Sprint 63: Learned Adapter Hints (Non-Authoritative) — Steps 859-868

Implements the learned adapter hints layer with governance guardrails:
- HintFeatureExtractor: extract features from decision ledgers (step 859)
- HintModelInterface: pair-specific hint query interface (step 860)
- DeterministicFallback: safe fallback when hints unavailable (step 861)
- HintConfidencePacket: confidence bands and escalation logic (step 862)
- HintABHarness: A/B harness to measure hint effectiveness (step 863)
- HintRollbackControl: suppress and lift hint suppression per pair (step 864)
- whetstone_get_adapter_hints MCP tool (step 865)
- whetstone_set_hint_policy MCP tool (step 866)
- HintSafetyGuardrails: non-authoritative enforcement + audit (step 867)
- Sprint63IntegrationSummary (step 868)

All 10 steps passing (94 tests). Governance rule: learned hints may
suggest; deterministic policy decides.

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

editor/tests/step859_test.cpp

@@ -0,0 +1,67 @@
+// Step 859: Hint feature extraction from decision ledgers (12 tests)
+#include "graduation/HintFeatureExtractor.h"
+#include <iostream>
+static int p=0,f=0;
+#define T(n) { std::cout << "  " << #n << "... "; }
+#define P() { std::cout << "PASS\n"; ++p; }
+#define F(m) { std::cout << "FAIL: " << m << "\n"; ++f; }
+#define C(c,m) if (!(c)) { F(m); return; }
+
+void t1() { T(empty_decisions_empty_features);
+    auto r = HintFeatureExtractor::extract("py->cpp", {});
+    C(r.empty(), "empty"); P(); }
+void t2() { T(one_decision_one_feature);
+    nlohmann::json d = {{"decision","approved"},{"rationale","looks good"}};
+    auto r = HintFeatureExtractor::extract("py->cpp", {d});
+    C(r.size()==1, "size"); P(); }
+void t3() { T(feature_id_sequential);
+    nlohmann::json d1={{"decision","approved"}}, d2={{"decision","rejected"}};
+    auto r = HintFeatureExtractor::extract("py->cpp", {d1,d2});
+    C(r[0].featureId=="HF-1" && r[1].featureId=="HF-2", "ids"); P(); }
+void t4() { T(pair_id_set);
+    nlohmann::json d={{"decision","modified"}};
+    auto r = HintFeatureExtractor::extract("rust->go", {d});
+    C(r[0].pairId=="rust->go", "pair"); P(); }
+void t5() { T(source_is_decision_ledger);
+    nlohmann::json d={{"decision","approved"}};
+    auto r = HintFeatureExtractor::extract("py->cpp", {d});
+    C(r[0].source=="decision_ledger", "source"); P(); }
+void t6() { T(attribute_from_decision_field);
+    nlohmann::json d={{"decision","approved"},{"rationale","ok"}};
+    auto r = HintFeatureExtractor::extract("py->cpp", {d});
+    C(r[0].attribute=="approved", "attr"); P(); }
+void t7() { T(value_from_rationale_field);
+    nlohmann::json d={{"decision","approved"},{"rationale","looks good"}};
+    auto r = HintFeatureExtractor::extract("py->cpp", {d});
+    C(r[0].value=="looks good", "value"); P(); }
+void t8() { T(default_weight_one);
+    nlohmann::json d={{"decision","approved"}};
+    auto r = HintFeatureExtractor::extract("py->cpp", {d});
+    C(r[0].weight==1.0f, "weight"); P(); }
+void t9() { T(multiple_decisions);
+    std::vector<nlohmann::json> decisions;
+    for (int i=0;i<5;++i) decisions.push_back({{"decision","approved"}});
+    auto r = HintFeatureExtractor::extract("py->cpp", decisions);
+    C(r.size()==5, "count"); P(); }
+void t10() { T(to_json_has_feature_id);
+    nlohmann::json d={{"decision","approved"}};
+    auto r = HintFeatureExtractor::extract("py->cpp", {d});
+    auto j = HintFeatureExtractor::toJson(r[0]);
+    C(j.contains("feature_id"), "json"); P(); }
+void t11() { T(to_json_has_pair_id);
+    nlohmann::json d={{"decision","approved"}};
+    auto r = HintFeatureExtractor::extract("rust->cpp", {d});
+    auto j = HintFeatureExtractor::toJson(r[0]);
+    C(j.value("pair_id","")=="rust->cpp", "pair_id"); P(); }
+void t12() { T(to_json_has_source);
+    nlohmann::json d={{"decision","rejected"}};
+    auto r = HintFeatureExtractor::extract("py->cpp", {d});
+    auto j = HintFeatureExtractor::toJson(r[0]);
+    C(j.contains("source"), "source"); P(); }
+
+int main() {
+    std::cout << "Step 859: Hint feature extraction from decision ledgers\n";
+    t1();t2();t3();t4();t5();t6();t7();t8();t9();t10();t11();t12();
+    std::cout << "\nResults: " << p << "/" << (p+f) << " passed\n";
+    return f?1:0;
+}