simple reason: a rising 20-day comes up to meet you, so the loss shrinks on its own as time passes. The target was the 60-day, and if price cleared and held above it, the 60-day would become the new stop.

Here are the numbers at entry.

The designed reward-to-risk was 0.98. Below 1. Under the risk framework I was using, that entry does not clear the gate at all. More on that shortly.

First, whether the design worked. The 20-day did come up.

The 20-day rose 8.8% from $64.36 to $69.40, and the stop distance shrank from -9.58% to -2.50%. The premise that a rising 20-day would shrink the loss played out exactly. It is not risk-free yet, since MA20 at $69.40 is still 2.50% below the $71.18 entry, but another week or two of strength takes it past.

The 60-day target got cleared on August 19.

The interesting part is that the 60-day did not sit still. It fell from $88.78 on July 22 to $73.41 on August 21 while the close came up, and they crossed on August 19. Three consecutive closes above it satisfies my hold condition.

One thing to state honestly. I did not take the exit at the 60-day. Selling at the $78.59 close on the breakout day would have made $29.64 on four shares instead of $68.32. I actually sold on August 21 at $88.26 for $68.32. I broke the plan and got a better result. That is not a brag, it is homework, because unless I decide in advance what to do the next time this setup appears, it is not a repeatable decision.

August 21 was a reset, not an add

I bought the same stock again, but it was a different act. Selling four and buying three on August 21 closed the old position and opened a new one.

The first was a breakout buy stopped below the 20-day. The second is trend continuation stopped at the 60-day, premised on price holding above it. Same ticker, different strategy. Using the prior high of $118.85 from May 22 as the target, the second entry carries a reward-to-risk of 1.76, better than the first at 0.98.

The trade-off is a wider stop at -18.32%. At three shares the dollar risk is only $49.39, but applying that same percentage to 11 shares would exceed what I can absorb. Widening the stop requires shrinking the size, and that pairing is the whole design. Which is why it was three shares and not eleven.

Fills and result

These are computed from fill prices rather than taken off a settled statement, so FX, commissions and taxes are excluded.

And my weekly plan said do not buy this

This is where the post gets to its point. The Circle document I wrote on August 17 says, in writing: "78.7% of volume above, and not one cleared gate."

Overhead supply said no. Reward-to-risk said no at 0.98. Seven sessions later it was up 23.60%.

That should not be read as the tools being useless. The conclusion I reached is different.

Overhead supply and reward-to-risk are not unimportant. The problem was that I could not read the sentiment attached to a policy expectation. Stocks move on numbers sometimes, but not always.

Overhead supply measures location, not strength

Overhead supply measures exactly one thing: how many people are trapped at that price. For it to act as resistance, one more condition has to hold. Those people have to want out.

The same 78.7% is a wall in one regime and a corridor in another. Overhead supply tells you where. Sentiment decides whether it matters.

There was a natural control group

I did not construct this. It was already in the chart. Circle touched the same $71 level twice.

After July 21 it broke down, -6.92% on 7/22 and -6.02% on 7/23. After August 11 it rose, +0.17% on 8/12 and +5.75% on 8/13.

Same price, same volume multiple, same supply profile. Opposite outcomes. July 21 even had more volume. Volume does not separate them either.

Exactly one thing differed. On August 7 the Senate opened voting procedure on the CLARITY Act, and on July 21 it had not. In July there was no reason, so trapped holders sold. In August there was a reason, so they did not. The supply sat at the same level both times. Only the intent of the holders changed.

And that condition, sentiment and policy expectation, appears nowhere in my weekly plan. The August 17 document calculated 78.7% correctly. It simply never measured whether that 78.7% wanted to sell.

Reward-to-risk has the same structure

The ratio is a snapshot of right now, and direction is set by sentiment.

A designed 0.98 became a realized 2.50, 2.6 times higher. But the ratio was not wrong. It said, correctly, that the structure as of August 11 capped the move at $77.87. What was wrong was the premise that the structure would hold. Sentiment arrived, gates above fell several per day, and the target line itself came down. Numerator and denominator both moved.

Both measure the terrain of this exact moment accurately. Neither measures whether the terrain will be the same tomorrow. And terrain is changed by sentiment. My weekly plan was measuring terrain only. Supply percentage, ratio, stop distance, gate locations. Nowhere in the document was there a line for the terrain changing inside a single week.

Three wrong turns worth blocking off in advance.

I counted 632 instances and got a coin flip

"Sometimes people sell into overhead supply and sometimes they do not" was my observation. I checked it with numbers rather than instinct. Across 10 held and watched names from January 2024 to August 2026, treating a touch of the 20-day high as reaching overhead supply:

53.5 to 46.5. A coin flip. And whether it broke through tells you nothing about what comes next.

The faded group actually did better at D+5. It flips by name too. Netflix runs -4.10% after a breakout and +3.02% after a fade, while Applied Materials runs +1.58% and +3.21%.

Splitting by volume helps a little and then stops.

Volume lifts the odds of clearing that day from 49% to 63%. It does not separate the next five days at all, and in fact runs backwards. Numbers speak up to "does it clear now." They say nothing about why it clears or whether it keeps going. That belongs to sentiment.

And that gate was structurally shut in my account

This is arithmetic, not preference.

For that to clear 2, one of two things must be true.

My account is neither. 22 of 26 positions, 85%, are in a downtrend. A downtrend means price sits below its own supply, which means the gates are stacked immediately overhead, which means the ratio never clears.

Reviewing the 19-name summary from August 17:

8 of 19, 42%, either could not produce a target or had one that locks in a loss. The document says Compass had "all five gates below R 1," Applied had "two of three gates below cost," and SpaceX had "$148 is below cost, so it is an exit."

Hold the gate and there is almost nothing in this account I am allowed to buy. And Circle on August 11 failed that gate at 0.98. Holding the line means no 23.6%.

The strategy review I wrote on August 21 concluded exactly that: "do nothing."

I do not think that was a wrong conclusion. It was a correct conclusion on the wrong horizon. On a weekly basis there genuinely is nothing to buy. The verdict is right. The problem was that I was asking weekly.

The stops were sitting inside daily noise

One more thing surfaced the same week. 13 of 31 positions touched their stop during the week and 10 of them closed back above it by Friday. SK Hynix triggered a close-based stop on August 19 and sat 7.4% above that line by August 21.

Calling that a whipsaw tape is the wrong label. The accurate one is the stops were inside the daily range of the instrument. Which brings up sigma.

Sigma is how much a name moves on an ordinary day, the standard deviation of daily returns. A sigma multiple is today's move divided by that ordinary move. The same -5% is a completely different event depending on the name.

AXT down 5% happens with no news at all. Walmart down 5% always has a reason. Writing it as a raw percentage hides that difference entirely, which is why I am switching to sigma.

My positions exceed 1 sigma on 20 to 32% of days, so 1 to 1.6 days out of every five. Put a stop inside 1 sigma and it gets hit every week. 13 of 31 touching and 10 coming back was not luck, it was the design working as built.

The reverse matters too. 1 sigma days happen constantly. Attach a story to something that happened inside 1 sigma and the story is usually wrong. I wrote quite a lot of those stories into my weekly plans.

The information was not missing, it was on a channel I was not watching

This is the turn. My first conclusion was this: "No matter how much an individual researches, the information is out of reach. Expecting to find it in the news is absurd, so short-term trading is basically fortune telling."

Checking after the fact, that diagnosis was only half right.

Adding the channel and the price reaction to the same events makes the contrast sharper.

August 18 is the one that stands out. It is the only one of the four that broke in the general press first, and the stock fell 3.83% that day. By the time something reaches the general press it is no longer the catalyst, or at minimum it is not a leading signal.

The August 7 breakout happened the same day the Senate opened voting procedure on the CLARITY Act. The reasonless breakout had a reason. And on August 14 the White House meeting was already reported, three sessions ahead of the August 19 move.

What matters is that this is an actionable fix, not resignation. Every sector I hold has outlets that cover only that sector, and regulated industries publish their legislative and regulatory calendars. For crypto that is CoinDesk and The Block. Senate scheduling is a public calendar.

The August 7 item in particular explains itself. A motion to proceed is procedure, not an event. Articles cover what happened. Procedure is about which stage you are at and what comes next. For that you read a tracker, not a story.

The rumor was wrong, and it was also late

What I watched live on August 19 was a stream saying the Circle and Coinbase CEOs had entered the White House. Checking it:

The information was wrong and the direction was right. Circle closed +9.56% that day. I am recording that as luck. Logging a case where an unverified rumor happened to point the right way as skill guarantees being wrong in the other direction next time. The August 7 buy, by contrast, rested on verifiable signals in price and volume. The repeatable part of this trade is the breakout accumulation, not the rumor.

The roles do differ, to be fair. The August 14 article told me what was scheduled. The August 19 stream told me it was happening right now.

The sequence holds up in the data

Another observation was that Circle moved first and Coinbase and Robinhood followed.

Indexed from Circle's August 7 breakout it is sharper still.

Up to the news only Circle was positive. That said, I still do not know why Circle led. The likely explanation is that a stablecoin issuer is the direct subject of the legislation, but a single instance cannot establish that. The sequence is a fact and the cause is a hypothesis. For what it is worth, ranked by final return Coinbase wins at +25.61%, which is a different question from who moved first.

Buying after the news still left half on the table

This measurement is reassuring. Buying the August 19 close at $78.59 still returned +11.95% through August 21. 52% of the total 22.88% move sat after the news.

I started accumulating well before that, around August 7, so entering post-news is a benchmark for the fallback rather than the path I took. It matters anyway, and the reason shows up later. Even with no leading information, securing the horizon captures a large share of the move.

On a long horizon information is cumulative, not asymmetric

This is the core of where I landed. Short-term, the individual is at an information disadvantage, and this episode confirmed it. But stretch the horizon and the nature of information itself changes.

How much would buying on the news have cost

First this question. My hypothesis was that buying on current news or current earnings means buying after most of the move.

Overwhelmingly confirmed. By the day AI became news, NVIDIA was already up 75.5x. Buying that day and holding to now still only gets 5.7x.

Missing it is not the same as making nothing, though. Even post-news you get 5.7x and 9.4x. Same structure as Circle leaving 52% on the table.

Steady accumulation really does produce a ten-bagger

Assuming $1,000 into the first trading day of every month from January 2015 through August 2026, 140 contributions totaling $140,000:

Monthly contributions alone produce 67x on NVIDIA and 10.3x on Tesla. Lump sum maximizes the multiple, since in a rising asset contributions keep lifting your average cost. But that is not DCA losing, it is a different game. Lump sum asks what if you had $140,000 in 2015. Contributions ask what if you earned $1,000 a month. The second one is reality.

And here survivorship bias has to be addressed

NVIDIA and Tesla are names we picked knowing the answer. Run the same strategy on other names that looked like the future in 2015:

Plug Power took 11 years of steady contributions and returned 83% of principal. Contributions did not prevent the loss. Intel was the semiconductor blue chip in 2015 and 11 years of contributions returned 2.38x, below the index at 3.37x.

So buying the future steadily is not a horizon strategy, it is a selection strategy. The horizon amplifies the gain on a correct pick and the loss on a wrong one. Which turns the question into: could I have told NVIDIA from Intel in 2015?

The answer is whether they delivered on guidance

Here is how I framed it. The reason to watch guidance over earnings is that even if guidance is a lie, whether they make it real is what remains.

Laying the two companies side by side makes them different companies. One thing to flag before the table: the two made different kinds of promise. NVIDIA's promise was a number called every quarter. Intel's decisive promise was a date on a process roadmap. So the same question, did they deliver, comes with different forms of evidence.

Intel carries one more consequence. After announcing the 7nm delay in 2020 it faced a securities class action. And 11 years of contributions split like this:

NVIDIA cleared its own numbers 17 quarters running. Intel pushed its own date back five years.

The gap in this comparison is worth stating plainly. I measured NVIDIA on quarterly numbers and Intel on roadmap dates, and I did not measure the reverse. I never checked how well Intel hit quarterly guidance, or how well NVIDIA held product roadmap dates. So this table is not two companies measured with one ruler. It is the single most decisive promise pulled from each. Measuring them with one ruler means filling all four cells, and I have not done that.

Asked whether I could have known in 2015, the accurate answer is no, but by 2018 yes.

One tells you nothing. Three tells you. This signal only becomes a signal by accumulating. And only someone on a long horizon can observe it. On a weekly basis all you see is one dot, this quarter's beat or miss.

On a long horizon information is cumulative rather than asymmetric, and cumulative information is something an individual can count. But you have to actually count. If you do not, a long horizon is just holding, and the result of that is Plug Power at 0.83x.

The limits of this signal, stated up front. First, companies have an incentive to guide low, so 17 straight beats could be skill or expectation management, which you judge by beat size. NVIDIA averaged +5.3%, which is large for pure management. Second, beating guidance guarantees nothing about the future. Intel was the byword for on-time process development before 10nm. Track records break. Third, my current account does not record this history anywhere.

Which is why you have to use the thing

A track record tells you they delivered up to now. It cannot tell you whether they deliver on this one. The only tool an individual has for that gap is using the product yourself.

This test is what separated the four names above.

Intel is the decisive case. AMD's first-generation Ryzen arrived in March 2017 with more cores at the same price while Intel stayed on 14nm. Anyone using a CPU could tell. And that timing overlaps with the track record answer, since the third and fourth 10nm misses land around 2018.

Two independent signals turned on between 2017 and 2018, and both were accessible to an individual. One was filings and roadmaps. The other was my own PC.

Plug Power gives you neither. A consumer cannot use a hydrogen fuel cell. What you cannot use gets judged on narrative alone, and 11 years of contributions returned 83% of principal.

This is exactly the spot Peter Lynch pointed at. The one place an individual beats an institution is not information volume or speed, it is observation as a consumer. Institutions pay for channel checks. Individuals just use the product. The people who felt Intel stalling when Ryzen launched were not analysts, they were people building PCs.

So I built a filter

The rule I set is this. Invest in what I can use or experience indirectly. A company that is too far out and running purely on sentiment, like Plug Power, should have been screened out.

Applied to the 32 positions I currently hold:

Three names trip the filter: AXT, ServiceNow and Compass. All three are US listings. Korea has zero in the "neither" bucket, since I use Samsung, Hyundai and Saempyo directly and the rest are domestic industries I can feel indirectly. The "neither" bucket only appears in US holdings, where remote, B2B and infrastructure names cluster.

And it may not be a coincidence. Those three overlap with the names flagged as problems in the August 17 review.

All three either could not produce a target or were being judged for survival. The names I cannot use turned out to be the same set as the names the numbers cannot answer. The reason is guessable. If you cannot use it, you cannot verify the reason to rise yourself, which leaves only numbers like supply and reward-to-risk. When those do not clear either, you are holding zero evidence.

But a future bet is not automatically a screen-out

Here is a counterexample against myself. I bought BOTZ on the future of physical AI, and I expect it to take a while. So how is that different from Plug Power?

The difference is not whether you bought the future but whether you knew the horizon when you bought it. Plug Power's failure was not buying hydrogen, it was believing it was arriving soon. Eleven years of contributions later it still has not.

Step three is the key. Without a written horizon, in a year or two it turns into "why isn't this going up," and at that point you either add or cut. Both contradict the original premise.

Indirect has sub-grades

I bought Rocket Lab, SpaceX and Cardinal Infrastructure after watching what they were doing through media, and Cardinal after researching the sector and its results. But the evidence behind those three differs in kind.

Activity observation is close to direct use. A rocket launch happens publicly whether it succeeds or fails, and you do not wait for a company announcement. Present-tense verification holds.

Earnings research is different in nature. It is essential for understanding an industry, but financials are the past. And this time I ran straight into that limit.

Litigation and share offerings are not on an income statement. Earnings screened the industry and the financials, but governance and financing risk are not things earnings measure. That is not a research failure, it is outside what the tool measures. The channel that measures it is SEC 8-K and S-1 filings, and both the investigations and the confirmed offering would have surfaced there first.

If you cannot use it and earnings do not prove it, it goes

The rule is simple. If you can use it or feel it indirectly, keep it. If not, check whether earnings prove it. If not, cut it. Applied to the three "neither" names:

All three pass. No name gets cut by this rule.

ServiceNow's price reaction is worth noting separately. It took -6.47% on the day of the July 22 print.

A textbook case of being wrong if you only look at D+1.

And I got this verdict wrong once

This is embarrassing but it needs to become a rule. The August 17 ServiceNow document had several red flags in it, and I read them as "earnings do not prove it." Taken one at a time, they were all something else.

All three were "I did not check." Not knowing and being bad are different things. Convert a failure to verify into a negative verdict and any name you failed to gather data on becomes an automatic sell. So the rule is: when you see a red flag, first separate whether it means unverified or bad. If it is unverified, go verify instead of judging.

One last thing. This rule does not mean earnings make a name safe. Cardinal is the counterexample: proven by earnings, then three lawsuits and a 4.6 million share offering. Earnings are a cutting criterion, not absolution.

The filter's own limits, stated. First, "I can use it" does not mean "good stock." I use Netflix directly and it was up 1.83% this week, while I cannot use AXT and it was up 21.0% in the first week of August. Second, the indirect grade is arbitrary. Calling Applied Materials indirectly felt through Hynix is loose. Strictly, that is reading a capex announcement, not experiencing anything. Third, the filter's value is in exclusion. It is not a tool for picking good names, it is a tool for removing names where I have no basis to judge. Screening out Plug Power is its purpose.

How well did instinct match the measurements

Before the conclusion, a scorecard. This post started from observations I jotted down on instinct right after the trade. Here is what happened when I checked each one against data. The wrong ones and the half-right ones are included as-is.

Of 19, 15 fully confirmed, 2 partial or half, and 2 where my own rule tripped my own account. Not a bad hit rate, but the misses matter more.

Number 2 is a case of my memory disagreeing with the actual fill date. I wrote that volume was low on the day I bought, and that day was 1.21x average. The low volume belonged to the box before it, and I transferred an impression of a stretch onto the memory of a specific day.

Number 4 is what changed this entire post. If I had not checked that "an individual cannot get the information" is only half true, the conclusion would have been resignation. Confirming that the information existed and was accessible but sat on a channel I was not watching turned changing channels into an actionable fix.

Number 16 is my own rule screening my own account. I wrote that I only buy what I can use, and I was holding three names outside that.

And two readings that must be kept separate, nailed down here in advance.

So I am shutting the weekly plan down

That is the evidence. The conclusion is five lines.

The eight measurements behind steps three and four, in one place.

The third one is the heaviest. Over a month I won, and I lost it on weekly and daily execution.

I did not lose because information was scarce. I lost because I kept making decisions on a horizon where information is scarce.

Two things that must stay separate

So this conclusion does not wander somewhere it should not.

First, long-term and averaging down are independent variables.

A long horizon works without any additional buying. Circle this week is the example: entered on the breakout, exited in steps, zero averaging down, +22.88%. The reason I rejected averaging down is capital requirements, not horizon. It needs 7.95 million won at -15% and 40.85 million won at -30%, which is 150% of the original cost. There is no ceiling. So the accurate name for this shift is from weekly per-stock planning to event-driven long holding, and it is not a shift to averaging down.

Second, stretching the horizon does not mean holding anything for a long time.

The horizon is an amplifier. It does not change direction. So entry conditions are still required. No reason to rise means no buy. Under an information disadvantage, the practical form of that condition is the four from the top of this post: a sideways box, a prior-high breakout on a green candle, with volume and public reasons both empty.

How the new format runs

What replaces the weekly plan is a weekly event and exposure map. Four rules.

On predicting size but not direction, precisely: size means the absolute move with the sign stripped off. You can say NVIDIA moves about 7% on earnings day. You cannot say up 7%. Knowing the size settles three things.

The second one is what hurt this time. A stop tighter than the expected range gets caught by noise. The third means realized movement exceeded what the market priced, so the options were cheap.

One caveat. The new format deliberately does not set prices, stops or targets inside that document. But this week's stop problem was not the frequency of decisions, it was the placement of the line. Changing the cadence does not automatically fix placement, so where stops go has to be settled separately.

Limits of this post

The cleaner a conclusion looks, the larger the limits section should be.

1. The Circle section is a single instance. One name, one cycle. The sequence from Circle to Coinbase to Robinhood is measured, but the cause is a hypothesis and repeatability is unverified.
2. NVIDIA and Tesla were picked knowing the outcome. Adding Intel and Plug Power as counterexamples corrects the bias only partly, since those two were also chosen after the fact.
3. The August 7 breakout coinciding with the Senate procedure is timing, not proof of causation. Whether the market traded on it that day is unverified.
4. The P&L is computed from fill prices. FX, commissions and taxes are excluded. The fills themselves are confirmed.
5. The contribution math ignores dividends, taxes and FX. It is a simple monthly buy at the first trading day's close.
6. NVIDIA and Intel were not measured with the same ruler. NVIDIA was judged on quarterly guidance delivery and Intel on roadmap dates. I never counted the opposite cells, so this is not the same metric applied evenly to both.
7. The cycle is not over. A Senate vote is scheduled for September 15 and Jackson Hole's theme is financial innovation and payments. This verdict is as of August 22.

Disclaimer

This post organizes one individual's trading records and reasoning process. It does not recommend buying or selling any security. The figures here are calculated from Yahoo Finance daily bars and public sources, and the fill-based P&L is an estimate that excludes FX, commissions and taxes. Past data does not guarantee future returns, and responsibility for investment decisions and their outcomes rests with the investor.

References

• CoinDesk, U.S. Senate opens first stage of crypto Clarity Act voting (2026-08-08)
• CoinDesk, Trump expected to attend White House meeting with crypto CEOs (2026-08-14)
• CoinDesk, Trump pushes Congress to move on Clarity Act during White House crypto event (2026-08-19)
• Bloomberg, Crypto Executives Join Trump in Push for US Digital Asset Legislation (attendee list)
• CryptoPotato, CLARITY Act Stalls in Senate, Pushed Into September (September 15 vote)
• CryptoSlate, Stablecoins are quickly becoming Kevin Warsh's Fed's next policy problem
• Benzinga, Nvidia Earnings Impact Tracker (17 straight revenue beats, average surprise +5.3%)
• Tom's Hardware, Intel's 10nm Is Broken, Delayed Until 2019
• Motley Fool, The Price of Intel Corporation's 10-Nanometer Failure
• Latham & Watkins, US Crypto Policy Tracker
• Own measurements: Yahoo Finance daily bars. NVDA, TSLA, INTC, PLUG and QQQ from January 2015 to August 2026. Contributions are $1,000 at the first trading day's close, 140 times, excluding dividends, taxes and FX. Circle, Coinbase and Robinhood use confirmed regular-session closes.