test: establish recognition challenge v2

docs/pack-maturity.md

@@ -21,8 +21,10 @@ holds approved support policy; GitHub issues hold planned work.
   Electromagnetism, thermodynamics/heat transfer, fluid mechanics, calculus,
   discrete mathematics, and optimization remain deliberately narrow slices;
   evaluated capability evidence is not a field-completeness claim.
-- The independent frozen challenge passes 24 of 24 cases across binding,
-  constraints, packs, presentation, resolution, and syntax.
+- The independent frozen challenge v2 contains 48 cases across binding,
+  constraints, packs, presentation, resolution, and syntax. Its initial
+  hard-cutover baseline passes 41 of 48; the seven exposed gaps are assigned to
+  issues #189 and #191 rather than hidden by fixture or threshold changes.
 
 These numbers describe the synthetic benchmark, not real-world prevalence or
 field completeness.

docs/public-api.md

@@ -1,6 +1,6 @@
 # Public API
 
-Semath is a library and language-service runtime. Protocol 4 is a deliberate
+Semath is a library and language-service runtime. Protocol 6 is a deliberate
 hard cutover to a small meaning-first API.
 
 | Export | Responsibility |
@@ -33,7 +33,7 @@ refusal explanation when appropriate. `context.assumptions` contains explicit,
 source-linked assumptions with their subjects; omission means none were
 established. Parser ASTs and legacy rewrite queries are not public.
 
-Protocol 4 identifies every `RoleInfo` by its open, pack-qualified `conceptId`.
+Protocol 6 identifies every `RoleInfo` by its open, pack-qualified `conceptId`.
 There is no closed role enum or unnamespaced compatibility field. Included-file
 role, shape, and quantity facts use the same records and retain their original
 evidence.

docs/semantic-quality-scorecards.md

@@ -33,10 +33,12 @@ assumption, evidence, refusal, and scope independently instead of blending them
 into one recognition number.
 
 The frozen recognition challenge is separate from those development and
-diversity fixtures. Its cases are manually authored and grouped by the layer
-that owns a failure: syntax, binding, constraint, pack, resolution, or
-presentation. Association, structure, constraint, recognition, evidence,
-refusal, scope, and navigation remain separate metrics. Default CI validates
+diversity fixtures. Version 2 has 48 manually authored cases, including twelve
+positive/refusal semantic boundary pairs, grouped by the layer that owns a
+failure: syntax, binding, constraint, pack, resolution, or presentation. Exact
+IDs and normalized document sources must not occur in development or foundation
+fixtures. Association, structure, constraint, recognition, evidence, refusal,
+scope, and navigation remain separate metrics. Default CI validates
 the challenge schema and coverage matrix through pure tests but does not execute
 the engine over the holdout.
 Evaluated laws require