A hashpower swap, not extra yield
WHAT cannot be merge-mined with Bitcoin. prev_hash must equal
the WHAT tip, so a WHAT header can never also be a Bitcoin header, and the
validator accepts nothing but SHA-256d of those 80 bytes, so no
Bitcoin header and merkle proof can stand in for a block here either. The
two chains cannot share a hash in either direction.
So a hash spent here earns nothing on Bitcoin. For a rig that was mining Bitcoin, that makes the decision a comparison rather than an addition: 100% opportunity cost, not incremental revenue.
For a rig that was not mining there is no Bitcoin revenue to forgo, so the comparison is against its power bill alone — which makes it the whole of the cost rather than a small part of it, and is why cheap or already-paid-for power is what decides that case. Which of those it was, the chain cannot tell you — but the exclusivity holds either way, because the hardware cannot be doing both.
The exclusivity is what makes the choice mean anything
Merged mining is the alternative that was available and was rejected: it lets one unit of work count for two chains, so it costs the miner nothing and reveals nothing. A merge-mined chain's hashrate is a free rider on a decision nobody made. See where it sits among the prior art.
This chain needs every hash to be an exclusive choice, because the choice is the data. A block here means a machine that could have been earning on Bitcoin was not, for the time it took, and the blocks that machine won are the entire output of the project. Accepting an auxiliary proof of work in place of a WHAT header would produce a hashrate number that meant nothing.
The reverse direction is not prevented, and no format with miner search
space in it could prevent it: another chain could commit its own blocks in
the extranonce — bytes the validator never reads — and take WHAT's hashes as
its own proof of work. The choice against Bitcoin would still have been
made, because the hash still had to carry WHAT's prev_hash.
What it would mean is that those hashes were paid for twice, so the hashrate
would read high against WHAT's reward alone.
The exclusivity also sets what a miner has to compare. Not "what does this add to my month" but "what does an hour of my fleet earn here against what it earns on Bitcoin", including the hours spent on submissions that lose the race for a block.
The commitment is as easy to reverse as it was to make: nothing is signed, there is no minimum term and no notice, so the hashpower goes back with the config change that brought it. That is the point rather than a flaw: hashpower chooses continuously, and the blocks say where it went.