A rule you can inspect.
An edge we must prove.
Personal research project · 16 October 2026 · strategy perp-pullback-1.0.0. Free access. No subscriptions, trading accounts or order execution. No validated win rate or profitability improvement.
What this is
A small technical signals board for Binance BTCUSDT, ETHUSDT and SOLUSDT USDⓈ-M perpetual contracts. Not a brokerage, continuous quote feed or portfolio manager. The strategy was developed with AI assistance; each scan runs deterministic rules. There is no continuously thinking AI analyst or verified news-catalyst feed.
The strategy hypothesis
Trend-following pullbacks on completed 15-minute candles, with completed 1-hour and 4-hour confirmation. Thresholds are engineering hypotheses, not calibrated probabilities. Extra filters can reduce the sample and overfit a strategy; complexity is not evidence of profitability.
| Gate | Rule |
|---|---|
| Source integrity | Active USDT-margined perpetual and tick size verified against Binance metadata. At least 200 valid, contiguous, aligned completed bars on each of 15m, 1h and 4h. In-progress bars excluded. |
| Trend regime | 1h EMA20 versus EMA50, with EMA20 sloping over three completed bars, must agree with 4h EMA20 versus EMA50. Long and short are symmetric. 1h Wilder ADX14 at least 20. |
| Trigger | EMA20 touched by one of the last two completed 15m candles; latest closes back on the trend side and beyond the previous candle’s high (long) or low (short). 15m EMA20 slopes in the same direction. |
| Participation | 15m Wilder RSI14: 48–68 long, 32–52 short. Volume at least 1.1 times the preceding 20-bar median. |
| Volatility | ATR14 between 0.15% and 2.5% of close. Trigger range no more than 2.5 ATR. Mark within 0.35 ATR of the trigger close. |
| Perpetual context | Mark/index and bid/ask updates no more than 120 seconds old. Spread at most 8 bps; absolute mark/index basis at most 0.5%. Latest adverse funding at most 0.03%; avoid the 10 minutes before next funding settlement. |
| Reference levels | Zone = trigger close ±0.1 ATR, rounded to ticks. Stop beyond the last six completed candles’ extreme plus 0.2 ATR, or 1.5 ATR from close, whichever is further. Risk from the worse end of the zone must be 1–3 ATR and at most 3% of reference. Target 2R from that worse entry. All levels must be positive. |
| Room and costs | Prior 48-bar extreme must be behind a breakout or at least 1.5R ahead. Require ≥1.5R after illustrative 6 bps fee + 4 bps adverse slippage per side and one adverse funding event. These are assumptions, not actual account fees or a complete funding model. |
| Expiry | Entry window ends 15 minutes after the trigger closes, even if the scan is late. No backdated calls. Skip after expiry, after price leaves the zone or after invalidation. The board does not monitor stop/target touches between checks. |
Risk protocol for evaluation
Start with paper observation, not live sizing. Before any real trade, check current price, spread, actual fees, funding, liquidation level and whether you can tolerate the entire planned loss. Leverage does not replace an invalidation level; stops do not guarantee fills. BTC, ETH and SOL are correlated, not three independent bets. Do not average into invalidated setups or increase size to recover losses.
Updates and outages
A signed-in owner can initiate a scan. With that tab open, the browser requests another check each quarter-hour after a candle-close grace period. A server lease allows one scan per window. Ordinary visitors only read saved results. A separate owner-connected cloud task is required for unattended updates; its connection and schedule must be verified independently. Always check the displayed scan time. Scheduling is best effort, not an execution service.
Failed requests, stale/gapped candles, wrong instruments, invalid quote/funding fields or clock mismatch suppress entries. No geographic restriction is bypassed. No spot, alternate-exchange or invented-price fallback. A snapshot older than 20 minutes cannot supply an active setup; each setup also has its own shorter expiry. This is not continuous position management.
Evidence before marketing
The ledger preserves original issue times, rules and levels. It is not a trade history or verified performance record: fills and wins/losses are not inferred. Expiry is not an outcome. The UI shows the latest 60 issues; older records remain stored. Zero-signal periods stay zero.
Before a success claim: freeze the version, use permitted historical data, predefine chronological training/holdout periods, account for fees, slippage, variable funding and ambiguous intrabar outcomes, compare a simple baseline, report all trials and drawdown, and forward-test out of sample. Win rate alone is insufficient: expectancy, uncertainty, sample size and risk matter. Claude’s independent review is requested, not yet received.
Access, privacy and rights
The free URL is public and unindexed, not password-private. Only non-personal derived research is stored. Do not submit exchange keys, positions, balances or trades. Owner sign-in authorizes the fixed data checker, not financial-account access. Existing sample-terminal journal records remain on their original browser; they are neither imported nor deleted.
No raw-price API, historical-candle export or continuous quote stream is offered. That does not establish an exemption from source terms or regulation. Commercial distribution and paid signals remain on hold pending the relevant rights and review. A “not financial advice” line is not a legal exemption.
Official Binance market-data API documentation ↗ · Surj privacy information ↗
No promise of profit
These experimental conditions may fail or produce no useful edge. Perpetuals can rapidly liquidate margin. You retain every decision; this app never places, closes or manages an order. Educational research, not personalized financial advice or guaranteed results.
Return to the board →