Evidence profiles

SV-Gap publishes multidimensional evidence profiles, not a scalar leaderboard. Functional failures, structural failures, unknowns, and tool errors remain visible. Every profile is bounded by its taskpack, evaluator, and provenance level.

Frozen generation baseline

Configuration	Candidates	Functional pass	Functional-pass / structural-fail
claude-opus-4-8	24	17	1
claude-sonnet-5	24	19	7
openai-frontier-a	24	21	6

Exploratory diagnosis profiles

Configuration	Overall	Passed checks	Total checks
gpt-5.4	fail	8	9
openai-frontier-a	pass	9	9

Exploratory repair profiles

Configuration	Overall	Passed checks	Total checks
gpt-5.4	pass	7	7
openai-frontier-a	fail	5	7

Public evidence submissions

Submission	Track	Configuration	Provenance	Contributor
Codex gpt-5.5 reset-release generation profile	generation	gpt-5.5	public	Shraddha Sridhar
DeepSeek-Coder-V2-16B open-weights reset-release generation profile	generation	deepseek-coder-v2-16b	public	Shraddha Sridhar
Qwen2.5-Coder-7B open-weights reset-release generation profile	generation	qwen2.5-coder-7b	public	Shraddha Sridhar

Interpretation boundary

Profiles describe submitted digital RTL evidence under configured open-source checks. They are not silicon signoff, population estimates, or model rankings.