What kept bothering me on OpenGradient was not TEE.
No. Worse.
It was ZKML path everybody ignored until OpenGradient review got expensive.
That's bad.
OpenGradient Vanilla is cheap.
TEE is practical. Alright.
ZKML is what everybody suddenly respects right after they picked something lighter and let OpenGradient panel move.
I’ve seen that choice look reasonable right up to bad hour.
Builder picks TEE because model is bigger. Because latency budget is real. nobody wants one OpenGradient inference turning into a small war about proof cost. Fine. Inference nodes return result. Review panel fills in. Queue moves. Full nodes can settle later. Good. Great even.
I hate how reasonable that choice looks.
Then somebody opens settlement trace.
Now risk wants OpenGradient settlement record. compliance wants to know why this OpenGradient path was attested and not proved.
Now the old "TEE was enough" line starts sounding a little elegant for what it bought.
That's the split.
$OPG Inference nodes do the fast part.
Full nodes settle later.
Lovely.
That's OpenGradient doing what it said. Fine. That irritates me.Inference nodes keep hot path moving.Full nodes settle later. Was always the deal. OpenGradient clears panel on one clock. The heavier proof path sits on another. Later, obviously.
Nobody minds that OpenGradient split until review does.No.they don't.
That's where it slips.
ZKML is too heavy for hot path right up until the OpenGradient result touches money.Or access? Or one calm panel clear that isn't calm anymore.
cheaper path ages badly.
I've watched that logic rot in real time.
OpenGradient Vanilla looks thin.
TEE starts looking practical the wrong way. Useful. Until review.
suddenly everybody wants mathematical certainty like it was free.Nice.
It wasn't.why they didn't pick it.
So TEE becomes normal OpenGradient answer. ZKML is too heavy for live path. Good.
I'm not sure teams are choosing an OpenGradient path anymore. Not really.
Might be choosing which later @OpenGradient trace fight they can live with?
Or worse.maybe.