Transparent method
Every profile separates popularity, evidence proximity, review depth, mechanism coverage, and time. Each answers a different question.
Directness, 0 to 3: conceptual analogy through target-like evidence.
Horizon, 0 to 3: not observed through longitudinal evidence.
The visible order sorts directness first, horizon second, and source ID only as a stable tie-breaker. It is not a quality verdict.
Global rank: observed GitHub stars across the reviewed set.
Category rank: observed stars within one of 12 broader implementation families.
Selection lens: the more specific mechanism that justified inclusion. It is descriptive, not another rank.
Stars are dated audience attention, not quality, safety, or effectiveness.
Mechanisms are marked visible, partial, not observed, or out of scope against a pinned revision. Visible means the reviewed evidence directly exposed the mechanism. Partial means the evidence exposed only part of it. Evidence depth states what was actually inspected. A README screen is not a code audit.
Each profile shows its last reviewed date and pinned source. The current ledger contains one verified review snapshot per repository. Future reviews will append dated changes rather than rewrite history.
Claim IDs connect repositories and research to bounded propositions. Each claim records support, limits, a null case, and the next test that could change the conclusion.
Read all claims in plain EnglishNo composite score certifies a repository. Source count is coverage, not confidence. Feature visibility does not prove collaboration benefit, authority quality, recovery, or lower cost.