Methodology

How cryptojargon.in computes what it shows: whale-flow netting, backtest rules, the setups record, funding intervals — and the negative results.

Whale flow

Trades of $50K+ ($75K for ETH, $100K for BTC) where 'buy' and 'sell' mean the aggressive taker side — who crossed the spread. On-chain exchange flow is netted per transaction, so an exchange shuffling coins between its own wallets never counts as flow; before netting, one venue alone showed 4.3M BTC of 90-day 'flow' that never left the building. Wallet labels are public attribution, not exchange-confirmed.

The negative results we publish

We backtested our own headline features. Whale-flow direction: no predictive edge in any signal we tested across 60 coins — activity predicts how far price travels, not which way. Lead-lag between coins: all 4,556 ordered pairs tested at 1h against shuffled controls; nothing survived beyond chance, so this site will never claim 'X follows Y'. A finding that dies in testing gets reported, not reworded.

Pattern backtests

Only mechanically-detectable patterns are tested — one that needs a human-drawn line cannot be backtested honestly. Detection on closed candles, entry at the NEXT open, signed forward returns versus an unconditional baseline and a 200-round shuffled control with matched instance counts. Target-vs-stop races count same-bar ties as stopped. Most results are nulls, and the pages say so; the fakeout rate is measured, not asserted.

The setups record

Every call the engine issues is resolved against price, losers included, and nothing that happened before a setup existed can resolve it. Every call the updated engine has closed is listed on the Radar with its coin, how it ended and the move in percent — the side and levels only for a funded partner account — and the running record is held back while it proves itself. Confidence tiers were retired on 5 Sep 2026: after 65 graded calls the high tier had won 18% against 25% for the rest, and the badge went as promised. How a setup is selected, and how it is settled, are the two things on this page we keep to ourselves — the record is what we ask to be judged on instead.

Funding, correlations, sectors

Funding is shown at each symbol's real interval — on 13 Sep 2026, 416 of Binance's 658 USDT perps funded every 4h or 1h, not 8h. Correlations are Pearson on 4h log returns over ~90 days, aligned on shared candles so newer listings can't poison the panel, shown on two windows because a relationship's direction of travel matters more than its level. Sector strength uses medians so one pumping coin can't drag a sector, and tokenised T-bill funds are excluded from directional reads.

What keeps it honest in production

Every push runs a typecheck, more than 600 tests (most of them named regressions of bugs this codebase actually shipped), and a build. Production is checked after every deploy, and every publish on our box must pass the same checks or it is rolled back. A watchdog on our own collector box alerts — and self-heals — when a data feed goes silent. Displayed numbers pass plausibility guards: a funding rate beyond 3% per interval renders as a dash and files an anomaly report instead of appearing as fact.

Crypto Jargon

Live numbers load with JavaScript and refresh every few minutes; the text above is the page at build time.