Method
MarketSonar shows only what can be measured. Below is every rule you will see in the “Lights” — and the ones we tested and rejected. Data: BTC and ETH, 1H Bybit candles, 2 years (10.2024–10.2026), 0.13% cost per trade. Measurements are repeated quarterly; new rules ship only after measurement. Last update: 2026-10-29.
We use
Basis: own journal — 548 BTC/ETH positions (Sep–Oct 2026): entries with the 1D trend ≈ $0 per position (+$15 total), no trend −$2.0/pos (−$330), against −$1.9/pos (−$319). A discipline rule confirmed on real trades, not a market edge: after the measurement correction (23 Oct) a mechanical with-trend entry at any hour does not win (PF 0.87), and a “against 1D” ban is a coin toss.
13:00–15:30: half of large 1H moves are erased within 2 h (52%/50%). 16:00–19:00: volume ×3, continuation 53%, whipsaws 26%. 20:00–24:00: continuation 36–44%. Time of day gives no directional edge — it is a noise warning, not an entry signal.
Our own journal of 529 BTC/ETH trades (Sep–Oct 2026) confirms the 1D rule (all losses came from no-trend or counter-trend entries) and shows that positions held at a loss beyond one hour account for most of the losses. Rule: losing after 60 min → close.
Information only
A single 1D candle with a wick ≥50% of range and ≥0.5 ATR: +1.5 ATR bounce in 47% (any candle: 36%). Clusters of 2+ wicks — too few cases (n=11) to judge.
73 episodes — too few for a “no trade” rule; volatility after a squeeze is not higher than usual (1.9% vs 2.1%). Shown as information.
Tested and rejected
8 years of daily readings (since 2018). After extreme fear BTC did not rise more than usual (30 days: −0.7 pp vs average, ETH +1.0 pp, both within chance). Large differences, e.g. +9 pp after extreme greed, come from greed occurring in bull-market years; within a given year extreme greed did worse than average in 4 of 6 years. The index shows the market phase, the same information as trend. Kept as context.
Buy at the lower 1M/3M average, sell at the upper one, target before the opposite edge, R:R ≥ 1.5 (BTC+ETH, 5 years of 1H, closed periods only). Result PF 0.91, worse than the same rules on levels randomly shifted by 3–8% (PF 1.13). Weekly averages 0.80, daily 0.95. Monthly levels are not stronger than daily ones in the data (break in 45% vs 43% of tests).
The measurement scripts took 1D/1W/1M values from the period IN PROGRESS (its close known only later). Everything was recomputed on closed periods: the “with the 1D trend” and “⚡” edges disappeared; the own journal, time of day, time stop and the v3 paper trader (PF 1.60 → 1.45) survived. Full before/after report in the measurements repository. We publish this because the rule is one: we show only what can be measured — including our own mistakes.
The 21 Oct measurement (PF 1.23) computed the 1D trend from the in-progress daily candle — knowing how the day would end. Corrected (closed days only): PF 0.94 with the trend, 0.99 against, +0.02% after 24 h on average. As a mechanical strategy: PF 0.85–1.08. It stays on the chart and in e-mails only as the information “1H closed beyond a level”, with no claim of an edge.
The 39% vs 48% result of 21 Oct came from the same error (1D trend from the in-progress day). Corrected: with 43%, no trend 43%, against 42% — no effect. The only survivor: a level after one defence breaks less often (33–40% vs 46%), too little for a rule.
20 Bybit markets, up to 5 years of 1H. Bounce from the line with the 1D trend: portfolio PF 1.15, but all profit from 2022, 2023–2026 combined negative; longs from support PF 0.49. Line breakout: PF 0.83, −8% of capital. ATR 1M lines were never touched. Shuffled candles give a similar result — no edge. The bot was not built.
448 live-settled forecasts of our own model: zones forecast at 0–20% broke 70% of the time, those at 80–100% — also 70%. The number carried no information. Removed from the site, the script and the e-mails.
Daily, weekly and monthly levels break identically: 43% / 43% / 43%.
Entry at the close of the candle touching the level, stop 0.5 ATR 4H, target 1.5 R: PF 0.70–0.93 in every variant, market and year. The number of earlier holds of the level does not change the result.
Momentum alone: move in the banned direction −0.03% (= baseline). When momentum runs against 1D, price reverts toward 1D by +0.5% per day — the rule was harmful. Withdrawn after one day.
PF 0.86–0.87 vs baseline 0.84. When 1W disagrees with 1D, 1D wins (PF 1.07).
A fast move (≥1.5 ATR 15m in 2 candles) into a daily+ level, wick touches, candle closes back → entry against the move. Short: PF 0.43–0.55 (n=571), long: 0.49–0.87 (n=680); at RMA21 1D without a 1D trend, short PF 0.37. A fast move into a level more often starts a move than ends it.
Price right below a descending line (≤0.8%): longs gained +0.32% (1H line) / +0.51% (4H) per day on average, shorts lost (PF 0.62). Price pressing on a line more often breaks it than bounces — the ban worked backwards. Lines stay on the chart as context.
Phase trend↑ → long PF 0.89 (baseline 0.85), trend↓ → short PF 0.91. A weaker version of the 1D trend; adds nothing when they disagree (n=48–240). Kept as a description on the card, no effect on the verdict.
Not measured yet
First settlement (11 days, 436 readings): scenario direction right 49% of the time after 24 h, “confident” scenarios 46%. Withdrawn from the public forecast on 22.10.2026; still computed, back only if a ≥3-month measurement shows an edge.
No reliable historical data to measure it. Shown as context only, with no effect on verdicts.
Method: definitions fixed before each measurement; Wilson 95% confidence intervals; time split (year 1 / year 2) and per market; transaction costs included. Full tables and code available on request. Educational material — not investment advice.