Fixed critical inference regression, restored throughput — 7.9M tokens

Fixed critical inference regression, restored throughput

qwen3-30b-in-c

DEEP WORK

First session on qwen3-30b-in-c

First build tracked for this project

7.9Mtokens

42prompts

33:39:59time

6,481lines

Claude Codeclaude-sonnet-5C$87.52

mostly building

Fixed loop order bug that broke cache sharing, restoring inference throughput from 9 to 46.87 tokens per second.

Tokens / prompt187.0k

Cost / line$0.014

Cache hit98%

Burn rate$2.60/hr

Derived from this session's token and cost data. Not shown on the feed.

Building 55%Verifying 4%Thinking 3%Delegating 38%