Decision support, not a promise

Analysis that explains itself

The analyzer turns market data into a structured read: a signal, the estimated chance it is right, how the model did on candles it had not seen, the figures it used and exactly what would prove it wrong. It never promises profit.

Run a sample analysis Start on demo
Anatomy of an AI read

Seven parts, each one checkable

Hover or focus a numbered pin to see what a part is, which field carries it and why it is there. This read was produced by the real analyzer from the live feed's candles.

AI readBTC/USDT · 1m
DelayedBinance · reconnecting
SignalBUY
50of 100
Estimated chanceAn estimate, never a promise

Leans BUY: about a 50% estimated chance the price is higher in 10 min. On unseen candles it has not beaten a coin flip here lately (45% of 190), so treat this as close to 50/50.

What drove it

  • Buyer vs seller flow (5 bars): Buyers took 36% more of the last 5 bars’ volume — leans up; it has tended to continue on this market.
  • Share of up bars: 4 of the last 10 bars closed higher — leans up; it has tended to reverse on this market.
  • Buyer vs seller flow (last bar): Buyers took 65% more of the last bar’s volume — leans down; it has tended to reverse on this market.
  • Move over 3 bars: Price moved +0.96 (scaled) over 3 bars — leans up; it has tended to continue on this market.

What would prove it wrong

  • A decisive close below 80,644.16 breaks the bullish read.
  • The fast average falling back under the slow one would weaken it.

Risk factors

  • An estimate from recent price behaviour. Markets change, so the chance shown can be lower on the next trade.
  • Volatility is moderate but can change quickly.
timeframe 1msource binance (delayed)quality goodlast candle 2026-11-05 00:43:00 UTC

  1. BUY, SELL or WAIT. WAIT is a real answer: when the model finds no edge on this market, or the lean is too weak to matter, standing aside is the disciplined output. NO TRADE appears when there is too little data to read.

  2. The model's estimate, from 0 to 100 and never above 80, that the call is right when the trade ends. The model is re-trained on this market's recent candles and first tested on candles it had not seen; the figure is pulled toward 50% by how much that test supports. It is an estimate, not a promise, and it is empty when there is too little history to test.

  3. One plain sentence: which way it leans, the estimated chance, and how the model did on the candles it had not seen.

  4. The measurements behind the call, ranked by how much each moved the estimate: moves over several windows, momentum, distance from the averages, place in the recent range and volatility. Each says whether it supports the call or argues against it, so you can see when the read is a close one.

  5. The conditions under which the read no longer holds, for example a close back below a swing low. If they happen, the read is void and the sensible move is to stand aside.

  6. Standing reminders: the chance is an estimate from recent price behaviour, markets change, and fixed-payout trades risk the whole stake.

  7. Every read records its timeframe, the time of the last candle, whether the data was simulated, live or delayed and how thick it was, so you can judge how much weight it deserves.

A strict output schema

Every read has the same shape, whichever engine produced it. Fixed fields are what let the app show evidence next to the chart and keep a history you can check later.

Signal

scenario
buy | sell | wait | no_tradeThe signal. WAIT when signals disagree, NO TRADE when data is insufficient.
market_state
bullish | bearish | neutral | insufficient_dataThe plain description of the sampled window.
win_probability
number 0 to 80, or nullThe estimated chance the call is right when the trade ends. Null when there is too little history to test the model.
edge.test
{ n, acc }How many candles the model was tested on without having seen them, and the share it called right.
confidence_explanation
stringSays in words what the estimate does and does not mean.

Reasoning

summary
stringOne sentence in plain language.
drivers
object[]Each measurement, whether it supports the call, and how much it moved the estimate.
evidence
string[]The same measurements written as plain sentences.
supporting_conditions
string[]Sample size and price range behind the read.
invalidation_conditions
string[]What would prove the read wrong.
risk_factors
string[]Reminders to weigh before acting.

Provenance

symbol, timeframe
stringWhich market and candle size the read is about.
data_timestamp
UTC datetimeTime of the last candle used.
data_source
stringsimulated when the built-in sample generator is in use, otherwise the live provider's name, with "delayed" added while its feed is catching up.
data_quality
good | limited | insufficientBased on how many candles were available.
model_name, model_version
stringWhich engine produced the read.
disclaimer_version, status
stringThe disclaimer text version, and ok or insufficient_data.
Example payloadDelayedBinance · reconnecting 
{"symbol": "BTCUSDT","timeframe": "1m","data_timestamp": "2026-11-05 00:43:00","data_source": "binance (delayed)","data_quality": "good","market_state": "bullish","scenario": "buy","win_probability": 50,"confidence_level": "low","edge": {"verdict": "none","horizon_seconds": 600,"test": {"n": 190,"acc": 0.4474000000000000198951966012828052043914794921875}},"summary": "Leans BUY: about a 50% estimated chance the price is higher in 10 min. On unseen candles it has not beaten a coin flip here lately (45% of 190), so treat this as close to 50/50.","evidence": ["Buyer vs seller flow (5 bars): Buyers took 36% more of the last 5 bars’ volume — leans up; it has tended to continue on this market.","Share of up bars: 4 of the last 10 bars closed higher — leans up; it has tended to reverse on this market."],"invalidation_conditions": ["A decisive close below 80,644.16 breaks the bullish read."],"risk_factors": ["An estimate from recent price behaviour. Markets change, so the chance shown can be lower on the next trade."],"status": "ok"}

Shortened for the page. Evidence and risk lists are longer in a full read.

Analyzer

Run the real engine yourself

Pick a market and a timeframe. The same analyzer that powers the app reads about 420 candles, tests itself on the newest ones and returns the full schema. No account needed, and nothing is saved.

TradeXReads about 420 candles and tests itself on the newest
DelayedBinance · reconnecting
  1. Read 420 recent candles
  2. Measured 18 signals on each one
  3. Tested the model on 190 candles it had not seen
  4. Estimated the chance and what would prove it wrong
SignalBUY
50of 100

Leans BUY: about a 50% estimated chance the price is higher in 10 min. On unseen candles it has not beaten a coin flip here lately (45% of 190), so treat this as close to 50/50.

What drove it

  • Buyer vs seller flow (5 bars): Buyers took 36% more of the last 5 bars’ volume — leans up; it has tended to continue on this market.
  • Share of up bars: 4 of the last 10 bars closed higher — leans up; it has tended to reverse on this market.
  • Buyer vs seller flow (last bar): Buyers took 65% more of the last bar’s volume — leans down; it has tended to reverse on this market.
  • Move over 3 bars: Price moved +0.96 (scaled) over 3 bars — leans up; it has tended to continue on this market.

What would prove it wrong

  • A decisive close below 80,644.16 breaks the bullish read.
  • The fast average falling back under the slow one would weaken it.

Risk factors

  • An estimate from recent price behaviour. Markets change, so the chance shown can be lower on the next trade.
  • Volatility is moderate but can change quickly.
  • Fixed-payout trades risk your whole stake. Never risk more than you can afford to lose.

Provenance

  • Source: binance (delayed)
  • Quality: good
  • Last candle: 2026-11-05 00:43:00 UTC

Estimated chance the call is right when the trade ends. The model is re-trained on this market’s recent candles and first tested on 190 calls it had not seen (right 45% of the time); the figure is pulled toward 50% by how much that test supports. It is an estimate, not a promise.

An estimate from recent price behaviour. Informational only, never a guarantee of profit and never financial advice.

How the estimate is made

The model is small on purpose, and it is checked before it is believed. It is re-trained on the market in front of it each time, then graded on candles it was never shown.

  1. MeasureEach recent candle is described by about sixteen signals: moves over several windows, momentum, distance from the averages, place in the range and volatility.
  2. LabelEvery past candle is marked by what happened next: was the price higher or lower a trade-length later?
  3. Train and testThe model learns from the older candles, then makes calls on the newest candles it has not seen. How often it was right on those is the honest score.
  4. EstimateThe chance for the current candle is pulled toward 50% by how little the test supports it, and never shown above 80%. If there is no edge, the answer is WAIT.
The test behind the read above
45%right on 190 unseen candles
Always guessing the same side66%
VerdictNo edge found

A test on recent candles is evidence, not proof: markets change, so the estimate can be lower on the next trade. That is why it is pulled toward 50% and capped.

AI analysis you can reason about

The analyzer never promises profit. It shows what it looked at, how it did when tested on candles it had not seen, the chance it estimates, and exactly what would invalidate the idea.

Evidence, not vibes

Every read cites the price data it used: moves over several windows, momentum, distance from the averages, range position and volatility.

The chance is an estimate

It is tested on recent candles the model had not seen, and pulled toward 50% when that test is weak. It is never a guarantee, and the page says so.

Invalidation levels

Each read states what would prove it wrong, so you know when to stand aside.

A record you can review

Analyses are timestamped and stored in your history so you can check them against what happened.

What it is, and what it is not

An AI read is a way to organise what the data shows. It can be wrong, and it should never be the only reason you place a trade.

It is

  • Decision support. A structured read of recent price data, tested before it is believed.
  • Checkable. Every claim cites a figure you can compare with the chart.
  • Honest about doubt. WAIT and NO TRADE are valid answers.
  • Kept apart from execution. It never places a trade on its own.

It is not

  • A promise. It is built from how price behaved recently, and the next trade can still lose.
  • A sure thing. The chance is an estimate that is pulled toward 50% and capped at 80%.
  • Financial advice. It does not know your goals or circumstances.
  • A certainty. Real candles make the inputs real. They do not make the outcome known.
Important. AI-generated analysis can be wrong, does not guarantee accuracy or profit and must not be treated as personal financial advice. Fixed-payout trades can lose your whole stake. Read the risk disclosure before you trade.

Read a market, then decide for yourself

Open a demo account to request AI reads inside the terminal, keep them in your history and check them against what happened.

Create free account Sign in
  • No deposit or card
  • Every read saved to your history
  • Data state always labelled