strategy-dev

AI Signals — read the verdict, judge whether to act

BUY / SELL / HOLD with confidence, expected range, P(Up) / P(Down) — and the 4-way matrix for deciding what to do

6 min · beginner

What you'll have when finished

  • Read every stat on the Signal Card without guessing
  • Distinguish "confidence" (indicator agreement) from "expected range" (forecast cone)
  • Use the 4-way matrix to decide whether to act, wait, or sit out
  • Avoid the most common AI-signal misreads

Before you start

  • No signal predicts the future. The engine describes what indicators say RIGHT NOW.
  • HOLD is a real signal, not a placeholder. Most retail traders lose by feeling they must always be in the market.
  • Confidence < 60% means the indicators are split. Sitting in cash IS a position.

Walkthrough

  1. Pick a symbol — start with what you actually want to trade

    Don't shop for signals. Decide what you want to trade FIRST, then check the signal. If you're considering buying ETH, load ETH/USDT. If you're considering trimming BTC, load BTC/USDT. The signal is for confirming or invalidating your existing thesis, not for generating ideas from nothing. **Anti-pattern:** loading 5 symbols in rotation looking for a BUY signal to act on. That's called "shopping for signals" and it's how impulse trades get justified after the fact.

    Success criteria: Loaded the symbol you actually want to trade · Did NOT shop multiple symbols looking for confirmation

  2. Read the main Signal Card — three things at once

    The big circular badge in the center tells you THREE things simultaneously: **Direction** — BUY / SELL / HOLD. The verdict. **Confidence %** — how unanimous the 8 underlying indicators are. The ring around the letter visualizes this. **Color** — green (BUY), red (SELL), amber (HOLD). **The most common misread:** treating direction and confidence as independent. They're not. A "BUY at 52% confidence" is barely different from HOLD — the engine is leaning but not committed. A "BUY at 85% confidence" is the indicators screaming in unison. Always read direction AND confidence together as a pair.

    Success criteria: Read direction + confidence together · Understood the ring visualizes confidence not strength

  3. The 4 mini-stats — Risk / Expected Range / P(Up vs Down) / Window

    **Risk Level** (Low / Medium / High / Extreme) — current volatility regime. Even a correct signal can have big drawdowns when Risk is High. Adjust size. **Expected Range** (e.g. -5% to +7%) — the forecast cone over the window. The actual move can fall outside in high-vol regimes. Treat as "this is what the model thinks is reasonable" not as "this is what will happen." **P(Up) / P(Down)** — probabilities. Don't expect them to sum to 100% — the remainder is the HOLD probability. P(Up) 65% / P(Down) 20% means the engine sees 15% probability of no clear move. **Window** — typically 24h. Don't apply a 24h signal to a 1-hour decision. Match your trade horizon to the signal horizon.

    Success criteria: Read all 4 mini-stats · Understood window matches forecast horizon, not trade duration

  4. The Reasoning panel — count the indicator votes

    The Signal Reasoning panel lists the WHY behind the verdict — numbered explanations like "RSI(14) at 28 = oversold" or "Price broke above 50-day EMA." Each reason is one of the 8 indicators voting. The more reasons listed, the higher the confidence: - **6+ reasons** = strong consensus, indicators broadly agree - **3-5 reasons** = moderate consensus, mixed evidence - **0-2 reasons** = weak signal, mostly silent indicators The **Key Indicators tags** below are color-coded by tone: - **Green** = bullish vote - **Red** = bearish vote - **Amber** = warning / overbought / oversold - **Blue** = informational / neutral Count the colors. If you see 6 greens and 1 red, the bull case is strong. If you see 3 greens and 3 reds, the engine is genuinely uncertain (which the headline confidence will reflect).

    Success criteria: Read each reason · Counted the indicator tag colors

  5. The 4-way decision matrix

    Combine **strength of signal** (high vs low confidence) with **direction** (BUY/SELL vs HOLD) to decide: | | High Confidence (75+) | Low Confidence (<60) | |----------------|-----------------------|----------------------| | **BUY/SELL** | Trade with the signal | Reduce size or wait | | **HOLD** | Do nothing — wait | Mixed market, sit out | **High confidence + BUY** → take the trade with normal sizing **High confidence + SELL** → take the short or trim existing longs **High confidence + HOLD** → genuinely no edge right now. The engine is unanimously confused — sit out **Low confidence + BUY/SELL** → engine is leaning but uncertain. Half-size or wait for higher confidence **Low confidence + HOLD** → the most common state. Markets are mostly noise; don't trade noise **For new users:** only trade on high-confidence BUY or SELL signals for the first month. You'll trade 1-2 times per week per symbol at most. That's the right pace.

    Success criteria: Applied the matrix to current signal · Decided act / half-size / sit out

  6. Detected Patterns — context, not a separate signal

    The Detected Patterns panel shows chart patterns the engine identified — Ascending Triangle, Double Bottom, Bull Flag, etc. Each pattern shows: - **Direction** (bull / bear / neutral) - **Confidence %** — how clean the pattern is - **Breakout Probability %** — likelihood it resolves as predicted - **Risk level** (low / high) **Important:** patterns are CONTEXT, not a separate signal to act on. The main signal (BUY/SELL/HOLD) already fuses 8 indicators. Patterns add color but don't override. **Use them when:** - The main signal is HOLD but you see a high-confidence bullish pattern → the engine sees something brewing - The main signal is BUY and a bullish pattern is detected → extra confirmation - The main signal is BUY but only bearish patterns appear → caution flag Empty patterns = no clear shapes right now. Common in choppy markets. Not a problem.

    Success criteria: Read pattern direction + confidence · Did NOT treat a pattern as a primary signal

  7. Signal History + Performance — verify the engine has been right for THIS pair

    The **Signal History** table shows the last 14 days of verdicts for this symbol. Use it to spot: - **Frequency** — is this pair generating signals daily or just once a week? - **Consistency** — has the engine been mostly BUY for the last week, or flipping between BUY/SELL? - **Accuracy** — Result column shows what happened (✓ correct, ✗ wrong, ◌ pending) The **Performance** cards above (Win Rate, Avg Return, Max DD, Total Signals) appear ONCE signals have completed their windows. Until then you see "Performance pending" — that's honest, not broken. **For platform-wide accuracy** (across all symbols and users), check [Forecast Accuracy](/forecast-accuracy). That page surfaces the engine's overall track record with Wilson-bounded statistics. **Action rule:** if the engine has been wrong 4 of the last 5 times on this pair, ignore today's signal. The engine isn't calibrated for this symbol right now.

    Success criteria: Reviewed the history table · Decided whether to trust today's signal based on recent performance

What's next

Once you've decided to act: - **For a one-shot trade with TP + SL** → [Smart Trade](/smart-trade) or [DEX Smart Trade](/dex-smart-trade) - **For an entry you want to size cautiously** → [Algo Orders → TWAP](/algo-orders) breaks size into time-spaced child orders - **For accumulating into a position** → [DCA Safety](/dca-safety) or [DEX DCA](/dex-dca) on a schedule - **For executing signals automatically** → wire AI Signals into a [Conditional Bot](/conditional-bots) rule For automated regime-switching (so you don't have to read this page manually every day), the [Conditional Bot](/conditional-bots) runs similar logic on a timer and activates the right inner bot based on the current regime.