fix(consult): check the evidence gate against the route the answer is on

七条路由的证据门都不是自己的：route_requirements 的键写成 relationship/finance，而
路由名是 marriage/wealth，另有 5 条路由压根没有条目，全部静默落到 general 的门。
missingLayers: [] 因此不表示证据齐备，只表示没检查过——婚姻的 UL 与财富的 D2 从未
进入检查。

同一函数另有两处判据也没接到权威来源。7 块正则用问题文本重猜领域，而领域早已由模型
声明并写进 route_packet，一句写作「情感」而非表里「感情」的提问在 marriage 路由上完全
拿不到性别解读边界。timing_layers_ready 读的是 missing_route_layers，该列表只装本路由
要求的层，于是对任何不要求 narayana_dasha 的路由恒为真，精确应期在该层根本没算出来时
也照样放行。三处的失败方向都是静默放宽，因此没有任何人报错。

三处都接回权威来源：10 条路由逐条显式列出必需层（层名限定为证据包真实构建的 section，
所以 wealth 不要求引擎不产出的 D11）、领域边界按 route 查表、出生时间边界从矫正闸门的
effective_accuracy 与 Lagna 敏感度派生、就绪判断直接读 section 状态。唯一保留文本探测
的是「用户有没有要一个具体日期」——服务端对此没有权威来源，改为 timing/annual 路由结构
性携带、文本仅作叠加，一次措辞漏判不再能把信号清零。

另外把 skillReferenceReadCount 暴露为回执的 skill.referenceReads（必填）与可观测日志的
skillReferenceReads。它此前数完即丢，而 skill_read 按设计不记成 runtime step，因此「模型
有没有真的翻开方法文档」在运行结束后无处可查。

Co-authored-by: Cursor <cursoragent@cursor.com>

frontend/src/app/api/consult/route.ts

@@ -681,7 +681,11 @@ export async function POST(request: Request) {
       const executionReceipt = (): AgentExecutionReceipt => ({
         runId: requestId,
         runtime: "mastra-agentic",
-        skill: { name: "jyotish-vedic-astrology", loaded: state.jyotishSkillLoaded },
+        skill: {
+          name: "jyotish-vedic-astrology",
+          loaded: state.jyotishSkillLoaded,
+          referenceReads: state.skillReferenceReadCount,
+        },
         steps: publicConsultationRuntimeSteps(state),
         stepBudget: consultationStepBudgetReceipt(state),
         workflow: workflowReceipt,
@@ -745,7 +749,11 @@ export async function POST(request: Request) {
     const executionReceipt = (): AgentExecutionReceipt => ({
       runId: requestId,
       runtime: "mastra-agentic",
-      skill: { name: "jyotish-vedic-astrology", loaded: state.jyotishSkillLoaded },
+      skill: {
+        name: "jyotish-vedic-astrology",
+        loaded: state.jyotishSkillLoaded,
+        referenceReads: state.skillReferenceReadCount,
+      },
       steps: publicConsultationRuntimeSteps(state),
       stepBudget: consultationStepBudgetReceipt(state),
       workflow: state.workflowReceipt ?? workflowReceipt,

frontend/src/lib/agent-observability.ts

@@ -132,6 +132,9 @@ export const agentObservabilityEventSchema = z.object({
   // every attempt. Both are enum-like machine values, never provider text.
   modelFinishReason: z.enum(agentModelFinishReasons).optional(),
   modelStepCount: countSchema.optional(),
+  // How many reference documents the model opened after loading the skill. Zero on a run that
+  // answered a domain question means the method was never consulted, which no other field shows.
+  skillReferenceReads: countSchema.optional(),
 
   inputTokens: tokenCountSchema.optional(),
   outputTokens: tokenCountSchema.optional(),

frontend/src/lib/consultation-agent-events.ts

@@ -51,6 +51,10 @@ export const agentExecutionReceiptSchema = z.object({
     name: z.literal("jyotish-vedic-astrology"),
     loaded: z.boolean(),
     version: z.string().max(120).optional(),
+    // How many reference documents the model opened after loading the skill. Required rather than
+    // optional: the count was tracked in runtime state and surfaced nowhere, so "did the model
+    // consult the method at all" was unanswerable from a finished run. Zero is a real answer.
+    referenceReads: z.number().int().min(0).max(64),
   }).strict(),
   steps: z.array(executionStepSchema).max(32),
   stepBudget: stepBudgetSchema.optional(),

frontend/src/mastra/consultation-tools.ts

@@ -123,6 +123,7 @@ export function createConsultationRuntimeState(options: { plannedSteps?: number;
 export function consultationModelStepTelemetry(state: ConsultationRuntimeState) {
   return {
     modelStepCount: state.modelStepCount,
+    skillReferenceReads: state.skillReferenceReadCount,
     ...(state.modelFinishReason === undefined ? {} : { modelFinishReason: state.modelFinishReason }),
   };
 }