Documentation

Quick Read

Start with the regime band, then read the table. The table ranks contracts by scanner observations, but the score is not a trade instruction.

Dashboard Overview

The scanner is organized from market context to coin-level detail.

┌─────────────────────────────────────────────────────────────┐
│ ● SENTINEL                         [Founding Beta] [More]  │
├─────────────────────────────────────────────────────────────┤
│ BTC regime read                                            │
│ ML EDGE / historical bucket read                           │
├─────────────────────────────────────────────────────────────┤
│ Mode / timeframe / correlation controls                    │
├─────────────────────────────────────────────────────────────┤
│ Search, filters, export actions                            │
├─────────────────────────────────────────────────────────────┤
│ #  ★ Coin      Price    24h   Alpha  Win%  vsBTC  Mom ... │
│ 1  ☆ BTCUSDT   ...      ...    ...    ...   ...    ...     │
│ 2  ★ ETHUSDT   ...      ...    ...    ...   ...    ...     │
└─────────────────────────────────────────────────────────────┘

Connection Status

The dot beside the SENTINEL mark shows whether the live scanner is connected.

Modes

Modes change how the scanner weighs observations and which historical window the readouts use.

Regime Band

The regime band appears above the controls so the first read is market context, not table density. It renders a neutral loading state first, then swaps to live values when data arrives.

The ML EDGE element is a historical-bucket readout, not a live prediction. It groups past observations into buckets by mode and score, then reports how that bucket behaved historically. It describes what similar setups did before; it never claims to know what this setup will do next.

Scanner Table

The table ranks tracked USDT perpetual contracts. Cold visitors start with a quiet column set; returning users keep their saved dense views.

Column Reference

Sorting and Views

Click a desktop column header to sort. The column picker can move from the quiet default to a denser desktop view. Mobile uses cards instead of the desktop table.

Alpha Score

Alpha Score is a composite observation from momentum, outperformance, volume, funding, open interest, correlation, and RSI context. It is a ranking signal, not a prediction.

Momentum Checklist

The Momentum Checklist is a compact six-point read shown in the table as the Pump column. It is simpler than Alpha Score: each condition is either present or absent, and the score is the count met.

Filtering

Filters help reduce table noise without changing the underlying scanner data.

Mobile

On mobile, the scanner uses cards instead of the dense desktop table. Tap a card to expand it.

• The compact row shows rank, symbol, main score, and current context.
• Expanded cards expose overview, technical, and trade-planning tabs.
• The trade tab shows risk math from current price and volatility. It is an informational worksheet.
• Cards keep touch targets at a comfortable mobile size.

Exports

The scanner can export the current view for your own research notes. Exports are local browser actions.

How It Learns

SENTINEL is not a fixed set of rules. The scoring engine retrains on fresh market data every day, keeps a versioned history of every model it has used, and runs a heavier challenger model in the background — so it can keep improving without ever risking your alerts on something unproven.

The daily retrain

Every morning at 06:00 desk-local time, the Alpha scoring model retrains on the latest market history and produces a new candidate set of factor weights. Each version is saved, so any change can be traced, compared, or rolled back.

The shadow challenger

Alongside the live model, a heavier gradient-boosted model scores every contract in “shadow” mode. Its scores are recorded and measured but are never used to send you an alert. It is a challenger on probation, earning the right to be promoted — or not.

What stops a bad model from reaching you

A new model cannot replace the live one just because it looked good in a backtest. It has to clear promotion gates that are declared in advance and measured on live-forward data only:

Tiers & Access

Not every read carries the same weight of evidence, so signals are organized into tiers by how thoroughly they have been validated.

How a signal graduates

A new idea starts life in shadow, where it is scored and tracked on live data but never delivered. It soaks for months. Only if it clears the promotion gates above does it move toward a customer-facing tier. Many never do.

Self-Healing

“Self-training” is the model learning. “Self-healing” is the system staying up. SENTINEL is built to recover from the routine failures of live infrastructure without waiting for someone to notice.

• Connection recovery — if the Telegram message loop hits a fatal error, it restarts itself instead of going quiet.
• Safe delivery — every alert is converted to safe formatting at the moment it is sent, so a stray character can never silently swallow a message.
• Scan guards — if a scan cycle is still running, the next one steps aside rather than stacking up and falling behind.
• Data integrity — the history database uses atomic writes, rotating backups, and periodic compaction so a crash cannot corrupt the record.
• Config-drift checks — the system watches for settings or keys drifting between environments and flags the mismatch.

Scope and Disclosures

This public guide is limited to reading the web scanner. Operational runbooks and implementation details are not part of this page.

FAQ

Do I need to connect an exchange account?

No. The public scanner page does not ask you for exchange credentials.

How often does data update?

The scanner updates progressively. Some fields appear before deeper enrichment fields are ready, and the page keeps working while live services warm up.

Why do some fields show a dot?

A dot means the value is unavailable for that contract or the enrichment source did not return it. Missing values are not treated as bullish or bearish by themselves.

Is Alpha Score predictive?

It is a historical, forward-return-calibrated ranking readout. It can help organize research, but it does not predict the next move and does not replace risk management.

Why is the mobile view different?

Mobile cards prioritize legibility and touch targets. The desktop table remains the dense view.

Does SENTINEL self-train?

Yes. The scoring model retrains on fresh market data every day and keeps a versioned history of its weights, and a heavier challenger model runs in shadow for comparison. See How It Learns.

Does it self-heal?

Operationally, yes. The Telegram bot recovers from connection failures on its own, alerts are made safe at the moment of sending so they are not silently dropped, scan cycles guard against pile-ups, and the history database uses atomic writes with rotating backups.

What stops a broken or overfit model from reaching me?

Promotion gates measured on live-forward data, not backtests: at least 30 days and 100 resolved trades, a profit factor of 1.20 or higher with a bootstrap lower bound above 1.05, no more than a 5% chance the profit factor is actually below 1, plus drawdown and Sharpe floors. The thresholds are fixed in advance, so they cannot be tuned to fit. Most candidates never clear them.

What is the difference between CORE and Research Lab?

CORE (sweet and liqsqueeze) are the calibrated, subscriber-visible tiers. Research Lab adds higher-conviction, experimental reads and requires Research Lab access plus a current risk acknowledgement. Experimental means promising but unproven.

Is any tier proven to be profitable?

No, and we do not claim it. We do not market returns we cannot prove on a live ledger — which is exactly why the promotion gates and public receipts exist. Research Lab is frontier access and transparency, not a profit promise.

New to Bybit? You can create an account here. (referral link — see our affiliate disclosure)