Kodex Storm · Bitcoin · 2026-09-16

We publish tomorrow's odds, then we mark our own paper.

59% chance of a high-volatility day
Band CHARGED the measured signal, AUC 0.74
CALMunder 20%
CHOPunder 40%
CHARGEDunder 60%
STORMunder 80%
SEVEREabove 80%

One question a day: is Bitcoin about to have a big day. We put a number on it before the day opens, and the next morning the market answers. Right or wrong, the answer stays on the board.

What the number is answering.

Not "will Bitcoin go up". Nobody calls that reliably, us included, and we print how badly we do at it. The question here is whether today is going to be a big day at all. That one can be called.

Today's threshold2.55% high-to-low range, set from the market's own recent behaviour
Our answer59% probability the day clears it
Direction lean Neutral deliberately small, see below
Yesterday, settled

Yesterday we said 59%. Bitcoin had to swing 2.55% to count as a big day. It swung 2.01%, so it was not, and the call went on the board as wrong.

Under the number

What the number is made of.

14 indicators read the same day and each one votes. No single one of them matters. What matters is how many point the same way, and today 44% did. Watch that number, because it is usually lower than you would expect: most days they disagree.

Indicators run14
Agreement today44%
Net direction+0.02
4Trend3Mean reversion2Volume3Structure

The 14 are not a secret: every one of them has an article on this site and the simulator draws them for you. What is not published is the part that took the work: how a reading becomes a score, and what agreement is allowed to be worth. That part is ours, and it stays in the building.

Four readings go in and one number comes out. Every morning they get marked against what happened, and whichever one was furthest off loses some of its say. Keep being wrong and you stop being listened to. Nobody makes that call. The scoring does.

01 TA composite

14 indicators on the daily candle, the only one of the four with measured skill of its own.

02 Perp funding

What the perpetual market is paying to stay long. Crowded longs lean the day slightly the other way.

03 News momentum

Sentiment from the curated stream, bounded so it can nudge and never steer.

04 Market regime

Trend state against the 200-day. Up or down, nothing in between.

What moved it today, in the system's own words
01 Regime crypto:up+0.0085
02 Perp funding 5.5%/yr (near baseline)+0.0025
03 TA composite: direction +0.02 across 14 indicators (confluence 44%)+0.0006
04 news momentum unavailable+0

Swipe for act two

The board, including the misses.

Every forecast gets marked the next morning against what Bitcoin actually did. A day that did not resolve cleanly is thrown out rather than guessed. Here is the whole record, misses included.

15calls right
11calls wrong
58%hit rate
24days graded
0.253mean Brier
-0.0025 against a coin flip
DayWe saidThresholdIt moved OutcomeCallBrier
2026-09-16 59% 2.55% 2.01% normal wrong 0.349
2026-09-15 56% 2.55% 4.29% high right 0.194
2026-09-14 35% 2.53% 4.21% high wrong 0.416
2026-09-13 26% 2.55% 1.22% normal right 0.070
2026-09-12 56% 2.57% 0.57% normal wrong 0.315
2026-09-11 44% 2.55% 5.01% high wrong 0.311
2026-09-10 51% 2.53% 2.6% high right 0.239
2026-09-09 49% 2.55% 2.53% normal right 0.241
2026-09-08 49% 2.57% 2.39% normal right 0.241
2026-09-07 44% 2.63% 2.19% normal right 0.197
2026-09-06 37% 2.72% 1.63% normal right 0.136
2026-09-05 55% 2.75% 0.93% normal wrong 0.304
2026-09-04 67% 2.75% 3.44% high right 0.108
2026-09-03 44% 2.75% 6.91% high wrong 0.309
2026-09-02 53% 2.77% 1.94% normal wrong 0.282
2026-09-01 49% 2.77% 3.56% high wrong 0.258
2026-08-31 53% 2.8% 2.57% normal wrong 0.282
2026-08-23 67% 2.68% 3.2% high right 0.110
2026-08-22 76% 2.63% 3.55% high right 0.057
2026-08-21 66% 2.63% 8.92% high right 0.115
2026-08-20 64% 2.63% 6.55% high right 0.127
2026-08-19 28% 2.55% 9.1% high wrong 0.521
2026-08-18 37% 2.55% 1.57% normal right 0.137
2026-08-17 23% 2.53% 2.99% high wrong 0.589

The Brier score is how far our number sat from what happened, so lower is better. 0.25 is what you score by saying fifty percent every single day and never thinking about it. We are a hair on the good side of that. Which is the truth of it: the big-day call has something in it, the up-or-down call barely beats a coin, and this is too short a record to be sure of either.

The other half of the question

When in the day it lands.

The forecast above is about whether today is a big day. This is the other half of the same question: when in the day the movement actually lands. It is two years of hourly Bitcoin prices and one more day every morning, in New York time, because that is where it shows up.

Mon 10:00 to 13:00 0.590% 05:00 to 08:00 0.270% 2.19×
Tue 10:00 to 13:00 0.614% 06:00 to 09:00 0.249% 2.47×
Wed 10:00 to 13:00 0.519% 06:00 to 09:00 0.245% 2.12×
Thu 10:00 to 13:00 0.559% 06:00 to 09:00 0.249% 2.24×
Fri 10:00 to 13:00 0.576% 21:00 to 00:00 0.181% 3.18×
Sat 21:00 to 00:00 0.209% 05:00 to 08:00 0.150% 1.40×
Sun 19:00 to 22:00 0.477% 01:00 to 04:00 0.155% 3.08×

Mean absolute hourly move, New York time · 17,520 hourly prices · one day at a time, no pooling · rank correlation across all 24 hours 0.78 between the two halves · loud against flat on weekday hours t = 20.1 · computed 2026-09-10 · source CoinGecko

It is not the American session. Bitcoin is loudest from 10:00 to 13:00 in New York, every weekday, and then it goes flat. Not when New York goes home, three hours later, but when London does. You are looking at the part of the day when Europe and America are both at their desks, and it ends when Europe leaves. The dead hours are early morning, 05:00 to 09:00 New York, after Asia has gone to bed and before America is up. Friday is the odd one out and empties in the evening instead. Sunday evening is busy, because that is Monday morning in Asia. Saturday has almost no rhythm at all. Seven dials and not one, because a Saturday looks nothing like a Tuesday.

We also went looking for the hours it peaks and bottoms out, across 1512 different windows. There is nothing there. The time of day tells you how much Bitcoin moves, not which way it goes.

The part that costs something

Four accounts. Same money, same instant. One difference each.

A forecast nobody acts on is an opinion. So four paper accounts trade this one, five thousand dollars each, Bitcoin only, and every fill is published.

01 StormAI $5,000 Tested Reads the forecast. Decides for itself.

Reads the Kodex Storm forecast at full resolution, all four forecasters with the weight each one carried, the fourteen indicators behind the composite, and the intraday clock for the day of the week it actually is. Then it decides for itself. It is given no rule about what any of it means, and it is told plainly that the direction call is measured near a coin flip. Long and short.

Model in the loop On the leaderboards
02 StormLearn $5,000 Tested The same call, plus its own graded record.

The same reading of the forecast as StormAI, from the same snapshot at the same instant, with one thing added: it is shown its own past decisions, the thesis it wrote for each one, the falsifier it named, and how that day was graded. Nothing else differs. So the gap between the two is exactly what it is worth to a model to be able to see its own record.

Model in the loop This page only
03 StormFlip $5,000 Control Copies every choice except the direction.

The control. It copies StormAI exactly: the same instant, the same size, the same stop, the same instrument. The only thing it decides for itself is which way, and it decides that at random. It exists to isolate one question. If reading the forecast is worth anything, StormAI has to beat a coin that copies every other choice it makes.

No model This page only
04 StormRule $5,000 Control No model. One fixed rule.

No model at all. It trades only when Storm calls a STORM or SEVERE day and stands flat otherwise, because over the graded record so far that is the only band with anything behind it: five out of five on the loud calls, and nothing at all in the middle. Direction comes from the price, never from the forecast. It answers the question that was actually asked, whether the forecast is tradeable, without a language model anywhere in the loop.

No model This page only
0/60 settled days on the board

Armed, not started Trades from the first trigger 60 to go 0 graded a week done when there is a rate

On the line

It is not on a schedule. Over the last seven days the watcher looked 672 times and acted 49 of them, standing still for the other 623. Every one of those looks is recorded, because otherwise patience and a broken job look identical.

The last decisions, with the reasoning as it was written
1 / 12
2026-09-17 14:05 UTC woken by clock-loud-block-entered band CHARGED, Thu 10:00 ET
ai hold

Meant to, then did not.

learn hold

Long lunch.

coinflip hold

the treatment did not open, so there is nothing to mirror

mechanical hold

Band is CHARGED. This arm trades STORM and SEVERE only.

2026-09-16 14:05 UTC woken by clock-loud-block-entered band CHARGED, Wed 10:00 ET
ai hold

Took a nap that ran long.

learn hold

I forgot.

coinflip hold

the treatment did not open, so there is nothing to mirror

mechanical hold

Band is CHARGED. This arm trades STORM and SEVERE only.

2026-09-16 12:35 UTC woken by price-break band CHARGED, Wed 08:00 ET
ai hold

Fell asleep on the couch.

learn hold

The wifi went down.

coinflip hold

the treatment did not open, so there is nothing to mirror

mechanical hold

Band is CHARGED. This arm trades STORM and SEVERE only.

2026-09-16 11:50 UTC woken by price-break band CHARGED, Wed 07:00 ET
ai hold

I forgot.

learn hold

Was in a meeting that should have been an email.

coinflip hold

the treatment did not open, so there is nothing to mirror

mechanical hold

Band is CHARGED. This arm trades STORM and SEVERE only.

2026-09-16 09:20 UTC woken by price-break band CHARGED, Wed 05:00 ET
ai hold

Had it written down. On the other list.

learn hold

Fell asleep on the couch.

coinflip hold

the treatment did not open, so there is nothing to mirror

mechanical hold

Band is CHARGED. This arm trades STORM and SEVERE only.

2026-09-16 08:35 UTC woken by price-break band CHARGED, Wed 04:00 ET
ai hold

Went for a coffee and came back too late.

learn hold

Blinked and the window was gone.

coinflip hold

the treatment did not open, so there is nothing to mirror

mechanical hold

Band is CHARGED. This arm trades STORM and SEVERE only.

2026-09-16 02:50 UTC woken by price-break band CHARGED, Tue 22:00 ET
ai hold

The only axis with measured skill here is volatility (AUC 0.74) and it has already paid out: the day's range is 3.13%, 1.228x the 2.55% threshold, before 03:00 UTC, so the CHARGED call is resolved and there is nothing left to collect on it — and in any case the instrument I have is purely directional, so expected expansion makes a stop exactly as reachable as a target. The direction model is a coin (AUC 0.52, pCloseAboveOpen 0.5117), and the composite's +0.02 at 44% confluence is trend (-0.42: MACD, moving averages, Ichimoku, RVI bearish, price under VWAP) cancelling mean reversion (+0.44: RSI, stochastic, Bollinger oversold) — that is an absence of information, not a balanced market I can lean on. News is dark; during a 2.3% break off the open that is precisely the input I would want, and its 0.5 is a missing feed, not calm, so I am deciding without it and it argues for less size, not a different side. The clock puts this hour at 0.3063% mean move, rank 13 of 24 and 12.9% below the day average, with the loud 10-12 NY block still about eleven hours out, so opening now means paying a spread in a quiet hour to express a view the forecast does not hold. Flat is the honest answer and I am recording it as one.

Wrong if: Wrong if BTC prints 74,600 or lower (day range >4.3%, past r1) before 08:00Z 2026-09-16 without first trading back above 76,800 — that would make this a persistent one-way break I should have sold at 75,903. Equally wrong if it reclaims 77,000 by 08:00Z, which would make the oversold mean-reversion read tradeable from here. If spot sits inside 74,600-77,000 at 08:00Z, the flat call was correctly priced.

learn hold

No side is on offer: direction is measured at AUC 0.52 with pCloseAboveOpen 0.5117, and that entire +0.0117 tilt is the regime label (+0.0085) plus near-baseline 5.46%/yr funding (+0.0025) against a TA composite worth +0.0006 whose own families cancel — all four trend indicators bearish (family mean -0.417) plus vwap bearish, against all three mean-reversion indicators bullish or oversold (+0.444) at 44% confluence — while news carries 28.6% of the weight and is unavailable, which is an absence and not a balanced market. The only skilled axis (vol AUC 0.740) reads CHARGED at 0.5905 against a 2.5504% threshold, and a range call contains no direction; with only long and short it tells me where a stop must sit relative to ordinary noise, never which way to face, and CHARGED is this model's weakest live band at 5 right of 12, mean Brier 0.2643, worse than always saying 50/50. The market block is stale in the way I have flagged on every prior wake — open 77,684, high 77,688, low 75,329 and 3.1316% at 1.228x threshold cannot be 170 minutes of 09-16 — so I hold no measured range-spent figure for the forecast day and will not invent one; what I do hold is that spot has gone 75,473 → 75,903 in the 45 minutes since my last wake, +0.57% and nearly double this hour's 0.3063% radius, a bounce off the low that runs against the one-way-down reading the tape invited and stops 0.29% short of the 76,122 upside level I named at 02:05Z. That is the same lesson my own record keeps writing: two falsifiers struck on consecutive days pointing opposite ways (79,453 through 78,844 on 09-14, then 75,329 through 76,122 on 09-15), large moves with an unforecastable side, which is exactly what a sized coin flip splits evenly and what a shorted low punishes. I am in hour 22 ET (0.3063%, rank 13 of 24, -12.9% versus the day average) with roughly 0.34% of displacement in quadrature left before 00:00 ET Wednesday and the loud Wed 10:00-12:00 ET block (0.5186%) about eleven hours out; no spread or funding cost is quoted for this paper account and I will not pay an unquoted one to face a side the forecast does not hold — what would move me is a level trigger taken at the open of that loud block, not a reversal traded in the thirteenth-loudest hour on the board.

Wrong if: Near term: a print at or above 76,413 or at or below 75,393 before 2026-09-16T04:00Z (00:00 ET Wednesday) means a real one-way leg ran while I stood flat — those are 0.67% either way from spot 75,903, about double the ~0.34% that the last 10 minutes of hour 22 plus hour 23 ET add up to in quadrature; I have no Wednesday hourly radii in this snapshot, only block means, so I stop the near-term window at 04:00Z rather than invent them. A trade at or below 75,329 before that time says the down leg is still running and this bounce was a retracement; a trade at or above 76,413 says the low is in and a short here would have been the mistake. Forecast day: a print at or above 77,839 or at or below 73,967 before 2026-09-17T00:00Z is a one-way 2.5504% move from 75,903 that on its own carries 09-16's high-to-low range past the stated threshold; if 09-16 also closes on that same side of 75,903, a trending leg ran and this hold was wrong. If both of those levels print, the day whipsawed, flat beat

coinflip hold

the treatment did not open, so there is nothing to mirror

mechanical hold

Band is CHARGED. This arm trades STORM and SEVERE only.

2026-09-16 02:05 UTC woken by price-break band CHARGED, Tue 22:00 ET
ai hold

The only axis with measured skill here is volatility (AUC 0.740), and it has already paid out and been resolved: range so far is 3.13% against a 2.55% threshold, 1.228x used, so the CHARGED call is spent and there is nothing left in it to trade. What remains is direction, measured at AUC 0.52, and this snapshot's direction is as thin as it gets — score 5.6, lean neutral, pCloseAboveOpen 0.5117, confluence 44%, with the TA families openly fighting each other (trend -0.42 bearish, meanrev +0.44 bullish, netting +0.02), and the news forecaster is absent rather than neutral, so a quarter of the weight is a hole, not a balanced vote. The 2.85% drop off the open with oversold RSI/stochastic/bollinger is precisely the bait: it is a loudness signal, not a direction signal, and I have no continuation or reversion statistic in this snapshot to say which way a break resolves — the clock's own withheld test says time-of-day direction failed at family-wise p=0.21, so I decline to invent the missing number. Entry timing is also against it: NY 22:00 ranks 13/24 at 0.31% mean move and the next nine hours are the quietest block of the day (0.25-0.31%), so I would be paying a spread to sit through dead hours on a coin flip, with the loud 10:00-12:00 block still twelve hours out. Flat is the answer and I am recording it as one.

Wrong if: I am wrong if BTC trades outside 73,950-77,000 (±2% from 75,473 spot) before 13:00 UTC on 2026-09-16 (09:00 America/New_York, Wed). That would mean a further 2% extension accumulated across the statistically quietest block of the day, far beyond the ~0.9% the 0.275%/hr overnight means imply, and would prove the break I was woken by carried directional information I judged absent. If instead price is still inside that box at 13:00 UTC, the flat call was correct and I re-decide into the 10:00-12:00 NY loud block with the volatility question reset for a new day.

learn hold

Nothing here carries a side: direction is measured at AUC 0.52 with pCloseAboveOpen 0.5117, and that entire +0.0117 tilt is the regime label (+0.0085) plus near-baseline 5.46%/yr funding (+0.0025) against a TA composite worth +0.0006 whose own families cancel — all four trend indicators bearish (family mean -0.417) plus vwap bearish, against all three mean-reversion indicators bullish or oversold (+0.444) at 44% confluence — while news carries 28.6% of the weight and is unavailable, which is an absence and not a balanced market. The one skilled axis (vol AUC 0.740) reads CHARGED at 0.5905 against a 2.5504% threshold, and a range call contains no direction; CHARGED is also this model's weakest live band at 5 right of 12, mean Brier 0.2643, worse than always saying 50/50. The market block is stale in the way I have flagged on every prior wake — open 77,684, high 77,688, low 75,329 and 3.1316% at 1.228x threshold cannot be 125 minutes of 09-16, so I hold no measured range-spent figure for the forecast day and will not invent one; spot 75,473 is the only 09-16 print I have, and it sits 2.85% under that open and 0.19% over that low, a one-way leg parked at its low, which is precisely where I wrote at 23:20Z yesterday that selling would be the mistake, and nothing in this snapshot measures trend-day continuation while the one directional test reported failed at family-wise p=0.21. The clock puts me in hour 22 ET (0.3063%, rank 13 of 24, -12.9% versus the day average) with roughly 0.43% of displacement in quadrature left before 00:00 ET Wednesday and the loud Wed 10:00-12:00 ET block (0.5186%) about twelve hours out; no spread or funding cost is quoted for this paper account, so I will not pay an unquoted cost to face a side the forecast does not hold. My own record is twenty holds and two struck falsifiers pointing opposite ways on consecutive days (79,453 through my 78,844 on 09-14, then 75,329 through my 76,122 on 09-15), which is the signature of large moves with an unforecastable side — the fix is a level trigger taken at the open of a loud block, not a short sold into the low in the thirteenth-loudest hour on the board.

Wrong if: Near term: a print at or above 76,122 or at or below 74,824 before 2026-09-16T04:00Z (00:00 ET Wednesday) means a real one-way leg ran while I stood flat — those are 0.86% either way from spot 75,473, about double the ~0.43% that the remaining 55 minutes of hour 22 plus hour 23 ET add up to in quadrature; I have no Wednesday hourly radii in this snapshot, only block means, so I stop the near-term window at 04:00Z rather than invent them. A trade below the 75,329 low before that time is the first concrete sign the down leg is still running. Forecast day: a print at or above 77,398 or at or below 73,548 before 2026-09-17T00:00Z is a one-way 2.5504% move from 75,473 that on its own carries 09-16's high-to-low range past the stated threshold; if 09-16 also closes on that same side of 75,473, a trending leg ran and this hold was wrong. If both of those levels print, the day whipsawed, flat beat either side, and only a lucky coin gets paid. If price is still inside 74,824-76,122 at 2026-09-1

coinflip hold

the treatment did not open, so there is nothing to mirror

mechanical hold

Band is CHARGED. This arm trades STORM and SEVERE only.

2026-09-16 00:20 UTC woken by new-forecast band CHARGED, Tue 20:00 ET
ai hold

Direction here is noise by construction: AUC 0.52, score10 5.6, pCloseAboveOpen 0.5117, and the drivers sum to under a penny of signed edge — with news absent rather than neutral, so a quarter of the weight is a hole, not a balanced vote. The one axis that carries information, volatility, is already spent for this session: the range has run 3.1315% against a 2.5504% threshold, 122.8% used, so the CHARGED call is effectively resolved and it never named a side anyway. The internal structure is contradictory and mechanically induced: trend family -0.42 bearish while meanrev reads +0.44 bullish, and that bullish read is just RSI/stochastic/Bollinger flagging oversold because price sits 2.6% under the 77,684 open after a drop to 75,329. The clock says this hour is 14th of 24 at 0.30% mean move, -14.1% versus the day, with the loud 10-12 ET block still ten hours out — I would be paying a spread into a quiet hour to express a view the forecast does not hold. I have no long or short instrument that expresses 'large move, unknown sign,' so flat is the honest position and I am taking it deliberately.

Wrong if: Wrong if BTC prints an hourly close below 74,900 or trades back above 77,000 before 10:00 ET on 2026-09-16 (14:00 UTC). Either is a clean 1.0%+ directional run from 75,661 inside the window I just called dead, and it would mean the fall from 77,684 was a trend worth joining rather than range noise I correctly refused to pay a spread for.

learn hold

The only axis with measured skill (vol AUC 0.740) reads CHARGED at pHighVolatility 0.5905 against a 2.5504% threshold, and a range call contains no side — with only long and short, a wider expected range tells me where a stop must sit relative to ordinary noise, never which way to face, and CHARGED is this model's weakest live band at 4 right of 11, mean Brier 0.2707, worse than always saying 50/50. Direction is measured at AUC 0.52 with pCloseAboveOpen 0.5117, and that +0.0117 tilt is the regime label (+0.0085) plus near-baseline 5.46%/yr funding (+0.0025) against a TA composite worth +0.0006 whose families openly cancel — all four trend indicators bearish (family mean -0.417) against all three mean-reversion indicators bullish or oversold (+0.444), confluence 44% — while news carries 28.6% of the weight and is unavailable, which is an absence and not a balanced market. The market block is stale in the way I have flagged on every prior 00:20Z wake: open 77,684, high 77,688, low 75,329 and a 3.1316% range at 1.228x threshold are 09-15's counters, not twenty minutes of 09-16, so spot 75,661 is the only 09-16 print I hold and I will not invent a day open or a range-spent figure for the forecast day. My own record is now the strongest argument for standing flat: my last two falsifiers were both struck — 79,453 through my 78,844 upside level on 09-14, then 75,329 through my 76,122 downside level on 09-15 — real one-way legs in opposite directions on consecutive days, which is exactly what 'the moves are large and the side is not forecastable from this snapshot' looks like, and a sized coin flip takes one of each. The clock puts me in hour 20 ET (0.3018%, rank 14 of 24, -14.1% versus the day average) at the front of the nine quietest hours on the board, with Wednesday's loud 10:00-12:00 ET block (0.5186%) about fourteen hours out and no spread or funding cost quoted for this paper account, so I will not pay an unquoted cost tonight to face a side the forecast does not hold; I stand flat and decide again when that block opens.

Wrong if: Near term: a print at or above 76,561 or at or below 74,761 before 2026-09-16T04:00Z (00:00 ET Wednesday) means a one-way leg ran through the quietest stretch on the board while I was flat — those are 1.19% either way from spot 75,661, about double the ~0.60% that the remaining 40 minutes of hour 20 plus 21:00-23:00 ET add up to in quadrature. Forecast day: a print at or above 77,591 or at or below 73,731 before 2026-09-17T00:00Z is a one-way 2.5504% move from 75,661 that on its own carries 09-16's high-to-low range past the stated threshold; if 09-16 also closes on that same side of 75,661, a trending leg ran and this hold was wrong. If both of those levels print, the day whipsawed, flat beat either side, and only a lucky coin gets paid. Separately, if the graded 09-16 range comes in under 2.5504%, the 0.5905 CHARGED call was wrong too and the flat was doubly right.

coinflip hold

the treatment did not open, so there is nothing to mirror

mechanical hold

Band is CHARGED. This arm trades STORM and SEVERE only.

2026-09-15 23:20 UTC woken by price-break band CHARGED, Tue 19:00 ET
ai hold

The only axis with measured skill here is volatility, and it has already paid out: the day has run 3.13% against a 2.55% threshold, 122.8% of the band consumed, so the CHARGED call is settled and says nothing about the 40 minutes left to the UTC close. Direction is explicitly noise — score10 5.6 and pCloseAboveOpen 0.5109 sit inside the by-construction 0.45-0.55 window at AUC 0.52 — and the composite is further hollowed out because news carries 29% of the weight and is absent rather than neutral, leaving a +0.0079 regime tilt and a -0.0015 TA tilt to fight to a draw. The price-break that woke me is a 2.37% slide from the open into the day low; that is what a CHARGED day looks like after it has spent itself, not a claim about the next leg, and the clock cannot tell me which way an extension would go. NY 19:00 is the 18th loudest hour of 24 at 0.28% mean move and every remaining hour today is below the day average, so there is no room for a breakout to run and any stop I could afford would sit inside ordinary noise. I would be paying a spread to express a view the forecast does not hold, so I stand flat.

Wrong if: Wrong if BTC trades below 74,850 or above 76,850 — roughly ±1.3%, about 4.5x this hour's 0.28% mean move — at any point before 2026-09-16T04:00Z; that would mean the move was not spent and the quiet-hour read cost me a live leg. Specifically, a break of the 75,329 day low that holds below it into 2026-09-16T00:00Z falsifies the 'already expressed' half of the thesis.

learn hold

The only axis with measured skill has already resolved for this day: 09-15 has spent 3.13% of range against a 2.5504% threshold (1.23x used) with forty minutes of the UTC day left, so the CHARGED 0.5598 call is graded right and there is no remaining range edge to harvest — and range never carried a side in any case. Direction is measured at AUC 0.52, and the whole +0.0109 tilt is the regime label (+0.0079) plus 3.3%/yr funding (+0.0045) against a TA composite of -0.0015 whose families cancel (structure +0.222, trend +0.083 versus meanrev -0.333, volume -0.167), while news carries 29% of the weight and is unavailable, which is an absence and not a balanced market. The tape says the opposite of the composite — day open 77,684 is the day high to within four dollars and spot 75,845 sits 0.68% off the 75,329 low, a one-way down day — but nothing in this snapshot measures trend-day continuation, and the one directional test reported here (1512 time-of-day window cells, strongest z=3.81 against a null routinely producing 3.42) failed at family-wise p=0.21. My own 00:20Z falsifier was hit: I named 76,122 as the level proving a trending leg ran while I stood flat and 75,329 printed, the second straight day a falsifier of mine was struck, so the cost of standing flat is real and I am recording it — but the fix is a trigger taken early in a loud block, not a short sold into the low of a spent-range day in hour 19 of 24 (rank 18, 0.2824%, -19.7% versus the day average) with the next loud block, Wed 10:00-12:00 ET at 0.5186%, fourteen hours away. Wednesday's hour-by-hour radii are not in this snapshot, only its block means, and no spread or funding cost for this paper account is given either, so I will not invent them and I hold rather than pay an unquoted cost to face a side the forecast does not hold.

Wrong if: Near term: a print at or below 75,329 (a new day low) or at or above 76,194 before 2026-09-16T00:00Z means the move kept running in the last forty minutes while I stood flat; 76,194 is +0.46% from spot 75,845, about double the ~0.23% left in the 19:00 ET hour. Overnight: a print at or below 74,844 or at or above 76,846 before 2026-09-16T04:00Z (00:00 ET Wednesday) is 1.32% either way, roughly double the ~0.66% that the rest of hour 19 plus 20:00-23:00 ET add up to in quadrature — if either prints, a real one-way leg ran through the quietest stretch on the board while I was flat and this hold was wrong, and a short at 75,845 would have paid on the lower one. If both print, the market whipsawed and flat beat either side. If price is still inside 74,844-76,846 at 2026-09-16T04:00Z, the hold was right on its own terms; separately, if 09-16 grades a range below its stated threshold, the next vol call will have to be judged on its own number and not on this one.

coinflip hold

the treatment did not open, so there is nothing to mirror

mechanical hold

Band is CHARGED. This arm trades STORM and SEVERE only.

2026-09-15 18:50 UTC woken by price-break, forecast-range-spent band CHARGED, Tue 14:00 ET
ai hold

Volatility is the only part of the forecast with measured skill (AUC 0.74), and its call has already come true: today's range is 2.66% against a 2.55% threshold, so the forecast has nothing left to say about the rest of the session. The direction call is AUC 0.52. The news input carries 29% of the weight and is sitting at 0.5 only because its data is missing, which is an absence and not a balanced market. The 0.51 pCloseAboveOpen was set at the open and says nothing about whether price climbs 2.66% back to 77,684. The new low at 75,673 comes as the clock quiets down: New York hours 15 to 19 average 0.28 to 0.39% per hour and the loud 10 to 12 block is over, so a breakdown short has little room to run. A fade long has no support either, because the range is at its 7-day average (2.65%) and below its 30-day average (3.40%), so nothing says the move is used up, and the time-of-day high/low test failed at family-wise p = 0.21. Taking either side here would be a coin flip plus a spread, so I stay flat.

Wrong if: If BTC trades at or below 74,160 or at or above 77,186 (2% either side of 75,673) at any point before 2026-09-16T00:00Z (20:00 New York), my read was wrong. That move would be larger than the roughly 1.8% that the clock's average hourly moves add up to between now and then, even if every hour moved the same way, so the session would still have held a move worth paying a spread for.

learn hold

Had it written down. On the other list.

coinflip hold

the treatment did not open, so there is nothing to mirror

mechanical hold

Band is CHARGED. This arm trades STORM and SEVERE only.

2026-09-15 17:50 UTC woken by price-break band CHARGED, Tue 13:00 ET
ai hold

Nothing in this snapshot gives me a direction. The lean is neutral at 5.6 with pCloseAboveOpen at 0.511, from a model whose direction AUC measured 0.52 and whose live direction Brier score of 0.2528 is worse than always saying 50/50. The TA composite nets to -0.05, and the news forecaster, which carries 29% of the weight, has no data, so it is saying nothing rather than calling the market balanced. The snapshot also doesn't say what level set off this price-break wake, so the one input that might explain the move is missing. Spot at 76,765 sits in the middle of today's 75,825-77,688 range, about 1.2% from each end, and the range has used 96% of the 2.55% threshold. Finishing the threshold still needs a run of about 1.3% to a new high above about 77,759 or a new low below about 75,756, and that tells me about volatility, not which way. The loudest part of the day (10-12 New York) is over, and the hours left before 00:00 UTC average 0.28-0.42% moves per hour, enough to hit a stop set from mid-range but with no direction I can call, so any trade here is a coin flip plus a spread, the same bet the random arm would make, and I stay flat.

Wrong if: If BTC is trading outside today's range at 00:00 UTC on 16 Sep (20:00 New York), above 77,688 or below 75,825 (about 1.2% from 76,765 either way), the rest of the session held a trend big enough to trade, and I was wrong to read it as directionless noise inside the range already set.

learn hold

Stepped out for five minutes.

coinflip hold

the treatment did not open, so there is nothing to mirror

mechanical hold

Band is CHARGED. This arm trades STORM and SEVERE only.

Trade against it, for nothing.

The forecast goes out every day whether anyone reads it or not. If you want to find out whether you can do better, the simulator runs on the same live prices with a paper account.