How to read this: each card's big badge is the IN/OUT read from the simple engine —
IN · EARLY UP (EMA12 > EMA26 and 4-bar momentum > +0.5% persisted 2 bars),
OUT · EARLY DOWN (EMA12 < EMA26 and momentum < −0.5% persisted 2 bars),
RANGE · CAUTION (Choppiness Index > 50 — the anti-chop guard: the engine stays flat in
chop instead of whipsawing you in and out). Click 📈 state for the three inputs' live values and
exact rule; click ⚙️ rule for the whole engine. Every box also shows the equity of following
the arrows vs buy&hold — the honest benchmark, so a losing engine is never hidden. Click a
per-tf strip row to chart that horizon: green ▲ IN / red ▼ OUT confirmed arrows by default, the
yellow range band, and an optional equity-curve toggle. The 🌡 line is the old equal-weight vote,
kept as context only. Everything here is a manual RESEARCH read — auto-execution
stays HOLD (nothing clears the ≥45-day out-of-sample gate). Full definitions of every term & signal: 📖 Guide tab.
📈 Price chart
Honest backtest matrix. Every signal scored against buy-&hold over the same window,
with PulseX fees (0.3%/side) + slippage, next-bar entry (no look-ahead). Green = positive; a row
only matters if it beats its buy-&hold column (★) AND clears the sample verdict
(results on <10 trades are flagged NOT MEANINGFUL). This is disclosure — the live strategy is
picked separately on held-out data, never by cherry-picking this table.
🧭 The IN/OUT engine behind the Live boxes (early-up / early-down / range — zero tuned parameters, honest equity vs buy&hold)
loading…
IN/OUT engine per asset × timeframe (following the arrows with 0.4% round-trip cost, vs buy&hold over the same window — the honest benchmark)
loading…
Per-asset live strategy (chosen on held-out data)
Best meaningful signal per asset (in-sample, disclosure only)
🛡️ Crash-preservation view (1d ≈ 4-6mo — this window CONTAINS the crash; research, not the live auto-strategy)
Why this exists: the coins really did crash ~60-73%, and a signal that sits in cash through it
does "win" at not losing — that instinct is correct. The old gate missed it because it only ever saw the
6-week rally (1h/4h) and scored total return (which rewards holding a rally and punishes cash).
Here every signal is re-ranked on the crash-containing 1d window by Calmar (return ÷ |max drawdown|);
cut = pp of the crash the rule rode less than buy&hold (+ is good).
Honest caveat: in-sample the dodgers below are real, but we have one crash in six months = a single observation of the protection regime, so no rule can be auto-certified out-of-sample for live trading without overfitting to that one crash — which is exactly why the live auto-strategy stays HOLD. These are shown as research. The combo "winners" from the 1h search are rally curve-fits (0/5 tokens survive this window — they rode the crash).
Honest caveat: in-sample the dodgers below are real, but we have one crash in six months = a single observation of the protection regime, so no rule can be auto-certified out-of-sample for live trading without overfitting to that one crash — which is exactly why the live auto-strategy stays HOLD. These are shown as research. The combo "winners" from the 1h search are rally curve-fits (0/5 tokens survive this window — they rode the crash).
Full matrix
Token:
Daily flow (green=buy, red=sell, blue=net)
Top wallets by volume
Top accumulators (net buy)
Top dumpers (net sell)
loading the encyclopedia…