How DeQrypto works

A Sell signal means the model sees meaningful probability that the price will be lower in 7 days. A minimum conviction threshold must be reached — weak bearish readings produce Hold, not Sell.

A Sell signal on an asset you don't hold is informational context. On an asset you do hold, it indicates the model sees short-term downside risk.

A Hold signal means the model does not have strong enough conviction in either direction to recommend action. This is the correct outcome when the market is sideways, consolidating, or when the evidence is mixed.

Hold is not a weak or failed signal — it is the model accurately describing uncertainty. Most weeks, the majority of assets will be Hold.

Signals are generated automatically every Sunday at midnight UTC, aligned to the weekly candle close. The model is retrained on the latest price history before each generation run, so it always reflects current market behaviour rather than patterns from weeks or months ago.

All users see the same signals — generation happens once per week for the entire platform, not per user.

The top 100 crypto assets by market capitalisation, sourced from CoinGecko each week — excluding sentiment-driven tokens (meme coins), whose prices are dominated by social-media activity rather than market fundamentals. The list updates automatically — assets that drop out of the top 100 are removed; new entrants are added. Assets with insufficient price history are skipped and marked separately.

Free accounts receive signals for the top 10 assets by market cap, Starter for the top 30, and Pro for full coverage.

Confidence reflects the strength of the model's directional prediction. A higher confidence means the model sees a clearer, larger expected move. A lower confidence means the expected move exists but is smaller or less clear.

Confidence and signal direction are independent — a 70% confidence Sell is a strong bearish view; a 55% confidence Buy is a modest bullish lean. Always read both together.

No model can predict crypto markets with certainty. The 95% cap is a deliberate design choice reflecting that residual uncertainty is always present. Any platform claiming 100% confidence on a market prediction should be treated with deep scepticism.

The average confidence score across all signals issued this week. Reflects how strongly the model feels about its calls overall — not just the direction, but the conviction behind them. A week with a high average confidence is generally more actionable than a week of mostly low-confidence signals.

It does not change the signal. A Buy with a High risk level means the model expects a rise, but the coin's price is unsettled right now — if you buy, consider a smaller amount than you normally would. A Buy with a Low risk level comes with calmer conditions behind it.

It is guidance, not an instruction. DeQrypto does not execute trades. You decide what to do with your portfolio.

Three things, all about the coin's market rather than the prediction:

Price swings: how large and erratic the coin's weekly price moves have been recently.

Trend: whether the coin's price is above or below its recent average — rising or falling.

Market stress: when the wider crypto market is under stress, every coin's risk level rises together.

Confidence is about the prediction: how sure the model is that the price will move the way it says. Risk level is about the conditions: how calm or rough the coin's price is right now. They come from separate parts of the model, so any combination can happen — a confident Buy can still have a High risk level.

The Market overview's risk level is different again: it describes the whole market, from the mix of Buy and Sell signals across all coins.

Crypto assets move through repeating cycles of expansion and contraction. The cycle zone tells you where the current price sits within its recent range — from the cycle low (bottom) to the cycle high (top). It provides interpretive context for each signal.

The three values shown on the cycle chart are the lowest price, current price, and highest price within the trailing 26-week (6-month) window. They define the cycle's boundaries and anchor the zone labels.

The Cycle Position % shows exactly where within this range the current price sits — 0% means at the low, 100% means at the high.

Win Rate = total verified Wins ÷ total verified outcomes × 100. Only signals where 7 days have passed and an outcome has been assigned count toward this figure. Pending signals are excluded.

Win rate is recalculated every Sunday before new signals are issued. The process is fully automated — the system fetches current prices, compares them to stored entry prices, and writes outcomes to the database without any human input or intervention.

Shown on the History and Results pages. The all-time win rate for each individual asset across every signal ever issued for it. A coin's rate is shown once it has at least 6 graded Buy/Sell signals.

Some assets will naturally outperform — the per-asset breakdown lets you identify which assets the model has historically been most accurate on.

The Results and History pages break down win rate separately for Buy and Sell signals. This shows whether the model is more reliable in one direction than another — for example, consistently strong on Buy calls but less so on Sell calls.

Hold signals are never graded, so they have no win rate. The Hold row is shown for completeness — to account for how many signals were Hold — but Hold is excluded from win rate and average return.

For each resolved Buy or Sell signal, the platform records the price at signal time and checks the price 7 days later. The actual percentage change is stored for every signal. Average return is the mean of all these values across all resolved Buy and Sell signals.

Hold signals are excluded from this calculation since no trade was recommended — including them would dilute and misrepresent the figure.

Each individual signal in the log also shows an Expected Return — the model's estimate at signal time of the likely price move magnitude. It is a forward-looking estimate, distinct from the actual 7-day return which is the verified outcome recorded after the fact.

Expected return is signed (positive = bullish, negative = bearish) and expressed as a percentage of the current price at signal time.

Max drawdown and Sharpe ratio are measured on a simple portfolio: one that followed every Buy and Sell signal each week in equal amounts. Each week's result is the average 7-day return of that week's signals. A Sell signal counts as a gain when the price falls.

Total gains from all winning signals divided by total losses from all losing signals. Above 1 means gains outweighed losses; 1.5 or higher is commonly considered good.

The largest fall from a peak in the value of the portfolio above, measured week to week. It shows the worst stretch someone following every signal would have been through.

For context it is shown next to Bitcoin's own drawdown over the same weeks, based on its weekly prices. Because values are checked weekly, a dip that recovers within the same week does not show up.

The portfolio's average weekly return divided by how much that return varied from week to week, annualised. Higher means steadier returns for the risk taken; above 1.0 is commonly considered good. No risk-free rate is deducted. Shown once at least 4 weeks of results are available.

7 days after each signal is issued, the current price is compared to the price recorded at signal time. The outcome is assigned as follows:

These rules never change and are applied automatically every Sunday. There is no human review or retroactive adjustment of any outcome.

Pending means the 7-day resolution window has not yet closed. The signal was issued this week and no outcome has been assigned yet. Once 7 days elapse and the next Sunday generation run executes, the outcome resolves to Win or Loss automatically.

A direct count of Buy, Sell, and Hold signals across all covered assets this week. The sub-label shows each count as a percentage of total covered assets — for example "42% of 85 covered assets". These are the raw signal tallies, nothing more.

The platform's all-time verified win rate — identical to the figure shown on the Portfolio Signal performance panel and History page. All three read from the same database. It shows "—" until sufficient signals have been resolved.

A 10×10 grid showing all covered assets colour-coded by this week's signal: green = Buy, red = Sell, yellow = Hold. Assets are sorted by market cap rank — top-left is the largest asset by market cap. Hovering a cell shows the asset name, signal, confidence, and cycle zone.

A horizontal bar chart showing the proportional split of Buy, Hold, and Sell signals this week. Gives an immediate visual read on how bullish or bearish the overall market picture is.

This week's signals grouped into three conviction bands: High (>70%), Moderate (40–70%), and Low (<40%). A week dominated by High-confidence signals is more decisively readable than a week of mostly Low-confidence ones.

A 0–100 score displayed as a needle gauge. It reflects how much of your connected portfolio's value is currently sitting in assets with Buy, Hold, or Sell signals. Sell-heavy portfolios score higher (more risk); Buy-heavy portfolios score lower.

The top 8 signals by confidence score this week, regardless of direction. Each row shows the asset, signal type, confidence, cycle zone, and expected return. These are the signals where the model has the highest conviction — they deserve the most attention this week, but remain context rather than instructions.

Per-holding bars showing what percentage of your total portfolio value is in each asset, colour-coded by that asset's current signal. The summary boxes below show your total portfolio percentage in Buy, Hold, and Sell assets respectively. This is the input that drives the Portfolio Risk Score.

Markets evolve continuously. A model trained once at launch gradually loses accuracy as conditions change — this is called concept drift. By retraining on the latest available price data every Sunday before generating signals, the model always reflects current market patterns rather than historical ones that may no longer apply.

This is the single most impactful operational decision in the platform's design. Internal testing showed win rate decaying meaningfully without weekly retraining, and recovering with it.

The model is trained and run exclusively on historical price and volume data. It does not use social media sentiment, news feeds, on-chain data, or external sources. All inputs are derived mathematically from each asset's own price history.

This means the model's conclusions are based only on what the market itself has done — not on what people are saying about it.

Every Sunday, before new signals are generated, the system automatically resolves all Pending signals from the prior week. It fetches the current price of each asset, compares it to the price recorded at signal time, and writes Win or Loss to the database according to the fixed outcome rules on this page.

This process has no manual steps. No signal outcome is ever retroactively changed. The full history — including every loss — is publicly visible on the Results page.

Asset price and volume data is sourced from public market data providers at weekly candle resolution. The asset universe (which top 100 assets are covered) is determined by CoinGecko market cap rankings, refreshed each generation cycle.