trashhalo

I built Bosun to address a specific issue: an agent’s memory expands as a knowledge graph, and without a mechanism to evaluate each new edge—whether it’s supported, non-redundant, or still true—the graph deteriorates into noise that overwhelms the model attempting to read it. Since no cost-effective method existed for that step, we developed one.

It’s a LoRA fine-tune of Qwen3-Reranker (0.6B and a 4B). You provide it with an instruction and two findings; it returns sigmoid(logit_yes − logit_no) ∈ [0,1]. “Warranted” isn’t a fixed rule, so you program it per graph with a sentence, and it generalizes to rules it hasn’t been trained on. The same structure applies to RAG filtering, deduplication, and moderation—the graph is just where it initially posed a problem for us.

We also open-sourced WarrantBench for evaluation, as FollowIR only includes relevance instructions. An honest limitation is that every rule it judges today is symmetric—it doesn’t handle direction (“A causes B”) yet; that’s our next step.

Weights: https://huggingface.co/Hanno-Labs/bosun-xs · Bosun-4B: https://huggingface.co/Hanno-Labs/bosun-4b · WarrantBench: https://github.com/Hanno-Labs/warrantbench · writeup: https://hannolabs.ai/field-notes/introducing-bosun. Feel free to ask any questions.

To give you a sense of it atlassian's current title is:

> Collaboration software for software, IT and business teams

The classifier gives that a 14/100

Rating every tag that showed up in `moonshotai/kimi-k2.6` thinking tokens the best one it could come up with was:

> Every Connection. Every Team. One Graph.

The classifier gives that a 83/100

I keep wanting to be hyped about elixer but the performance benchmarks confuse me. If you look at the tech emperor benchmarks[1] phoenix makes the list at #46 registering 16% of the performance of the top framework. I'm willing to sacrifice performance for readable maintainable code but it just surprises me that its that slow.

Anyone know whats going on there?

1. https://www.techempower.com/benchmarks/#section=data-r17&hw=...

Totally agree on the project discovery bit. Noticed the same thing when I played with it. I added a couple tickets to try and encourage them to fix it.

* Increase the prominence of gitlab hosted projects on about.gitlab. Project search, trending, curated lists. - https://gitlab.com/gitlab-com/www-gitlab-com/issues/558

* Trending project data by language (js, ruby etc) - https://gitlab.com/gitlab-org/gitlab-ce/issues/13475

* Allow projects to flag themselves as looking for contributors - https://gitlab.com/gitlab-org/gitlab-ce/issues/13502

Give them a +1 maybe they will put resources on making gitlab more oss friendly