This document explains the philosophy behind how Aminoid Alpha turns 22 raw indicator readings into a single trade decision. Exact numeric thresholds, weights, and formulas are intentionally not published here — what follows explains the logic, not the constants.
A simpler design would pick a direction early (say, from the 4H trend) and then just check how many other indicators "agree" with that choice. Aminoid Alpha deliberately avoids this, because it quietly hides disagreement — every indicator that doesn't fit the pre-chosen direction just gets ignored rather than counted against it.
Instead, every one of the 22 indicators is evaluated independently for both sides. There is a running long score and a running short score, and each indicator adds to whichever side it actually supports — not both, and not a predetermined one. A coin can (and often does) have indicators firing on both sides at once; that's normal market noise, not an error state.
Direction is decided last, simply by whichever score ends up higher. The gap between the two scores (the "margin") is itself meaningful — a signal that wins 8-to-1 is structurally different from one that wins 5-to-4, even if both technically clear the same minimum score.
Clearing a minimum score is necessary but not sufficient. A signal has to pass all of the following simultaneously:
1. Score threshold The winning side's score has to clear a minimum bar. (The exact number is tuned internally and not published — publishing it would let someone reverse-engineer how "loose" or "strict" the system currently is, which is exactly the kind of tuning knob that took months of live iteration to get right.)
2. Layer participation Confirmation has to come from a healthy spread across the 5 layers (Trend, Reversal, Momentum, Entry Timing, Microstructure) — not from one layer alone stacking up enough individual indicator hits to clear the score bar by itself. This stops the system from calling, say, "four different momentum indicators all agreeing" the same thing as genuine multi-factor confluence.
3. Momentum-or-reversal confirmation (dual-path) A signal can qualify through either of two routes:
- Trend route — the move is already confirmed by real, currently-happening momentum (volume, MACD, VWAP positioning all pointing the same way).
- Reversal route — for setups that are catching a turn before momentum has caught up (since momentum indicators are, by nature, always a little late), a specific cluster of earlier-firing indicators (divergence, proximity to a key level, a fast oscillator cross) can substitute for momentum confirmation instead.
This exists because a purely momentum-gated system tends to alert after a move has already run — the classic "signal arrives once the coin already pumped" complaint. The reversal route exists specifically to catch genuine early setups without loosening the gate for everything else.
4. Disagreement cap If too many indicators are actively firing against the winning direction, the signal is rejected outright — regardless of how high the winning score is. A high score built on top of heavy internal conflict is a different (much riskier) situation than the same score with little disagreement, and the system treats it that way rather than blending the two into one number.
Every signal carries a confidence score (roughly reflecting how strong and clean the setup is) and a suggested leverage cap. Both are calculated by fixed rules from the same inputs used in scoring — layer participation, margin between long/short scores, whether a "BTC shock" is currently active, current market regime, and how much internal disagreement exists. Neither number is ever generated or estimated by the AI reasoning layer; the AI receives them as already-final facts.
As a general (not exact) shape of the logic:
- Higher total score + wider margin + more layers agreeing → higher confidence, higher allowable leverage.
- Any meaningful internal disagreement → leverage is capped hard, regardless of score.
- A currently-active BTC shock → leverage is stepped down further, since sudden BTC volatility tends to whipsail altcoins regardless of their own individual setup.
- Reversal-route signals get one extra step of leverage caution versus trend-route signals of the same score, since they're structurally a bet on a move that hasn't fully started yet.
Entry zone, stop-loss, and take-profit levels are not flat percentage offsets. The system first looks for a genuine, nearby support/resistance level (identified from real recent price structure) to anchor the stop-loss and first target to — and only falls back to a volatility-based (ATR-scaled) offset when no structurally sound level exists nearby. A final consistency pass then guarantees the stop sits outside the entry zone, the first target clears the entry zone, and the second target sits beyond the first — regardless of which method produced each number.
A signal isn't just gated on whether an hour has passed since the last one for that coin/direction. If a coin re-clears the gate after that window but price hasn't actually moved a meaningful amount since the previous signal, it's treated as a stale repeat and suppressed — not re-sent. "Meaningful" is scaled to that specific coin's own typical target distance, not a flat number, since a highly volatile small-cap and a slow-moving major don't share the same definition of "moved enough to matter."
Rather than presenting a raw score, signals are bucketed into a small number of human-readable strength tiers, so the practical difference between a borderline signal and an exceptionally clean one is obvious at a glance without needing to interpret a raw decimal score.