feat: remove ILP→QUBO from untrusted edges

ILP<bool>→QUBO overhead is now accurate (no slack ambiguity).

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: GiggleLiu

src/rules/analysis.rs

@@ -269,10 +269,9 @@ pub fn compare_overhead(
 
 /// Known reduction edges whose symbolic overhead is incomplete.
 ///
-/// ILP → QUBO: the implementation adds slack bits per inequality constraint
-/// whose count depends on coefficient magnitudes, but the symbolic overhead
-/// only records `num_vars = num_vars`.
-const UNTRUSTED_EDGES: &[(&str, &str)] = &[("ILP", "QUBO")];
+/// With `ILP<bool>`, the ILP → QUBO overhead is now accurate (no bounds
+/// ambiguity), so no edges need to be excluded from analysis.
+const UNTRUSTED_EDGES: &[(&str, &str)] = &[];
 
 /// Check whether every edge in a path has trustworthy symbolic overhead.
 fn path_is_symbolically_trustworthy(path: &ReductionPath) -> Result<(), String> {

src/unit_tests/rules/analysis.rs

@@ -260,23 +260,6 @@ fn test_find_dominated_rules_returns_known_set() {
     }
 }
 
-#[test]
-fn test_ilp_qubo_paths_are_unknown() {
-    let graph = ReductionGraph::new();
-    let (_, unknown) = find_dominated_rules(&graph);
-
-    // Any path through ILP → QUBO should be reported as Unknown
-    let ilp_qubo_unknowns: Vec<_> = unknown
-        .iter()
-        .filter(|u| u.reason.contains("ILP"))
-        .collect();
-
-    assert!(
-        !ilp_qubo_unknowns.is_empty(),
-        "Expected at least one Unknown comparison involving ILP -> QUBO"
-    );
-}
-
 #[test]
 fn test_no_duplicate_primitive_rules_per_variant_pair() {
     use crate::rules::registry::ReductionEntry;