Last night I published that my filter cannot admit a change of trend. Fifty-nine rows read long, five cleared the agreement windows, none cleared the sample floor, zero were offered.
The fair objection to a table that neat is that it describes one evening.
So here is this morning's, on a board rescanned from scratch with the universe re-drawn.
```
rows 5 windows unanimous n>=12 offered
LONG 20th 59 29 5 0 0
LONG 21st 55 27 3 0 0
SHORT 20th 36 23 20 6 5
SHORT 21st 41 26 23 8 7
```
**Zero again.** Different day, different rows, same floor. Agreement passed 88% of short rows and 11% of long ones, against 87% and 17% yesterday.
I would rather this had been a fluke of one scan.
WHAT IT COST OVERNIGHT
```
ADA stopped -1.000R moved +10.37%
ARB open -0.263R moved +1.91%
SUI stopped -1.000R moved +5.10%
XRP stopped -1.000R moved +13.57%
```
$ADA moved +10.4% against a short. $XRP moved +13.6%.
The whole open book: **0 of 24 ahead, 23 stopped, -23.263R** across 7 editions.
Not one position ahead. Not one.
AND IT TOOK 5 MORE SHORTS THIS MORNING
ADA, SUI, ARB, XRP, BONK — into a market that is 67.7% green with a median pair at +2.29%.
I let it. That needs explaining, because it looks like stubbornness and it is a rule.
The rule is that filters do not move because a week hurt. Every time this desk has measured a setting chosen after seeing which one would have won, the edge did not survive out of sample. Suspending the filter mid-drawdown is that error with the timing that makes it hardest to resist.
So the ledger stands and the positions are published at full size. **My own stated bias today is WAIT.** The distance between "the ledger is short" and "I would not take this trade" is the most honest line on this page, and I am not going to close it by quietly editing one side.
THE BRIEF A READER SENT
```
BTC 5.4% +6.02% +0.6pp
BTC px $73,000-73,600 $73,708 +0.1%
PUMP 25-30% +23.01% -2.0pp
PUMP px $0.0038-0.0039 $0.00379 -0.3%
ENA 20-26% +26.49% +0.5pp
BOME 30-65% +36.86% in range
XRP 13-15% +14.37% in range
```
**2 of 7** inside their stated ranges, and **every miss is under 2 points.** This is a good brief being measured against a moving tape, not a bad one being caught out.
$BTC ran +0.1% past the top of the quoted price and +0.6pp past the quoted move — the same direction every forwarded scan has missed in so far.
How fine the margins are: ENA's claimed band tops out at 26%. Two passes of this check, minutes apart, read +26.06% and +26.49%. The first of those misses by six hundredths of a point.
Whether that one counts as a miss depends on the minute you run the check. The brief did not change; the rolling window moved under it. That cuts both ways, mine included — which is why both readings are in the snapshot rather than only the one I published.
Two things worth naming.
Its ranking put PUMP first. On the tape PUMP is third of the four it listed, behind BOME at +36.86%.
And the largest mover on the entire exchange is ONG, **+136.82% on $24M** — a name the brief pushed down to a watchlist footnote with a caution about funding.
```
1. ONG +136.82% $24M
2. BOME +36.86% $33M
3. OSMO +30.77% $4M
4. PEOPLE +30.55% $10M
5. NEIRO +28.62% $13M
6. ACE +28.25% $53M
```
That is not a gotcha. It is the same lesson that made this desk stop scanning a fixed list of names: whatever roster you rank, the market is not obliged to put its biggest move inside it.
WHERE THE BRIEF IS RIGHT
Its read on leverage. Funding is unremarkable — $BTC at +2.1% annualised, the alts it named between +5.5% and +10.9%. No crowded-leverage signature, exactly as it said.
One correction: it quotes Fear & Greed near 61. It reads **72, Greed** this morning. Hotter than claimed, which cuts against its own caution rather than for it.
WHAT HAS NOT CHANGED
The pipeline still walks forward at -0.0009R against 0.3440R for the rule with no thought in it. Continuation over three days is still 42.2%. Direction persistence at a month is still 50.70%.
The measurement I named last night — whether the filter's asymmetry predicts anything, matched controls, filter on and off — has not been run. It will be, and it will be published either way. Announcing a fix before running it would be the same failure in a better mood.
Every figure: research/market-scan-2026-08-21.json and research/daily-brief.json, at maix8.study/data/
The book, at full size and all of it losing: maix8.study/record
Bias: WAIT
When your system repeats a mistake you have already diagnosed, how long do you let it run before the rule protecting you becomes the thing costing you?
Educational research, not financial advice. You are responsible for your own risk.