The honest limits of onchain analysis

I would rather you trust this tool a little less and understand it a lot more. Onchain analysis is genuinely useful, but the space is full of people selling it as a crystal ball, and it is not one. Here is where it falls short, including the version I built.

The edge is real but modest

When I backtest the score properly, using independent windows rather than overlapping samples that inflate the numbers, higher readings really have led to better forward returns on average. But the effect is modest, not overwhelming. A bullish reading tilts the odds. It does not stack the deck. Anyone showing you an onchain signal with a suspiciously clean track record is either overfitting or hiding the windows that did not work.

Proxies are proxies

Some of the cleanest concepts, spent output profit and miner position, need paid, UTXO level data to compute properly. The free versions are approximations, and the tool labels them as such. They add colour, but I do not lean on them the way I lean on the metrics built from real, direct data.

Regimes change

Every metric here is measured against its recent normal. That works beautifully until normal itself shifts, during a structural change in the market or the network. In those moments the baseline lags reality and the readings can mislead until they recalibrate. This is not a bug I can fix, it is a property of measuring anything against its own history.

It cannot see everything

Onchain data misses whatever happens off chain. Exchange internal ledgers, over the counter deals, regulation, and pure narrative can all move price in ways no flow will show you in advance. That is exactly why the tool sits alongside a separate macro read and a sentiment read rather than pretending to be the whole picture. Treat the score as one honest input among several, and it will serve you well. Treat it as gospel, and the market will teach you the lesson itself.