MVRV on BTC: 8-year walk-forward + held-out validation, full data

Hello. Long-time lurker, first technical post. Posting here because this
is one of the few places where people will actually read a falsification
report instead of just the conclusion.

Context

Over the last six months I tested five families of crypto trading
strategies on the same protocol: walk-forward folds plus a held-out
selection test. The held-out test asks whether a rule like "pick the
parameter combo that worked best on the oldest fold" actually beats
random selection on later folds. Most backtests don't pass this, because
the visible edge dissolves once you subtract the implicit cost of having
chosen parameters by looking at the data first.

Four of the five families failed. One survived. I'm sharing the survivor
and the full report so people can push back on the methodology.

The four that died

Signal-based 15m on USDC majors (MACD/RSI/ADX, six different
architectures). Random pair selection beat the rule in 64.5% of
trials.
Same family on 4h and 1d. 73-131% OOS decay at 4h. 1d had only
4 trades in 4 years - too sparse to claim anything.
CNN walk-forward on 9 candidate architectures. Top 9 all failed
outside the training-adjacent fold. Best one lost $423 of $2k
notional on fold 2.
Stat-arb cointegration on 9 majors, 36 pairs, rolling
   Engle-Granger 365d, z-score reversion, 60bps fees:
   - Rule "fold 0 positive" (8 pairs): -5.86% on OOS folds
   - Random 1000 trials mean: -4.03%
   - Oracle (impossible hindsight): +1.07%
   Random beat the rule in 93.8% of trials. The rule was actively
   anti-predictive.

The one that survived

MVRV applied to BTC alone. Long when MVRV is undervalued, exit when
overvalued. Same exact thresholds in-sample and out-of-sample, never
re-tuned.

IS 2018-2021 best combo: CAGR +76%, Sharpe 1.62
OOS 2022-2026 same combo: CAGR +27%, max DD -36%, Sharpe 0.91, 1 trade
Full 2018-2026 strategy: CAGR +48%, max DD -36%
Full 2018-2026 buy & hold: CAGR +24%, max DD -81%
64/64 threshold combinations had positive CAGR
52/64 beat buy & hold

The edge is not return enhancement. Long BTC is hard to beat in CAGR
during a bull cycle. The edge is drawdown reduction by sitting
in cash roughly 60% of each cycle.

Things I want people on this forum specifically to push back on

N=3 trades in 8 years. Statistically thin in isolation. The
meaningful signal is the breadth: 52/64 combos beating B&H. But it's
still N=3 cycles, not 30.
BTC only. I tried the same MVRV approach on 8 altcoins.
5 lose money, 3 beat B&H only because B&H is terrible (DOT -31% vs
B&H -47% is not a strategy, it's a less-bad loss).
Cycle Sharpe is declining: 1.96 -> 1.81 -> 1.40. Spot ETF
   launched in 2024 and the on-chain behavioral pool has shifted -
   long-term holders selling differently, ETF flows muting reflexive
   distribution at tops. Realistic forward expectation should be CAGR
   +20-40%, DD -30-40%, Sharpe 1.0-1.4, not the IS Sharpe 1.62.
The free INTOTHEBLOCK feed is opaque about exactly how it
computes MVRV. CoinMetrics has a more rigorous CapMVRVCur but it's
behind a paid API. This is a known limitation.

What I shipped

A free Pine Script v6 indicator on GitHub
(MIT). It pulls INTOTHEBLOCK:BTC_MVRV, plots the validated thresholds,
colors the background by regime, and has three alerts on threshold
crosses. The README contains the full walk-forward report, the
multi-asset falsification table, and the cycle-decay disclosure.

What I deliberately did not ship

No "buy signal generator." No altcoin version (it was falsified). No
paid Discord. No premium tier with locked-up edge. None of those passed
the same test so I won't sell them.

Why I'm posting

Honestly: I want the methodology torn apart by people who have seen
every kind of cycle indicator since 2013. If MVRV regime models have a
known fatal flaw I haven't caught, I want to know now. If the held-out
validation protocol itself is too lenient or too strict, I want that
challenged too.

I will reply to technical pushback. I won't reply to "moon" / "scam" /
"price prediction" comments because they're not what the post is about.

Repo with the full report and indicator:
https://github.com/lonelyobserver0/mvrv-regime-monitor