That view is no longer tenable
I've attentively followed every stray tidbit to cross my radar about the shadow sector since the publication of The Puzzle Palace, about the peripheral ghosts of which my algebra professor had direct experience.
The gold box agencies can do traffic analysis at scale. They can model metadata at scale. They can't break every damn cipher at scale—neither can they employ the rubber hose password-getter at large scale (the Soviets managed to cover about 10% of their population with blue welts over a thirty year period, but ultimately this did no favours to their economy).
The best approach to scaling crackers is to leak key bits in the purportedly pseudo-random number nonce stream. This is the hardest tampering to identify from the outside of a black box. Even when the black box is reverse engineered and one discovers that random is far from uniformly random (with no stray key-space correlates), some idiot applies Hanlon's razor: Never attribute to malice that which can be adequately explained by stupidity.
How about we agree to make a small exception for the industrial-scale tainting of purportedly random numbers, where discerning the difference between malice and stupidity achieves an elite level of algebraic epsilon? Oh, look, one digit in the source code for the random number generator has a wrong digit. Must have been a careless mistake—as if careless mistakes are a dime a dozen in the land where a poor man's nonce is a persistent agency's key-space collapsing back-channel.
The NSA does not randomly shoot holes in the protection of the American public. Worse than having no back door is having a back-door that somehow becomes shared with the wrong people. What they want is to inject a weakness that only they can exploit, even when their adversaries discover their handiwork.
Just off the top of my head, one way to achieve this is to require that exploiting the leak requires having the intercept history of the channel in hand since day one. The unfortunate flip side is that the specificity of these methods of single-party Achilles-heal exploitation becomes a smoking gun to the presence of a far-from-blind watch master. No ruse is totally perfect.
But you can always keep 90% of the population busy debating whether metadata has any value, such that any debate that makes any progress at all contains only those people who were already sophisticated cranks (recruitment/rubber-hose scale, to mention the carrot and stick). It all works out.
If scale matters, assigning a scant value to metadata can not be so much as trivially entertained by a thinking person. Pity we have so few.