The $OPG usage rewards don’t feel as linear as they look on paper.
Ran OpenGradient across 2 wallets for 3 days, mostly stress-testing repeat inference instead of one-off prompts. Kept sessions small, around 7–9 prompts each, just to see if rewards actually track usage in a predictable way or drift once activity stacks up.
At low frequency, the numbers look clean enough — roughly 0.02 to 0.05 $OPG per interaction depending on timing. But once I crossed ~30–40 interactions in a day, the slope didn’t hold. Same activity pattern, different outputs. Not wildly different, but enough to notice when you’re tracking it closely.
Even repeated prompts weren’t stable. I reran identical queries twice and saw ~10–15% variance in reward output. That’s the part that changes behavior more than expected — I started spacing requests, batching them, even skipping obvious calls because they felt “less efficient” in reward terms.
That’s probably fine technically — distributed systems rarely behave cleanly at the edges — but it introduces this subtle second layer where you’re not just using the tool, you’re also estimating how the tool will score the usage.
Still early though. The idea of tying OPG directly to usage is interesting enough that the noise almost feels part of the experiment. Just not sure yet if the variability smooths out with scale or becomes something you learn to game around…
Right now it feels more like watching a meter flicker than reading a balance sheet — useful, but not yet something you trust for decisions.