Some edges can be quantified easily, and others really can't be. One canonical example of the former is the bid-offer imbalance. That's because autocorrelation

Some edges can be quantified easily, and others really can't be. One canonical example of the former is the bid-offer imbalance. That's because autocorrelation is low, variability is high, the signal can be encoded easily, the distribution is bell-shaped, the heteroskedasticity is low, and we have a tremendous amount of historical data to work with. TA exists on the very opposite side of this gulf, where its profound difficulty to study using the traditional statistics/ML toolbox is conflated with no edge by many dogmatists in the industry. It's unbelievably difficult to come up with a good encoding of what's visually obvious to a human, and beyond that such encodings are very difficult to backtest because of the typical nasties (autocorrelated data and so forth). In my strong opinion, the failure of TA is more a failure of the quant industry at present. I believe they will figure out how to monetize that edge, but it's beyond the scope of their tools and capabilities at present.