Transparent Track Record · methodology-first
Historical signal measurements — measured, not promised. Scoring was frozen on 30 Jun 2026, so almost all of this record is out-of-sample, and out-of-sample it is roughly break-even before costs. Every rate on this page ships with its confidence interval, its profit factor and its expectancy, and names the window it was measured over. These are not guaranteed trading profits.
Period now drives the Forward-Tested Track Record tiles and equity curve as well as the windowed analysis further down. It starts on Full span — no window applied — and every figure names the window it was measured over. The Confirmation-signal accuracy card just below is the one exception: it is a fixed trailing 30-day measurement from /v1/stats and says so on its own face.
This page draws two different records, and “full span” means “no window applied” to each — not the same dates. logged signals (one row per emission) drive the summary cards, the windowed analysis and the per-symbol cards. confirmation calls (5-minute re-emissions collapsed) drive the Forward-Tested Track Record and start later. Each section prints its own first→last dates on its own face.
Our engine produces two kinds of signals with very different quality, and we report them separately and honestly: the higher-quality Confirmation signals (Type 1) — a real-time whale-aggregation and confirmation layer with measured in-sample directional accuracy, though not a standalone profit engine after costs — and raw Regime-Flip events (Type 2) — positioning context that shows what whales are currently doing, not a tradeable prediction (no measured out-of-sample edge). Every headline number on this page is Type 1 only. Type 2 lives in its own clearly-marked section further down.
Directional accuracy of the smart_money_confirm subset — the same numbers shown on the homepage, sourced from /v1/stats. This card is a trailing 30-day measurement at a fixed 24h horizon, not the full record — the Period control above drives the track record and the windowed analysis, not this card. For the whole record with no window applied, read the Forward-Tested Track Record below; it is materially worse. Each rate is shown with its 95% confidence interval, its profit factor and its expectancy, because a rate on its own does not say whether the trades made money.
Window: trailing 30 days, fixed 24h horizon · dedup Rule B (episodes keyed on symbol + direction + confidence tier). This card does not follow the Period control above. Loading live values…
Whiskers = 95% confidence interval (Wilson) from the resolved sample size — the honest uncertainty around each rate. The dashed line marks 50%, what a coin flip would score.
Methodology. These figures are the smart_money_confirm subset, measured over distinct signal calls (— calls, — resolved outcomes at the fixed 24h horizon — not a 4–24h blend), direction-adjusted. A live call is re-confirmed every few minutes while it persists — the raw feed logs thousands of rows — but each episode counts once here (a >12h gap starts a new call), so the rate cannot be inflated by how long a call stays live or how often a bot polls it.
The collapse is a selection, not just a guard. Entry is the first emission of an episode, and first emissions score differently from the average emission — so collapsing re-emissions moves the measured rate rather than merely protecting it. Measured live, both ways, over the same window: loading… We publish both so you can see which way it moved.
Span and sample. Signals on record since Jun 10, 2026, a span of —. Scoring was frozen on 30 Jun 2026, so only the first three weeks are in-sample and is out-of-sample. The tiles above are a trailing 30-day slice of that out-of-sample record, and a favourable one: over the full span the same engine measures materially worse. Treat any of it as descriptive of one market regime, not as a forward guarantee.
Frozen forward holdout — every signal since the scorer freeze, at the fixed 24h horizon, dedup Rule B: win rate loading…, 95% CI —, profit factor —, expectancy — per call. The rate is only printed when the profit factor and expectancy can be printed with it. A profit factor at or below 1.0 with a negative expectancy is what "no durable edge" looks like when only the rate is quoted — the calls are right slightly more often and wrong slightly bigger.
Historical directional accuracy of confirmation signals — not trading profit. Past performance does not guarantee future results. Not financial advice.
Every smart_money_confirm signal was published in real time and scored against the actual subsequent price — a forward test, not a hindsight backtest. Measured over distinct signal calls (5-minute re-emissions of the same call collapsed to one), the standalone directional record is roughly break-even before costs and slightly negative after — which is exactly why we position these as a confirmation / veto layer on top of your own setup, not a standalone strategy. We show the full curve, honestly, either way.
Window: full span of the confirmation-call record — every distinct confirmation call on record. This is a different, shorter record than the one the Period control’s day count names (logged signals, which starts earlier); the same button un-windows both, and each prints its own dates. Set by the Period control above. Loading live values…
The first three tiles are pre-fee; max drawdown follows the Cost basis selector below and is currently net of fees. They are not the same basis and the row does not pretend otherwise.
Liquidation. Any figure whose levered value passes −100% of the margin is shown as liquidated, not as an impossible number — at 100x the worst recorded move on this record is roughly −590%, i.e. the position is gone several times over. Equity lines are likewise stopped and marked where the compounded account reaches −100%. At high leverage most of this record wipes out; that is the honest answer, and it is the most useful thing this control can tell you.Not modelled: funding / borrow cost, a real liquidation engine (maintenance margin liquidates you before −100%, not at it), margin calls, exchange leverage limits, partial fills, and the fact that these signals overlap heavily (real margin could not carry every one at once). Per-position liquidation is not modelled either: the replay lets one signal lose more than the margin posted behind it, which is what cross margin does to an account — on isolated margin the loss would stop at that position’s margin and the trade would simply be over. The two views can therefore disagree, and both are shown: a per-signal row reads liquidated as soon as the move exceeds the margin behind that one position, while a “Return @ N% size” row is about the whole bankroll and only wipes out when the compounded account does. The per-signal statistics are the server’s gross (pre-cost) numbers scaled by leverage — the cost is charged on notional, so at Nx the round-trip fee is N × 0.08% of your margin too; the equity curves below do carry the cost basis you selected.
Each line is a different position size per signal (5 / 10 / 20 / 30 / 40% of bankroll, compounded); the muted dashed line is BTC buy-and-hold over the same window (both start = 100). The 5 / 10 / 20% lines are the published server-computed curves; 30% and 40% replay the same recorded per-signal returns at a larger fraction — no new measurements. Both return and drawdown scale with size — and since this standalone record is around break-even, larger sizing mostly just amplifies the swings. Where the selected window reaches back past the 30 Jun 2026 scorer freeze, the left-hand region is shaded in-sample (scoring was tuned there) and everything to its right is the forward holdout. A window that lies entirely after the freeze is entirely out-of-sample and nothing is shaded — the caption below the chart says which of the two you are looking at. The per-signal stats above don’t depend on position size — but they do scale with leverage (a 1% move at 10x is 10% of the margin behind that trade), while the ratio stats (win rate, profit factor, payoff, Sharpe-like) cannot move at all, because leverage multiplies wins, losses and fees by the same constant. This is why the signal is built to confirm your setup, not to be traded blindly.
Forward test, not a backtest. Signals were generated live and scored on real subsequent price — no hindsight, no curve-fit entries. Scoring was frozen on 30 Jun 2026: only the first three weeks of this record are in-sample, and is out-of-sample. The frozen forward holdout reads loading…
The de-duplication is a selection. A live call re-fires every ~5 minutes; we collapse each run into one episode and enter at its first emission. First emissions do not score like average emissions, so this moves the measured rate — sometimes up, sometimes down — rather than only protecting it against poll-count inflation. Measured on the window shown: loading…
Three de-duplication rules are live on this page, and each figure names its own. Rule A (these tiles and the equity curve) keys episodes on symbol + direction and scores the average of the 4h, 12h and 24h moves. Rule B (the confidence-tier card at the top of the page and the forward holdout) adds the confidence tier to the key, so a call upgraded MEDIUM→HIGH mid-run counts as two episodes, and scores one fixed 24h horizon. Rule C is Rule B scored for payoff only. They are answers to different questions, which is why the same signals can read as three different counts — that is the reason the counts differ, not a discrepancy hidden behind one word.
Pre-fee, no stop-loss. Figures use the raw 4–24h move; a stop would cap the −4%+ loss tail further. Drawdown depends on position sizing (shown above).
Past performance does not guarantee future results. Not financial advice.
The unfiltered feed — every smart_money_confirm signal as it fires, with symbol, side, entry price, time and result. Nothing cherry-picked.
— raw rows = the same calls re-firing every ~5 min while they stay live. The track-record tiles collapse these into — distinct calls over the window selected above (full confirmation-call span) under Rule A (a >12h gap starts a new call, keyed on symbol + direction, scored on the average 4–24h move) — that is the number in the Win rate tile above. The confidence-tier card at the top of the page counts — instead, because Rule B also keys on the confidence tier and scores a fixed 24h horizon. Two counts, two questions, both real. The hard-coded “41” that used to sit here belonged to neither. See the methodology for the full breakdown.
| Generated · age | Symbol | Side | Entry | Avg 4-24h | 24h | 48h | 72h |
|---|---|---|---|---|---|---|---|
| Loading live feed… | |||||||
Each column is the direction-adjusted move at that fixed holding period (green ✓ = correct call). Open signals show a live ◉ LIVE mark in the 24h column until they close at the 7-day mark. Same data behind the charts — download it (4h–7d) and check the math.
Only fires when several independent sources agree: derivatives positioning (funding, long/short ratio, open-interest momentum) + whale consensus + on-chain metrics + price-zone (premium/discount inside the 24h range). A stacking gate requires ≥2 factor families to confirm the same direction, and a confidence threshold drops the rest. Fewer calls, and the profit factor for each tier is measured rather than asserted here — the live figures are in the track-record tiles above, and they move.
Fires whenever the aggregate whale consensus flips net direction (long↔short) on a symbol between snapshots. It describes what whales are currently doing — a lagging snapshot of crowd positioning, not a forecast. It catches every wobble, most of which mean-revert. On its own it has no measured out-of-sample directional edge; its value is situational awareness, and as one input the Confirmation engine weighs.
Why show a losing signal at all? Because hiding it would be dishonest — and because the gap between Type 1 and Type 2 is the product: the Confirmation engine's entire job is to filter this Type-2 noise down to the Type-1 signal. Figures computed live from the same public /v1/signals feed, direction-adjusted 4–24h, pre-fee.
How a stop-loss policy + leverage reshape either signal type's outcome distribution — win rate, profit factor, expectancy, drawdown, and liquidation rate. Click to expand.
Every signal is replayed on its 15-minute price path. Pick a stop-loss policy and a leverage to see how the outcome distribution changes. Numbers are pre-fee, in-sample, over distinct calls (bot-poll repeats collapsed). This is a transparency tool, not a promise — many rows lose money (profit factor below 1) because these signals have thin-to-no edge once stops, leverage and cost are applied. Losing rows are shown and labeled, not hidden.
The Hold to horizon & close row (highlighted) is the current site scoring baseline: no stop, close in profit or loss at the horizon. Every stop level is a separate row. Click any row to select it for the summary and sweep below.
| Stop-loss policy | n | Win % | Profit factor | Expectancy % | Avg win % | Avg loss % | Max DD % | Liq % |
|---|
Same policy, leverage swept 1x→10x. Watch the liquidation rate and max drawdown climb. At 1x there is no liquidation.
| Leverage | Liq threshold | Win % | Profit factor | Expectancy % (ROE) | Liq rate % | Max DD % |
|---|
Everything below this line is re-measured over the selected period. The flagship cards above it are fixed full-sample measurements and never move.
Measuring…
Complementary to the HIGH-confidence confirmation subset above — this is a broader, rolling … measurement over all actionable signals (a different, larger dataset), so its numbers differ from the confirmation-signal figures at the top. This table contains two different accuracy measurements that answer different questions. Time-window accuracy measures what percentage of all signals were correct when checked at a specific time horizon (1h, 4h, 12h, 24h). Confidence-level accuracy measures correctness broken out by how confident the system was — HIGH or MEDIUM — regardless of when the outcome was evaluated.
| Metric | Accuracy | Correct | Wrong | Signals |
|---|---|---|---|---|
| HIGH confidence (example) | 71% | — | — | — |
| Sample row — live accuracy data loads when you open this page. | ||||
Directional accuracy of all actionable signals measured at each evaluation window — the same resolved outcomes shown in the time-window table above, drawn straight from live tracking with no smoothing. Accuracy tends to be highest at short horizons and decay as the window lengthens; we show that honestly rather than cherry-picking the best number. This is the "was it right when checked at 1h / 4h / 12h / 24h?" question — separate from the confidence-tier split (HIGH vs MEDIUM), which asks "how sure was the system when it fired?"
Directional win rate at each horizon (1h / 4h / 12h / 24h), computed from resolved outcomes only. Whiskers = 95% confidence interval; dashed line = coin flip (50%). Hover a bar for the resolved sample size.
Correct (green) vs wrong (red) signal counts per horizon — the raw resolved sample behind each accuracy bar. Taller stacks = more signals scored at that window.
Cumulative vs daily win rate — real tracked data
smart_money_confirm win rate over timeRolling 20-call directional win rate over the resolved distinct calls in the selected Period above — the same collapse the tiles use (a >12h gap starts a new call), not the raw 5-minute re-emissions (direction-adjusted, average of the 4h, 12h and 24h moves — the same measure as the track-record tiles, not the best of the three). It was computed as the best of the three horizons until 2026-08-30, which is look-ahead and drew this line near 74% while the tiles read 54%. Almost all of this record is out-of-sample (scoring froze 30 Jun 2026). Dips below the 50% coin-flip line are real and shown.
Breakdown by confidence level
Signal accuracy (4h/12h/24h resolved outcomes) — only symbols with 10+ scored signals shown
Running win rate for each symbol as outcomes accumulate through the selected period — each point is that symbol's record up to that date, not its result on that date. Click a chip to show or hide a symbol. Gaps in a line are days with no scored signal; they are left empty rather than bridged.
Built only from the logged-signal record (performance_log — one row per emission): never the collapsed confirmation calls the Forward-Tested Track Record measures, and never the regime-flip rows spliced into the feed further down. Independent of the Period control above.
Showing full history · all directions · all confidence tiers. The whole logged-signal record covers every logged signal on record.
The badge is raw logged signals for that symbol — one row per row in performance_log, with no episode collapsing. The percentage is correct / (correct + wrong); neutral outcomes (moves under the threshold) are excluded from the denominator, never counted as wins, and shown separately.
There are three denominators on a card, and a count that is not one of them. Correct + Wrong is the win rate’s denominator. Correct + Wrong + Neutral — the resolved outcomes — is the denominator of the all-outcome reading at the foot of the card. The logged badge is a count of rows and is never divided by. The gap between them is Pending: signals that have been logged but not scored at 4h, 12h or 24h yet. A pending signal is not an outcome, and it is never counted as a neutral one.
The badge and the percentage are two different denominators, and the card says which is which. Under every rate is the line that reconciles them — e.g. “12 of 14 directional; 27 neutral excluded” beside a badge of 41 logged — and at the foot of the card is the same symbol scored over all its resolved outcomes (LTC: 12/41 = 29.3%, because all 41 of its logged rows have resolved), so both readings are on the card and neither has to be inferred. A neutral outcome is “price did not move past the threshold in either direction”, which is not a loss; that is why it is excluded from the rate rather than folded into it, and why it is never hidden.
Every rate carries a profit factor — gross favourable moves ÷ gross adverse moves, computed over exactly the same directional calls as the win rate, at the same horizon that produced each outcome, pre-fee. A profit factor cannot hide a denominator the way a rate can, and it is the number that says whether a high win rate was worth anything. Its own sample size (n=) is printed with it, and a thin sample is tagged on the payoff line as loudly as on the rate.
This is a different measure from the flagship confirmation card at the top of the page, which collapses a live call that is re-confirmed every few minutes into a single episode (a >12h gap starts a new one) and reports distinct calls. Both numbers are correct for what they measure; they are not interchangeable, so they are never blended.
The Today toggle and the Direction / Confidence buttons change the scope, never the population. With no filter the cards are the server’s own performance_log aggregation; with a filter they are re-counted in your browser from the signal feed. That feed also carries confirmation calls and regime-flip rows, which are a different record — the size of that splice is counted from the payload this page loaded, and is written here once the feed arrives. Those rows are filtered out before any card is built, and the line under the heading says which scope you are looking at and how many rows it counted.
Low totals are not a retention bug. Nothing is pruned. A symbol like LTC shows a low count because the engine has not fired on it recently, not because history was deleted — its logged rows all still exist and are all counted here.
deriv040 — live derivatives account tracked in real time
| # | Symbol | Dir | Entry | Exit | PnL | Duration | Exit Reason |
|---|---|---|---|---|---|---|---|
| 1 | BTC (example) | LONG | — | — | — | — | Sample — live trades load on open |
Our system monitors — major exchanges (Bybit, Binance, Hyperliquid) across — symbols in real time, tracking derivatives data, — whale wallets, funding rates, open interest, and liquidation levels. Every signal passes a 5-factor confirmation pipeline before publication.
Every 5 minutes, our engine scans — symbols for whale consensus regime flips. Each detected flip passes through a 5-factor confirmation pipeline: cross-asset alignment (BTC/ETH agreement), derivatives confirmation (funding rates, long/short ratios, OI momentum), price zone analysis (premium vs discount within 24h range), multi-snapshot consistency, and whale momentum acceleration. Only signals scoring above 0.45 confidence are published. Each signal includes direction (LONG/SHORT), a confidence score from 0.0 to 1.0, and entry price.
We auto-discover and monitor — whale wallets from Hyperliquid — the largest on-chain perpetual DEX. See what smart money is actually doing: position sizes, entry/exit points, which tokens they're accumulating, and when they're reducing exposure. This isn't aggregated data — it's individual wallet-level intelligence updated in real time.
Full derivatives screener for — symbols with sortable columns: open interest (absolute + 1h/24h change), funding rates across exchanges, long/short ratios, and top trader sentiment. Funding rate heatmap shows at a glance where leverage is building. OI rankings reveal which assets have the most speculative interest right now.
Every signal is tracked against actual Binance spot prices at 1h, 4h, 12h, and 24h intervals. These are live-tracked measurements, not backtested-only claims. The accuracy you see on this page is computed from real signals issued in real time, evaluated against real market data. You can click any signal to see its full outcome breakdown at each time window.
40+ API endpoints covering signals, derivatives, whale data, on-chain metrics (TVL, stablecoins, DEX volumes, gas), options flow (BTC/ETH PCR, max pain, OI by strike), ETF flows (BTC/ETH daily fund-level breakdown), and market indices. WebSocket streaming for real-time updates. Python and JavaScript SDKs. Integrate into your trading bot, dashboard, or research pipeline in minutes.
DeFi TVL tracking, stablecoin supply and flow analysis, DEX volume monitoring, BTC hash rate and difficulty, Ethereum gas prices, mempool congestion, and DeFi yield comparisons — all from a single API. See where liquidity is moving across chains and protocols. Identify capital rotation before it shows up on price charts.
Before entering any position, check if smart money signals agree with your thesis. A high-confidence signal (0.65+) means 5 independent factors confirmed the direction. A medium signal (0.45-0.64) provides partial support. Use the confidence score to gauge conviction strength.
Connect the API to your trading bot via REST or WebSocket. Filter signals by confidence score threshold and symbol. Execute trades automatically when conditions are met. Each signal already passed 5 confirmation gates — set your own minimum confidence (e.g. 0.60+) for additional selectivity.
Monitor liquidation levels, funding rates, and OI changes across your portfolio. Low confidence signals indicate weak multi-factor confirmation. Track stablecoin flows and exchange reserves as early warning indicators for market-wide risk events.
Track what the biggest wallets on Hyperliquid are doing in real time. See position openings, size increases, and exits. When multiple whales accumulate the same asset simultaneously, it's often a leading indicator of significant price movement.
Build custom dashboards combining on-chain data, derivatives metrics, and signal performance. Analyze funding rate arbitrage opportunities across exchanges. Study correlations between whale behavior, OI changes, and subsequent price action over historical data.
Monitor BTC and ETH options data including put/call ratio, max pain, and OI by strike. Track daily ETF inflows/outflows per fund with cumulative tracking and streak indicators. Understand institutional positioning through regulated product flows.
Visualize liquidation clusters across exchanges. Identify price levels where large cascading liquidations could trigger. Use this to set better stop-losses, avoid crowded liquidation zones, and anticipate sudden volatility spikes.
Combine long/short ratios, funding rates, and social sentiment to gauge market mood. When extreme greed meets high funding rates and leveraged longs, it's often a reversal signal. Our multi-factor confidence system quantifies this across 5 independent data sources.
Every 5 minutes, the Smart Money engine scans — symbols for whale consensus regime flips — moments when the short/long ratio shifts violently against the 7-day median. Each raw detection then passes through a 5-gate confirmation pipeline: (1) cross-asset check (BTC/ETH must not strongly contradict), (2) derivatives alignment (funding rate, global and top-trader LSR, OI momentum over 1h), (3) price zone filter (premium/discount within 24h range), (4) multi-snapshot consistency (4-6 whale snapshots moving monotonically), and (5) whale momentum acceleration. Signals below 0.45 confidence are rejected. Only confirmed signals are published with a direction, confidence score (0.0 to 1.0), and entry price.
Every signal is automatically tracked against Binance spot prices at four fixed intervals after issuance: 1 hour, 4 hours, 12 hours, and 24 hours. A LONG signal is marked correct if price is higher at the measured window; a SHORT signal is marked correct if price is lower. There is no manual curation — all outcomes are computed automatically and published as-is, including losses.
Win rate for each symbol and timeframe is computed as correct outcomes divided by total resolved signals for that window. A signal is unresolved until the measurement window has passed. The system caps recent signals at 2000 to keep the feed performant. Historical data beyond the displayed period is archived and available for download on Pro tier accounts.
1-hour accuracy is noisiest — even a correct directional signal can show a temporary opposite move within the first hour due to natural volatility. High 1h accuracy means the signal is capturing very immediate momentum shifts.
4-hour accuracy is the most actionable window for discretionary traders. It's long enough for a setup to play out with meaningful price displacement, but short enough that the original market regime — the conditions the signal was generated under — hasn't fundamentally changed. If you use signals for trade entries, prioritize symbols with high 4h win rates.
12h and 24h accuracy reflects the signal's ability to predict broader trend continuation. These windows are most relevant for swing traders and for evaluating whether the underlying thesis (derivatives positioning, whale accumulation) has lasting predictive value.
A win rate alone is not sufficient to evaluate a signal. A system that is right 60% of the time but loses twice as much on losers as it gains on winners has negative expected value. Use the accuracy percentage alongside the average return figures to estimate the quality of each signal category.
HIGH confidence signals (0.65+) are issued less frequently because they require all 5 confirmation gates to score strongly: cross-asset agreement, derivatives alignment, favorable price zone, consistent whale snapshots, and momentum acceleration. MEDIUM signals (0.45-0.64) passed the minimum threshold but with weaker multi-factor support. You should generally expect HIGH signals to carry higher win rates.
The per-symbol accuracy chart lets you spot which assets the system models well. Assets with consistently high accuracy across multiple timeframes are strong candidates for systematic trading. Assets with low accuracy despite sufficient sample size may have microstructure characteristics that resist the current signal methodology.
Every 5 minutes the engine scans — symbols for whale-consensus regime flips — moments when the aggregate long/short ratio shifts sharply against the 7-day median. Each candidate then passes a 5-factor confirmation pipeline before it is published:
Signals that do not reach a composite confidence score of 0.45 or above are rejected and never published. Only confirmed signals are recorded, stamped with a direction (LONG / SHORT), a confidence score from 0.0 to 1.0, and the market price at the moment of issuance.
All outcomes are tracked against Binance spot prices, captured automatically at four fixed intervals after each signal:
No prices are curated or adjusted after the fact. Price captures run on a fixed schedule and are stored as-is, including during volatile periods.
Each evaluation window applies a small deadband threshold to filter out micro-noise:
Win rate = wins ÷ (wins + losses) on fully resolved signals. Neutral outcomes are excluded from the denominator so they neither inflate nor deflate the rate.
A win rate computed from 2 or 3 signals carries no statistical meaning — yet it can look like 100 % or 0 %. To prevent misleading headlines, symbols and horizons with fewer than 10 directional outcomes — correct + wrong, the win rate’s own denominator, not the larger correct + wrong + neutral resolved count — are flagged as low-sample ("thin data" badge) and are excluded from the headline summary cards. They are still visible in the per-symbol grid with a muted colour so you can see the raw count, but they are not surfaced as "best" or "worst" performers.
The numbers shown on this page represent historical directional accuracy of the confirmation-signal pipeline, not trading profit or loss. Fees, slippage, position sizing, and execution timing are not modelled.
Honest baseline
Real system-wide win rates across all periods and confidence levels have measured in the range of approximately 50–54 % — only modestly above a coin flip. High-confidence signals (0.65+) tend toward the upper end of this range; medium-confidence signals (0.45–0.64) toward the lower end. Do not assume the latest short-window figure represents a durable edge.
This is not financial advice. Past performance does not guarantee future results.
All accuracy statistics shown here reflect real signals generated and tracked in real time. However, past signal performance does not guarantee future accuracy. Market regimes change — a signal methodology that works well in trending conditions may underperform in ranging markets. Always evaluate recent performance (last 7d or 30d) alongside all-time statistics.
Accuracy statistics are only meaningful with sufficient sample size. Symbols with fewer than 20 resolved signals in the selected period should be treated with caution — small sample sizes produce large statistical variance in win rate. Focus on symbols with 50+ signals per window for reliable accuracy estimates. The system continues logging signals as long as derivatives data is available, so sample sizes grow over time.
Signal accuracy is measured against Binance spot prices and does not account for slippage, trading fees, or the difficulty of executing at the exact entry price shown. Actual realized returns from trading these signals will differ from theoretical accuracy-based estimates. A signal being "correct" means price moved in the right direction, not that a specific trade would have been profitable after costs.
Disclaimer: Signal performance data is published for transparency and informational purposes only. It does not constitute financial advice or a solicitation to trade. Past accuracy rates are not indicative of future results. Crypto markets are highly volatile and carry substantial risk of loss. Always apply your own risk management rules, use stop-losses, and trade only with capital you can afford to lose entirely.
Free tier available — no credit card required.