What we tested
and learned

A technical entry on hourly candles (momentum, mean-reversion, or a regime-aware blend) can pick BTC turning points often enough to pay for itself after fees and slippage.

Under the tested risk geometry and crypto-perp cost assumptions, the five headline entries are net-negative. With zero modeled cost, three turn gross-positive in this sample. The assumed roughly 0.4% round trip is larger than the observed edge. One positive modeled-cost cell in the 108-cell edge map, classic momentum with wide 4R targets on H1, reached 0.03 to 0.05% per trade but was too thin to support deployment. Final verdict: this registered test found no deployable single-asset OHLCV timing edge.

If a hidden-Markov model labels the market as trend, volatile, or range in real time, routing each entry only into the regime that suits it recovers an edge the blended signal buries.

The regime model is honest about itself: its states separate volatility, not direction, so there is no trend premium to route toward. On paper the gate fails. The one routed cell with an adequate sample, classic momentum in the trend regime, is negative on all three symbols. Every cell that prints positive is thin, under 30 trades, and the positives disagree across symbols. Not a trustworthy edge.

Test whether a slow 28-day daily-trend rule survives the published cost assumptions across ten majors.

The signal-flip config reads +24.92% on average, but two launch-era single-asset flukes carry it. Strip SOL and AVAX and the typical major loses 5.25%. Four of ten symbols are positive, below the six-of-ten bar, and fold stability is 38% with most fold cells too thin to trust. The geometry-exit variants are flatly negative. No configuration clears the deployable bar.

Stop timing price. Harvest the structural funding premium: short the perp, hold the spot, collect the funding, stay delta-neutral. This is the strongest raw crypto edge on record.

In this backtest, always-on BTC returns +39.50% over 6.75 unlevered years, with 0.73% modeled max drawdown and all four folds positive. It fails the registered magnitude gate: full-window yield is 5.84% per year versus an 8% threshold, and the recent 24 months produce 2.35% per year. The observed premium compresses over the registered window. Under the same assumptions, adding a timing overlay (A2) or rotation (A3) worsens the result.

Rank the 30-major universe by trailing return each week and hold the top five. Rotation should beat passively holding the same basket, or it is only churn.

The full-window +1325% modeled result exceeds the basket at +759%, but the excess is concentrated in one fold, the 2020 to 2021 launch run measured over today’s surviving 30. Fold stability fails, two of four. In the bias-mitigated subwindow the long-only simulation returns −46.99% against a basket at −50.29%, with one of four folds positive. The long-short subwindow PASS is reported as the frozen gate computed it, then set aside because its modeled result loses 12.15%; cash is the more relevant benchmark for a market-neutral book.

A daily close-above-EMA200 long/flat gate on BTC and ETH, with size scaled by the existing HMM structural-regime classifier, can improve drawdown-adjusted return after modeled costs. Buy-and-hold remained the raw-return benchmark; absolute-return outperformance was explicitly not pre-claimed.

The 50/50 vol-targeted portfolio did not beat buy-and-hold on CAGR: 22.8% versus 28.1%. Vol targeting improved modeled drawdown-adjusted results over the full window: Calmar 0.61 versus 0.32, Sharpe 0.87 versus 0.71, and max drawdown 37.4% versus 86.5%. The study did not establish uniform raw-return outperformance across both assets: the BTC vol-targeted sleeve lagged hold, while the ETH sleeve exceeded it, and the plain EMA200 gate had isolated raw-CAGR wins. HMM sizing underperformed the hold benchmark on drawdown-adjusted metrics: 50/50 Calmar was 0.21 versus 0.32 for hold, and its frequent exposure changes accumulated the highest modeled cost. These are sandbox-only OHLCV outcomes with modeled spot costs, not live results, not broker fills, and not a trading recommendation.

Two hypotheses were written down before any query ran, both restating this project’s founder’s own brief, which the registered spec quotes: do not follow the mass, and trade the trending chart rather than the sideways one. First, that signals entered in trending regimes carry materially better net expectancy than signals entered in sideways regimes, so filtering to trending charts would turn the published record positive. Second, that the crowd loses, so inverting every signal would be profitable. The detector thresholds, the reconciliation tolerances and the decision rule were fixed in the same commit as the hypotheses, before the data was read. Publishing this entry was not pre-registered: the spec listed site publication as out of scope, to be decided once results existed.

No regime bucket was net-positive under any detector. Trend alignment did not even improve gross expectancy over counter-trend, +0.0979R against +0.0965R at the ADX 20 cut, and under the efficiency-ratio cut the trend-aligned bucket was gross-negative at −0.0871R, on 102 trades, below this study’s 300-trade bar for a conclusive cell. Inverting every signal produced a 62.6% win rate and a worse net result, −0.5207R against −0.5019R, because flipping the direction flips the returns and keeps the cost. What the study leaves standing is the cost geometry: modeled cost per trade is round-trip cost divided by stop width, so the published stream’s gross edge of +0.0094R sits roughly 54 times below its 0.5113R cost wall, and that wall shrinks only as stop width and holding period grow. That is an analytic rescale of the cost term alone: win and loss distributions at wider stops were not simulated and are not knowable from this dataset, so the wider-stop figures are a lower bound on what a trade must clear, not a projected result. Horizon, not filtering, is the lever the data leaves open. The regime results cover crypto only, 1,162 of 3,157 trades; 1,995 trades across 20 non-crypto pairs have no daily candle coverage in this repository, so their entry context cannot be independently re-derived here, and their outcomes remain counted, resolved trades. These are observed-OHLCV outcomes under modeled cost assumptions, not broker fills, and not a trading recommendation. No strategy was activated or deactivated by this study.

The inherited close-above-EMA200 long/flat rule, paired with a deliberately wide fixed stop, can remain net-positive after modeled production crypto-perpetual costs and match or beat buy-and-hold Calmar in at least three of four continuous-state folds without breaching the registered frequency ceiling.

On the exact frozen historical sample and modeled production crypto-perpetual costs, the pre-registered rule passed its build gate. The fixed 50/50 result beat buy-and-hold on both net return and full-window Calmar, while the registered Calmar fold rule passed exactly three of four folds, not all four. The result is an OHLCV simulation with modeled fills and fixed funding, and the BTC/ETH-only universe carries survivor-selection risk. It is not live performance, not broker fills, and not a trading recommendation. The owner separately approved promotion to a live tracked signal lane on 2026-08-09. That approval does not enable broker order execution, does not backfill historical live rows, and does not bypass the existing fail-closed broadcast evidence gate.