feat: sources wired as recency-gated injectables; transitive closure, NOT scoping, recompile-scope uninstall; targeted parse errors

BEHAVES AS transitive now emits bounded multi_hop configs (direct checks and
evidence references), fixing a silent no-op. NOT builds keep _subjectAsObject
scoping so unary predicates negate the right node, and value-typed evidence
objects gate by exact edge value. Recompiling a scope uninstalls its stale
relation configs (compileMultiple coexistence preserved). Sources become
injectable relations honored by requiredFacts with a within-X recency gate.
Duplicate definition fields and three common declaration mistakes (within on a
fact, two BEHAVES clauses, limit on a non-pattern body) now produce targeted
errors. Provider edges referencing unknown nodes are warned and dropped.

tests/DSLRuntimeTyping.test.js

@@ -93,4 +93,44 @@ describe('DSLRuntime typing', () => {
     rt.registerFact('owns', async () => ({ possibility: 2.0 }));
     await assert.rejects(() => rt.check('u:1', 'can_read', 'doc:9'), /invalid possibility/);
   });
+
+  it('enforces a literal value in evidence as an exact edge-value gate', async () => {
+    const rt = new DSLRuntime(new Arbiter()).compile(`
+      definition Employee { id: string? }
+      fact balance(user: Employee, amount: number)
+      evidence can_afford(user: Employee) { balance(user, 5) }
+    `, 'rt-expected-value');
+    rt.addNode('u:1', 'Employee', {});
+    // An edge carrying amount 3 must NOT satisfy balance(user, 5) — without the
+    // gate every balance edge matched regardless of amount (silent over-grant).
+    rt.addRelation('u:1', 'balance', 'u:1', { possibility: 1.0, value: 3 });
+    const denied = await rt.check('u:1', 'can_afford', 'u:1');
+    assert.equal(denied.possibility, 0, 'value-3 edge must not satisfy balance(user, 5)');
+    rt.addRelation('u:1', 'balance', 'u:1', { possibility: 0.8, value: 5 });
+    const granted = await rt.check('u:1', 'can_afford', 'u:1');
+    assert.equal(granted.possibility, 0.8, 'value-5 edge must satisfy balance(user, 5)');
+  });
+
+  it('treats a value-typed evidence OBJECT param as the expected edge value', async () => {
+    const rt = new DSLRuntime(new Arbiter()).compile(`
+      definition Employee { id: string? }
+      fact balance(user: Employee, amount: number)
+      evidence can_withdraw(user: Employee, amount: number) { balance(user, amount) }
+    `, 'rt-value-object');
+    rt.addNode('u:1', 'Employee', {});
+    rt.addRelation('u:1', 'balance', 'u:1', { possibility: 0.9, value: 5 });
+    // The check object is the VALUE, not a node key: grant only on exact match.
+    const granted = await rt.check('u:1', 'can_withdraw', 5);
+    assert.equal(granted.possibility, 0.9, 'value-5 check must match the value-5 edge');
+    const denied = await rt.check('u:1', 'can_withdraw', 3);
+    assert.equal(denied.possibility, 0, 'value-3 check must not match the value-5 edge');
+    // Re-checking the granted value must not hit a value-3 cache entry.
+    const again = await rt.check('u:1', 'can_withdraw', 5);
+    assert.equal(again.possibility, 0.9, 'value-5 re-check must not be served the value-3 result');
+    // Direct fact check with a value object works the same way.
+    const direct = await rt.check('u:1', 'balance', 5);
+    assert.equal(direct.possibility, 0.9, 'direct balance(user, 5) must match the value-5 edge');
+    // A non-scalar object for a value-typed param is rejected loudly.
+    await assert.rejects(() => rt.check('u:1', 'can_withdraw', 'u:1'), /must be number/);
+  });
 });