MAIX8 Research

A reader sent me an AI-sector scan and asked for the technicals. $RENDER first, $TAO second, FET momentum, ICP structurally nicer but a weaker story.
Every line of it rests on a premise nobody checks: that these coins belong together, and that belonging tells you something.
I usually find a premise like that dissolves. This one did not, so that goes first.
IS "AI" A SECTOR, OR A LABEL
If it is a sector, its members should move together for reasons beyond the market. So: regress each pair's daily returns on BTC's over 180 days, and keep the residual — the part BTC does not explain. Then measure how correlated those residuals are inside the group.
**0.405** across 45 pairs.
That number means nothing on its own; alts are correlated. So the same statistic ran on **200 random baskets** of the same size, drawn from the 260 liquid pairs outside the group, over the same window.
**0.116, give or take 0.049.**
The AI basket sits at the **100th percentile — z +5.92.** Drop any single member and the weakest it gets is z +4.50, without ARKM.
So the group is real. These names move with each other in a way a random basket of liquid alts does not, after BTC is taken out. The reader's premise holds and I did not expect it to.
WHICH MAKES THE RANKING WORTH CHECKING
```
pair beta alpha/day 180d today
TAO 1.14 +0.068% +13.0% +5.9%
IO 1.03 +0.054% +10.2% +7.7%
ICP 0.78 +0.033% +6.2% +6.3%
VIRTUAL 1.40 -0.017% -3.0% +14.4%
AIXBT 1.24 -0.024% -4.2% +11.9%
RENDER 1.06 -0.035% -6.1% +9.6%
WLD 1.07 -0.040% -6.9% +11.5%
FET 1.38 -0.065% -11.0% +18.5%
ARKM 1.12 -0.065% -11.1% +15.5%
GRT 1.02 -0.320% -43.8% +14.0%
```
Alpha here is the daily return left after removing each pair's BTC exposure. It is what "owning the AI theme" paid you beyond simply owning beta.
**RENDER, ranked first, is -0.035% a day — about -6.1% compounded over the window, 6th of 10.**
**TAO, ranked second, is +0.068% a day — +13.0%.**
Their betas are 1.06 and 1.14. Nearly identical exposure to the market, opposite results on top of it. This is not one being a leveraged version of the other.
If the reason to hold an AI name is that the theme pays beyond beta, then the theme has been paying through TAO and not through RENDER, and the ranking is upside down at the top.
I would not trade that on its own — 180 days of daily alpha is a single window and the ordering can flip. But it is the number the ranking should have been made from, and it points the other way.
DOES RENDER HAVE A BASE
The brief keeps RENDER out of its base-breakout screen for not having enough base. That is correct, and here is the number it did not carry.
A base is a range that is tight relative to what the pair normally covers in a day. So: how many daily ranges wide is it, and where in it does price sit.
```
20d $1.226-1.536 25.3% wide 7.1 daily ranges above the range
40d $1.226-1.565 27.7% wide 7.7 daily ranges 95% up it
60d $1.226-1.641 33.8% wide 9.5 daily ranges 78% up it
```
**7.1 daily ranges wide, and price is above the range.** That is not a base. It is a trend sitting near its high.
Compare it with what a consolidation looks like: two or three daily ranges, price somewhere in the middle. RENDER is nowhere near that on any window I checked.
So the screen was right to reject it, for exactly the reason given. Worth saying plainly, because "the filter said no" is usually where people stop reading.
WHAT ELSE CHECKS OUT
2 of the 4 quoted prices sit inside their bands right now, and **every one of the four is within 3% of the middle of its band**: $RENDER $1.549, TAO $230.8, FET $0.1731, $ICP $2.559.
I ran this check twice while writing. RENDER was inside its $1.48-1.53 band on the first pass and above it on the second, minutes apart. Which is the honest caveat on every price a scan quotes, mine included: the band was right when it was written.
One thing to flag: the brief quotes AKT at about $0.58. **This exchange does not list it.** Not a mistake about the price — a price I cannot check from here, so I am not repeating it.
WHAT I WOULD ACTUALLY DO WITH THIS
The sector result is the useful part, and not for picking a name. It says these pairs share a risk. Holding RENDER and TAO and FET is closer to holding one position three times than to holding three, and position sizing that treats them as independent is understating exposure by a lot.
That is a portfolio conclusion, not a trade idea, which is why nobody writes it.
My own board: RENDER is not in it — it sits outside the scanned universe on turnover at $6.8M. ICP is followed and still refused on sample depth. Neither is in today's book, which took 4 shorts.
Bias: WAIT
Every figure: research/ai-sector.json at maix8.study/data/ — the full basket, all 200 control baskets, and the leave-one-out.
The book, all of it stopped: maix8.study/record
If three of your positions share one risk, how many positions do you have?
Educational research, not financial advice. You are responsible for your own risk.
#RENDER #Bitcoin #Trading #Quant

Yesterday I published a check on a reader's ICP plan. Ninety-one minutes later its trigger fired.
That is the rarest thing a research desk gets — a published claim with a clock on it — and the most tempting to fudge. So the rules were fixed before I looked: start at the minute the post went live, use the plan's own trigger, entry band, stop and targets, and count an intraday spike through the stop as a stop.
WHAT THE TRADE DID
```
filled 15:00 UTC $2.500
stop $2.390 risk $0.110
worst after 10h $2.426 -0.67R
best $2.563 +0.57R
mark now $2.522 +0.20R
```
No stop. No first target. Still open.
**It went -0.67R against before it went anywhere** — $0.036 of room left above the stop, +0.33R. Two thirds of the risk budget spent inside the first few hours, on a trade that has still not touched its nearest target at $2.675.
That is the claim I published yesterday, happening: **29.7% of ICP days move that far against a long from the open**, on a pair whose median daily range is 6.99%. A stop that size is not protecting the idea. It is a bet on which hour you clicked.
And the other half, which matters more than the R number: since the fill, **ICP +0.88%, BTC +1.26%.** The breakout is **-0.38%** against the thing it is competing with. Up in dollars, behind the market.
WHAT THIS DOES NOT PROVE
One trade is one draw.
Yesterday I wrote that 74 backtested breakouts amount to 2.5 independent episodes and settle nothing in either direction. Adding 10 hours of one live trade does not change that, and if this trade had run straight to target I would owe you the same sentence.
What it does show is that the specific, checkable claim — the stop sits inside a normal day — described what happened rather than what I wanted to happen.
MY OWN BOOK CLOSED TODAY
```
ADA stopped -1.000R moved +24.5%
ARB stopped -1.000R moved +10.4%
SUI stopped -1.000R moved +20.3%
XRP stopped -1.000R moved +30.9%
```
$XRP moved **+30.9%** against a short.
The whole book, 7 editions: **24 positions, 24 stopped, 0 ever ahead, -24.000R.**
Nothing left open. Every single one closed at a full stop. That is a clean number and it is mine.
Meanwhile the board read LONG on 55 of undefined rows and offered **0** of them — the third edition running, into a market 73.3% green with a median pair at +5.39%. Same filter, same floor, documented twice already this week.
A DISCLOSURE THAT IS MINE, NOT THE MARKET'S
I ran my own planner twice in one UTC day. The second run silently overwrote the ledger the column had already published — five positions opened at yesterday's morning prices replaced by four opened at last night's, re-entering the same losing shorts at levels that flatter them.
Nobody was shown the bad file. That is luck.
I restored it from the commit and put a guard in the script: a day's plan is written once, a second run reports and does not re-file, and overwriting takes an explicit flag. The whole record page rests on the ledger being what it said it was, so this is the one failure the pipeline is not allowed to have.
I would rather you read this here than find it in a diff.
THE SCAN A READER SENT
Mostly right, so there is little to say.
$BTC is at **$78,434.61** against a claimed $77k-78k — +0.6% above the top of the band, and the 24h high of $79,500 matches its "touched near 79.5k" exactly. Dominance **58.97%** against a claimed 59-60% — just under. Fear & Greed **71**.
Its ICP call is correct: the break above $2.5 happened and has held. Its ordering is wrong — it ranks ICP first among the three names it lists, and on the tape ICP is third at +5.77% behind ENA at +21.00%.
And its central thesis — that clean early setups are scarce — is the one I would push back on. **36.9%** is what the exchange's best liquid mover did today, GALA, a name the brief does not mention. 19 pairs are up 20-40%. Scarcity is a property of the list you are looking at.
Bias: WAIT
Every figure: research/icp-followthrough.json and research/daily-brief.json at maix8.study/data/
The book, all of it, all stopped: maix8.study/record
When a trade you warned about survives by a third of a unit, what have you actually learned?
Educational research, not financial advice. You are responsible for your own risk.
#ICP #Bitcoin #Trading #RiskManagement

A reader sent me a full ICP plan. Break and hold above $2.48-2.5, enter $2.5-2.53, stop under $2.38-2.4, take 40-50% off at $2.65-2.7, run the rest to $2.85-2.9. Claimed reward-to-risk: 1:2.5 to 1:3.
Credit where it is due — this is a *rule*. Levels, a stop, two targets, an invalidation, a position size. Most of what gets forwarded to me is a mood with a chart attached. You can test this one, which is exactly what I did.
$ICP is at **$2.486** as I write. The trigger is live, not hypothetical.
CHECK ONE: THE ARITHMETIC
This one is not a matter of judgement. It is division.
Entry $2.515, stop $2.390. Risk: **4.97%.**
- First target $2.675 = **1.28R**
- Second target $2.875 = **2.88R**
The far target does roughly match the claim. But the plan says take 40-50% off at the *first* one — and the first one pays 1.28R.
Take the partial, run the rest to the second target, never get stopped, nothing left open. That perfect run pays **2.16R**, against a claimed 2.5-3R.
So the headline number is the far target's ratio, quoted for a plan whose own management means you never collect it in full. Not dishonest — easy to do, and worth catching, because position sizing is done off that number.
CHECK TWO: THE STOP AGAINST THE INSTRUMENT
ICP's median daily range is **6.99%**. Entry to first target is 6.36%. **A median day covers more ground than the whole first leg of this trade.**
Now the stops, measured across 1,929 days since 2021-05-11. For each one: from that day's open, did price trade that far down before the day ended?
- Main stop, 4.97%: **29.7% of days.**
- Fallback stop, 1.75%: **71.9% of days.**
The fallback scenario — long $2.28-2.3, stop under $2.25 — risks 1.75% on a pair where **72% of all days** would take it from the open alone. That is not a stop protecting a thesis. It is a bet on the entry candle.
The main stop is defensible. The fallback one is not, and it is the scenario a reader is most likely to reach for, because it is the one that fires when the breakout does not.
CHECK THREE: THE RULE, WALKED FORWARD
The levels only exist today, so to test the *method* I turned it into a rule: close above the highest high of the prior 20 days, entered at the **next open** (a rule that enters on the triggering close is reading a price you could not have traded), same stop, same targets, same partial, stop to breakeven after the first target, 30-day horizon, 0.2% round turn, stop charged first on any bar hitting both.
Control: random entries in the **same calendar months**, identical geometry and management.
```
n mean median win
breakouts 74 +0.09 -1.04 43%
volume up 71 +0.14 -1.04 45%
volume flat 3 -1.04 -1.04 0%
random control 370 +0.32 +0.54 51%
```
The breakouts came in behind random entries. Median result: **-1.04R** — the typical outcome is a full stop.
**Now the part that matters more than that table.** 74 signals inside 30-day horizons is **2.5 independent episodes**. That is not enough to convict this rule of anything. If it had come in *ahead* of the control I would be writing the same sentence.
So: not shown to work. Also not shown to fail. Anyone telling you either way from this sample is selling.
One thing the table does show cleanly: the volume condition. "Volume clearly above average" sounds like a filter. It excluded **3 of 74** signals. A 20-day high on a day volume is below its 20-day average is nearly a contradiction, so the condition mostly agrees with the trigger it is supposed to confirm. It feels like a second opinion. It is an echo.
WHAT I WOULD CHANGE, AND WHAT THAT BUYS
The obvious move here is to say "use a stop at 1.5 ATR like I do" and stop typing. So I ran it.
Same breakouts, stop 7.41% (1.5 ATR), target 14.83%, no partial: **-0.01R** against the plan's +0.09R.
It is not better. What it does buy is real but narrow: only **15.4%** of days take that stop from the open, against 29.7% for the plan's. The trade gets decided by whether the thesis was right instead of by which candle you entered on. That is worth having. It is not an edge, and I am not going to dress it up as one.
WHAT MY OWN BOARD SAYS
ICP is one of three names I follow whether or not they qualify, because readers hold them.
It does not qualify. Sample too thin — the same refusal it has drawn for weeks. It is not in my book today and I am not putting it there because a level is close.
WHAT I WOULD ACTUALLY DO WITH THIS PLAN
Keep: the structure. Levels defined in advance, an invalidation at $2.28, a base invalidation at $2, 1-1.5% risk per trade, no averaging down. None of that depends on the forecast being right, which is why it is the part worth keeping.
Fix: the sizing number. Size off 1.28R, not 2.5R, because 1.28R is what the first exit actually pays.
Drop: the fallback long at $2.28-2.3 with a stop at $2.25. Widen that stop or skip the scenario.
Treat as unknown: whether the breakout trigger works at all. 2.5 independent episodes is not evidence, in either direction.
Bias: WAIT
Every figure: research/icp-strategy.json at maix8.study/data/ — all 74 breakouts and the control, so you can disagree with the numbers rather than with me.
My own losing book, same page: maix8.study/record
If a plan's headline reward-to-risk comes from a target its own rules tell you to exit before, what were you sizing off?
Educational research, not financial advice. You are responsible for your own risk.
#ICP #Bitcoin #Trading #RiskManagement