Tomjosetj31

"Reading in the factory the books that eventually helped me get out of it" — that's a hell of a backstory for a side project, and it shows. The best tools come from a real, specific frustration rather than a market gap someone identified on a spreadsheet. The 0.86 inversion value being tuned for hours of actual reading rather than just looking right in a demo is exactly that kind of detail.

The instinct to keep the system "small in concept" is the right one, and it's harder to maintain than it sounds — most CMS projects start focused and then bloat themselves into irrelevance trying to be everything. The tell will be how you handle feature requests that are genuinely useful but push against that philosophy. What's already been the hardest "no" to say?

Impressive result on HLE if the methodology holds up. One thing I'd want to understand better: how much of the gain comes from the entropy weighting specifically vs. simply having more compute via parallel inference? Would be curious to see an ablation — same models, same budget, but with naive majority voting instead. That would isolate the actual contribution of your confidence-weighting approach.