Regime-aware signal engine · BTC/USDT · 1H

Most indicators react to price. This one reads the state you're trading in.

A Kalman filter separates trend from noise. A hidden-Markov model names the regime. GARCH forecasts the volatility you're about to walk into. Then a stacked ensemble answers one question per bar, what's the probability the next five close higher?

Behaviour in this regime
Direction taken
89% short, never long
Average exposure
14.0% of equity
Share of bars
16% of bars
Sharpe in this state
0.32
Share of total profit
2.6% of all profit
Detected state, directional Out-of-sample · 2022–2026
P(UP IN 5 BARS)
Scroll
29
Engineered features
3
Market regimes
0
Repainted signals
9yr
Tested history

The market, right now

BTC
Bitcoin / TetherBTCUSDT
, loading
Open full chart ↗
Fetching the last 24 hours…

Convert live rate

Range 24 hours

Low , , High ,

 

Market live

Loading market data…

Performance from daily closes

Seasonals each year rebased to 1 January = 0%

Every year starts at zero so the shapes can be compared. This is a description of what happened, not a pattern to trade, four years of a young asset is far too little to call anything seasonal.

What the standard indicators say

Computing…
BTC/USDT · computed in your browser

This is a readout, not a recommendation. Every figure above is computed here from Binance candles using the published formula for each indicator, and the Buy / Sell / Neutral verdicts follow the standard rules for each one. They are mechanical, they lag, and they disagree with each other constantly. Which is precisely why LATENT exists rather than a moving-average crossover. Nothing here is investment advice.

Bitcoin headlines

Last 24 hours live
Loading headlines…
   

The pipeline

Five layers. Nothing downstream can see the future.

Raw candles go in on the left, a sized order comes out on the right. Each layer has one job, and if any single model fails to converge the pipeline degrades instead of crashing.

01. DATA

Ingest & shape

OHLCV in, log returns and eight momentum lookbacks out. Everything is dimensionless, so nothing anchors to a price level and the same rules hold at $4,000 or $100,000.

OHLCV → returns
volatility-scaled
closed bars only
02. SIGNAL

Read the state

A Kalman filter for direction and velocity, a two-state HMM for whether the tape is directional or range-bound, and a GARCH(1,1) volatility estimate. All three are displayed, none of them gates the signal.

Kalman filter
2-state HMM
GARCH(1,1)
03. MODEL

Vote on direction

Eight momentum lookbacks, 20 to 365 days, each vote on direction by sign. The net vote is the signal. A 29-feature machine-learning stack was built and tested against this and scored worse; the ensemble ships.

8 lookbacks
sign vote
train / test split
04. RISK

Decide the size

Position size targets a fixed 25% annualised volatility, so exposure shrinks as the tape gets wild. A circuit breaker halves size when realised volatility is in its top decile. No leverage.

25% vol target
vol circuit breaker
no leverage
05. EXIT

Place and manage

Positions are held until the vote flips. There is no stop-loss and no take-profit, every stop variant tested cost between 0.17 and 0.31 Sharpe, because exiting a trend early is how trend-following loses money.

hold until flip
no stop-loss
rebalance band

What's actually doing the work

Three ideas the standard indicator set doesn't have.

Weight14%

Kalman velocity

A moving average tells you where price has been. A Kalman filter separates the underlying level from the noise and hands you its velocity, how fast the real trend is moving right now, not fifty bars ago.

Weight11%

Regime classification

The HMM sorts the tape into two states without ever being told what they look like. They did not resolve as bull versus bear. They separated directional from range-bound, and the indicator labels them that way. It is displayed context, not a filter: gating the signal on it cost 0.30 to 0.39 Sharpe in every combination tested.

Weight9%

GARCH sizing

Volatility clusters, quiet hours follow quiet hours, violent ones follow violent ones. GARCH forecasts tomorrow's variance from today's, so your size shrinks before the storm instead of after it.

Pair & timeframe
BTC/USDT, 1-hour bars. Tested on Binance data.
Repainting
None. Every signal is computed on a closed bar and never revised.
Features
Eight momentum lookbacks: 20, 40, 60, 90, 120, 180, 270 and 365 days. A 29-feature set was built and tested; it did not beat these eight.
Validation
Single train / test split: models fitted on 2017–2022, scored on 2022–2026 which was never touched during fitting. Train Sharpe 1.44, out-of-sample 1.17.
Costs modelled
Slippage and commission are subtracted inside the backtest, not bolted on afterwards.
Reporting
Sharpe with its 95% confidence interval, max drawdown, longest stretch underwater, beta and alpha to BTC, and every enhancement that was tested and failed.

What the testing produced

The whole result, not the flattering bit.

These are the figures for the indicator you can buy, measured on 3,240 daily bars of Binance BTCUSDT from August 2017 to June 2026. Models were fitted on the first half and scored on a held-out second half they never saw. Commission of 0.045% per side is subtracted from every fill. Nothing here is live trading.

Size it to youThese are percentages. Put your own capital in and see them in money.

This rescales a historical backtest to your account size. It is arithmetic, not a projection. It does not say you would have made this, and it does not account for the fact that larger orders move the market against you.

Annual return
+20.3%
+$20,300 per year on your capital
Sharpe ratio
1.31
1.44 in-sample · 1.17 out-of-sample
Max drawdown
14.5%
−$14,500 at the worst point
Buy & hold, same window
−83.2%
its drawdown, against our 14.5%

Buy and hold made more money. Over the same nine years it returned 34.3% a year against our 20.3%. But with a Sharpe of 0.78 against 1.31, an 83.2% maximum drawdown against 14.5%, and a worst calendar year of −73.0% against our worst of +2.3%. Scaled down to the same volatility so the comparison is fair, buy and hold turns $100,000 into $254,899 while LATENT turns it into $516,570. We win on risk, not on raw return. If you can hold through an 83% drawdown, buy and hold is the better instrument and you should not buy this.

Equity curve 2022–2026 · out-of-sample
122k 114k 106k 98k
LATENT Buy and hold, same period
Quality breakdown 0–100
CONSISTENCY WIN RATE RISK / REWARD DRAWDOWN SHARPE
78
deflated Sharpe probability, ×100
Average win
+$12,389
per winning trade
Win ratio
40.0%
50 round trips in nine years
Average loss
−$667
per losing trade
VaR 95%
−$1,016
95% of days are better than this
Performance by regime share of trades · win rate
Bull41% of trades
Sharpe 1.49, 44% of bars, same momentum rule
Sideways39% of trades
Sharpe 1.27, 55% of bars, same momentum rule
Bear20% of trades
40% win rate overall. Winners are 18× bigger than losers
Other ratios 9-year backtest
Sortino1.67downside risk only
Calmar1.40return vs worst drawdown
Avg holding49 bars~7 weeks on daily bars
Longest losing run5 round tripsexpect worse live
Time in market76%flat the rest of the time
Annualised volatility15.0%target is 25%, realised is lower
Trade log, sample backtest · BINANCE:BTCUSDT daily
SideOpenedEntryExitRegimeSizeExit reasonP&L
Long2025-10-24111,005111,646Range-bound16.7%vote flipped
Long2025-10-26114,559108,323Range-bound16.7%vote flipped
Long2025-10-31109,608110,098Range-bound17.2%vote flipped
Short2025-11-04101,49795,414Range-bound15.8%vote flipped
Long2026-01-1395,41495,605Range-bound17.7%vote flipped
Short2026-01-2088,42858,625Range-bound17.5%vote flipped

These are backtest figures, fitted on 2017–2022 and scored on 2022–2026, net of a slippage and commission model, not a live track record, and not a forecast. The 95% confidence interval on that Sharpe runs from 0.65 to 1.97, because nine years contains only eight independent annual observations. Every figure is measured on BINANCE:BTCUSDT with the settings stated in each tooltip. All indicator figures are computed on BINANCE:BTCUSDT, other feeds, including TradingView’s BTCUSD INDEX, use different closes and will produce different results.

What it would have done to your capital real years, not a projection

Deepest drawdown over these nine years was 14.5%, the point at which your balance sat furthest below its previous peak. Ending balances say what you would have kept; that number says what you would have had to sit through to keep it.

Starting capital
Started investing
You put in,
LATENT would have made,
Ending balance,
Buy & hold, same period,

These are the actual calendar-year returns from the backtest applied to your number, in order. Not an average, not a projection, and not a promise. The sequence is fixed history and it will never repeat. Two of these nine years returned under 4%, the short grey bars. Change the start year to see how much the entry point mattered.

From the research system

One dial, and what it costs you.

Target volatility decides how much capital rides on each signal. Drag it. Watch the return climb. And then watch the drawdown climb with it.

This dial is a real setting in the indicator. Every figure below is measured on BINANCE:BTCUSDT daily bars, 2017 to 2026, at 0.045% cost per side. The shipped default is 25%. Sharpe is almost flat across the whole range. This dial buys return with drawdown, not with edge. Margin cost and liquidation risk are not modelled.

25% target
volatility
0%20%40%60%80%
Rebalances
361
Sharpe
1.31
Net return
+20.3%
Max drawdown
14.5%

Return and drawdown, across every volatility target 9-year backtest
+48% +32% +16% 0% 0% 20% 50% 80% ━ CAGR ┄ MAX DRAWDOWN

What this shows leverage is not edge. Push the dial right and the return climbs to 38.7% a year, while the drawdown grows to 28.5% and the Sharpe barely moves. You are not buying a better system, you are buying a bigger position in the same one. Anyone advertising a return without showing you the drawdown has chosen the number that flatters them.

From the research system

The slower you go, the more often you’re right.

Same rules, same eight years of Bitcoin. Only the chart interval changes. Pick one and watch how often it asks you to do anything.

Synthetic path, measured cadence. The line is generated, not live market data. What is real is the rate: signals arrive at the frequency measured for each interval, and the figures beneath come from four TradingView runs of the shipped indicator on BINANCE:BTCUSDT, October 2018 to July 2026.

,

Trades a year

42

Win rate

49.2%

Return a year

24.4%

Where every published figure is measured.

Raw price is mostly noise.OHLCV in
A line through it only tells you where it has been.Reacting
A Kalman filter tells you where it is, and how fast.Estimating
Two states wrap the same estimate.Classifying
LATENT.The state you are trading in

What is coming

The releases that move it.

Bitcoin does not trade in a vacuum. CPI prints, rate decisions and payroll numbers move it hard and on schedule. This is the week ahead, marked for the ones that historically matter to crypto.

Time
Currency
Event
Actual
Forecast
Previous

Loading the week…

High impact Medium Low Times shown in your local timezone

Access

Tested the hard way. Sold as a finished tool.

No Telegram group, no VIP tier, no upsells. Pick a length, pay once up front, nothing renews automatically, and check every figure on this page with your own chart in your first month.

Included in every plan

BUY / SELL / SHORT / COVERClear signals on the chart, plus a Kalman trend line and HMM regime shading.
Alerts on everythingEntry, exit and regime change, delivered however TradingView delivers them.
Three fitted models, liveKalman filter, GARCH(1,1) and a 2-state HMM running on your chart, fitted parameters, not the training code.
Simple and advanced modesFixed position size or volatility-targeted. Long and short. Runs on 1H and slower. 15m and 30m are not supported. Intraday needs roughly 425 days of chart history loaded before the first signal, so a free TradingView plan may show nothing on 1H or 4H.

Requires a 1-hour chart or slower. The indicator runs on 1H, 2H, 4H, 6H, 8H, 12H, 1D and 1W. It will not run on 15-minute or 30-minute charts. Its longest lookback is a full year, which needs more historical bars than TradingView will load at those speeds. That is a hard platform limit, not a setting. If you only trade 15m, this is not for you.

Try the system

Quarter

$149
billed $149 every 3 months
  • Full indicator, every feature, no restrictions
  • All signals and alerts
  • Simple and advanced modes, long and short
  • Email support
Start 3 months

30-day refund on faults

Best value
Long enough to judge it

Half year

$249
billed $249 every 6 months · save 16%
  • Everything in Quarter
  • Parameter refits as they land
  • Priority email support
  • Six months is long enough to see the system sit out a bad stretch, which three is not
Start 6 months

30-day refund on faults

For the long run

Year

$419
billed $419 every 12 months · save 30%
  • Everything in Half year
  • Early access to new versions
  • A direct line for feature requests
  • Your renewal price is locked at $419 for as long as you keep renewing
Start 12 months

30-day refund on faults

How to check it in your first week

1

Load it on BINANCE:BTCUSDT and use TradingView bar replay. Step through nine years and watch every signal it would have given, one bar at a time.

2

Check the strategy tester against the figures on this page. Same symbol, same timeframe, same 0.045% commission. If they do not match, ask for your money back.

3

Enter your TradingView username at checkout. Access is granted by hand, usually within 12 hours.

Why 30 days rather than a free trial? Because a month is long enough to replay nine years of history on your own chart and confirm the indicator does what this page says, and short enough that the signal set stays worth paying for. You are not asked to trust a screenshot. You are asked to check it. There is still no live track record, only out-of-sample backtests, and that is stated on every page rather than buried.

5% per sale

Refer & earn

Bring someone with you

Your link earns five percent of whatever they buy, and again each time they renew. Sixty days to convert, twenty dollars to cash out, no minimum audience to apply.

Small on purpose. We would rather pay less and have people quote the drawdown. The reasoning is on the programme page.

Programme details ↗

Already an affiliate? Your link is emailed to you on approval

Before you start

The questions worth asking.

Will this make me money?

I don't know, and anyone who tells you they do is selling you something worse than this. What I can tell you is exactly how it was built, exactly how it was tested, and exactly what those tests produced. Then you decide. Backtest results are evidence about the past, not a promise about the future. Markets change and edges decay; that's the normal life cycle of a strategy, not a defect.

Does it repaint?

No. Every value is computed from closed bars only, and once a signal prints it never changes. You can verify this yourself by replaying history bar by bar, the setup guide shows you how.

Refunds?

Thirty days, if the indicator does not work as described: it fails to load, the signals do not match the published figures, or something is genuinely broken. Email the address in the footer with your order number and chart access is removed once the refund is processed. What is not refundable is disappointing trading results; that is the risk this site warns about throughout, not a fault in the product. If you are in the EU or UK your statutory 14-day withdrawal right applies regardless of this policy.

Do I need to know how to code?

No. You add it to a chart, pick simple or advanced mode, and set your alerts. That is the whole job. There is nothing to install, nothing to run and no code to touch.

Can I use it on other pairs or timeframes?

Timeframes: 1 hour and slower only, 1H, 2H, 4H, 6H, 8H, 12H, 1D and 1W all work. 15-minute and 30-minute charts will not run it, because a one-year lookback needs more bars than TradingView loads at those speeds. The script tells you plainly rather than failing quietly. Pairs: the published figures are BINANCE:BTCUSDT and that is what the parameters were fitted on. It will produce output on other markets, but treat that as untested.

How does the indicator reach me?

It is published on TradingView as an invite-only script, so there is nothing to install. You enter your TradingView username at checkout and access is attached to your account by hand, usually within 12 hours. It then appears under Indicators → Invite-only scripts.

Why isn't there a live track record?

Because there isn't one yet, and inventing one would be fraud. What exists is out-of-sample walk-forward testing, a much stricter bar than the train-once-test-once backtests most indicators are sold on. But it's still testing, not trading. That distinction is stated plainly rather than buried.

How do I know the figures on this page are real?

Load it on BINANCE:BTCUSDT, open the strategy tester and set the same 0.045% commission. The numbers should match what is published here. Use bar replay to step through nine years of history and watch every signal it would have given. If the indicator does not work as described, ask for a refund inside 30 days.

Risk disclosure

LATENT is an analytical tool. It is not investment advice, financial advice, or a recommendation to buy or sell any asset, and nothing on this page should be read as one. No advisory relationship is created by your purchase.

Trading cryptocurrency involves substantial risk of loss, including loss of your entire capital. Leveraged trading can produce losses exceeding your deposit. You are solely responsible for your own trading decisions and for complying with the laws and regulations of your jurisdiction.

All performance figures on this page are derived from historical backtests. Hypothetical and simulated results have well-documented limitations: they are prepared with hindsight, they do not represent actual trading, and they cannot fully account for the effect of financial risk in live markets. Past performance, real or simulated. Does not indicate future results.

About Bitcoin

The asset underneath all of this.

Background, for anyone who wants it. The figures below are read from the blockchain itself rather than typed in, so they cannot go stale.

What is Bitcoin?

A protocol for transferring value between people without a bank, a payment processor or any other intermediary in the middle. Transactions are broadcast to a network of independent computers, grouped into blocks roughly every ten minutes, and appended to a shared public ledger that every participant can verify for themselves.

The design was published in a nine-page paper on 31 October 2008. The first block was mined on 3 January 2009. Nothing about it has been controlled by a company since.

Who created it?

Someone using the name Satoshi Nakamoto. The name is a pseudonym, the identity behind it has never been established, and the last widely accepted communications from the account were in 2010 and 2011.

An early mining address believed to belong to them holds a large amount of bitcoin that has never moved. Whether that is deliberate, lost, or something else is unknown, and anyone telling you otherwise is guessing.

What makes it different from ordinary money?

Three things. The supply is capped at 21 million and that cap is enforced by every node on the network, not by policy. No single party can create more, freeze an address, or reverse a settled transaction. And the ledger is public, so anyone can audit the entire monetary base without permission.

Those same properties are also its downsides: an irreversible payment sent to the wrong place is gone, and a lost key is a lost balance.

How many bitcoin exist?

Issuance is fully deterministic. Each block creates a fixed number of new coins, and that number halves every 210,000 blocks. The figures at the top of this section are computed from the current block height using exactly that rule, which is why they agree with the blockchain rather than approximating it.

The twenty-millionth bitcoin was mined in March 2026. The remaining supply takes until roughly 2140 to issue, because each halving slows the rate further.

What is the halving?

Every 210,000 blocks, about four years, the reward paid to miners for each block is cut in half. It has happened in 2012, 2016, 2020 and April 2024, which reduced the reward from 6.25 to 3.125 BTC. The next one occurs at block 1,050,000.

It is often described as a bullish event. What can be said with confidence is only that it halves the rate of new supply; anything beyond that is a claim about how markets will react, and four observations is not a pattern.

How is the network secured?

By proof of work. Miners compete to find a hash below a target value, which takes an enormous amount of computation and no shortcuts. Rewriting a past block would mean redoing that work for every block since, faster than the rest of the network is extending the chain.

The difficulty of the puzzle readjusts every 2,016 blocks, roughly every two weeks, so blocks keep arriving about every ten minutes regardless of how much hardware joins or leaves.

Is it a store of value?

That is a claim, not a fact, and it deserves to be treated as one. The argument for it is fixed supply, portability and settlement that does not depend on any institution. The argument against is that it has repeatedly fallen more than 70% from its highs, which is difficult to reconcile with the ordinary meaning of the phrase.

What is measurable is that its volatility is far higher than gold or major currencies. Whether that resolves over decades is genuinely unknown.

How does it get upgraded?

Slowly and by consent. Changes are proposed as Bitcoin Improvement Proposals, implemented in node software, and activated only once enough of the network is running code that supports them. There is no company that can ship an update.

The two most significant recent ones are SegWit in August 2017, which restructured how transaction data is stored, and Taproot in November 2021.

What is Taproot?

An upgrade activated in November 2021 that introduced Schnorr signatures and a more efficient way of expressing spending conditions. In practice it makes complex transactions, multi-signature arrangements, for instance. Cheaper, and makes them look identical on-chain to ordinary ones, which is a privacy improvement as much as an efficiency one.

What is the Lightning Network?

A second layer built on top of Bitcoin for small, fast payments. Two parties open a channel by committing funds on-chain, then transact between themselves as many times as they like off-chain, and settle the net result back to the main chain when they close it.

It makes payments effectively instant and very cheap. The trade-offs are that funds must be locked in channels and the routing between distant parties is more fragile than a single on-chain transfer.

Who holds the most?

The largest known corporate holder is Strategy, the company formerly called MicroStrategy, by a very wide margin over any other public company. Several bitcoin ETFs and the major exchanges also custody large amounts on behalf of clients.

Deliberately no figure here. Published estimates of these holdings disagreed with each other by hundreds of thousands of coins when this page was written, and a number that wrong is worse than no number. Bitcoin Treasuries tracks it if you want the current position.

What are the actual risks?

Price volatility large enough to halve a position in weeks. Regulation that varies by country and changes. Exchange failure, which has wiped out customer balances repeatedly. Irreversible transactions, so a mistyped address is final. And self-custody, where losing a key loses the funds with no recourse.

None of these are hypothetical. Each has cost people a great deal of money already.

Nothing in this section is investment advice. It is background on how the protocol works, written to be checkable. The live figures are computed from the current block height using the protocol's own issuance rule; the prose describes design and history rather than predictions.

Check it yourself.

Nine years of history, replayable on your own chart. Thirty days to decide whether the numbers hold up.