Version Lineage

17 models trained, 8 teachers, 4 architectures, $1.76 total compute.

Ver	Teacher / Method	Heretic	keep_rate	Status
v2	— (base)	0.975	0.897	precision ceiling
v3	self-label, Q&A	0.942	0.728	first self-label
v4	self-label, domain	0.967	0.823	override internalized
v5	self-label, domain	0.961	—	converged
v6	generator, agent-dist	0.962	0.854	dead end
v7	sliding-window	0.956	0.868	dead end
v8	Qwen2.5-7B C3 (λ=3)	0.955	0.854	PRODUCTION
v9	Qwen2.5-7B, C3-only	0.921	—	overfit
v10	Qwen2.5-7B, scaled C3	0.947	0.891	diminishing
v11	Qwen2.5-7B, large enc	0.917	0.517	capacity ≠ precision
v12	Qwen3-Coder	0.949	0.949	too conservative
v13	regex, GLM	0.951	0.951	too conservative
v14	council, v8+GLM	0.882	—	proof-of-concept
v15	mixed (983 pairs)	0.878	—	data-scale regression
v16	Qwen2.5 C3 (λ=10)	0.972	0.972	best precision, 2.8% compression
v17	Qwen2.5 C3 (λ=5)	0.963	0.963	Pareto middle

Key finding

Label quality is the bottleneck, not model capacity or data quantity.