feat(consult): six-field evidence-card record in agent observability (log only)

After each natal turn the route logs a separate [agent-observability] event
{runId, agentVersion, evidenceCard: {domains, cardVersion, cardChars,
cardTokenEstimate, citedFieldIds, feedback}}. Cited field ids follow the
research R5 rule (ISO date, degree, planet-in-sign phrase of 8+ chars
appearing verbatim in a finished answer); matched text is discarded. Thumbs
are client state only today, so feedback is "none"; storing thumbs per turn
needs a table and is left to a follow-up. The strict schema has no user,
session, question or answer field.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017eEAG8HD3mm8gsKXgk8uU8

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

@@ -62,6 +62,7 @@ import {
   createWindowConsultationAgentContext,
   precomputeWindowConsultation,
   windowPrecomputedPacketMessage,
+  consultationEvidenceCardTelemetry,
   consultationModelStepTelemetry,
   consultationStepBudgetReceipt,
   createConsultationRuntimeHooks,
@@ -879,6 +880,10 @@ export async function POST(request: Request) {
     let firstActivityMs = -1;
     let firstTextMs = -1;
     let logged = false;
+    // The finished natal answer, read once by the evidence-card record to find
+    // which card fields it quoted; never logged. A failed or cut run leaves it
+    // empty, so only finished answers count citations.
+    let answerForTelemetry = "";
     const markFirstActivity = () => { if (firstActivityMs < 0) firstActivityMs = Date.now() - agentStartedAt; };
     const markFirstText = () => { if (firstTextMs < 0) firstTextMs = Date.now() - agentStartedAt; };
     const consultEntrypoint = parsed.success ? parsed.data.entrypoint : undefined;
@@ -948,6 +953,15 @@ export async function POST(request: Request) {
         themeCoverage: state.workflowReceipt?.domains ?? [consultationTheme],
         billingSettlementResult: settlementResult,
       });
+      // D7: a separate six-field record, no session or user id in it.
+      const evidenceCard = consultationEvidenceCardTelemetry(state, answerForTelemetry);
+      if (evidenceCard) {
+        logAgentObservability({
+          runId: requestId,
+          agentVersion: "consultation-evidence-card-v1",
+          evidenceCard,
+        });
+      }
     };
     const settleRun = async (
       action: () => Promise<AgentSettlementResult>,
@@ -1385,15 +1399,18 @@ export async function POST(request: Request) {
       headers: { "x-jyotish-birth-time-mode": consultationMode },
       onFirstActivity: markFirstActivity,
       onFirstOutput: markFirstText,
-      onComplete: (output, agentExecutionReceipt, thinkingText, thinkingSections) => settleRun(() => completeResponse(
-        output,
-        mergeUsage(usages),
-        state.techniqueTruth ?? "unknown",
-        state.workflowReceipt ?? workflowReceipt,
-        agentExecutionReceipt,
-        thinkingText,
-        thinkingSections,
-      ), undefined),
+      onComplete: (output, agentExecutionReceipt, thinkingText, thinkingSections) => {
+        answerForTelemetry = output;
+        return settleRun(() => completeResponse(
+          output,
+          mergeUsage(usages),
+          state.techniqueTruth ?? "unknown",
+          state.workflowReceipt ?? workflowReceipt,
+          agentExecutionReceipt,
+          thinkingText,
+          thinkingSections,
+        ), undefined);
+      },
       onError: (error) => settleRun(
         cancel,
         toAgentObservabilityErrorCode(error),

frontend/src/lib/agent-observability.ts

@@ -113,6 +113,33 @@ export const agentObservabilityContractPhaseSchema = z.object({
   status: z.enum(agentObservabilityStepStatuses),
 }).strict().readonly();
 
+/**
+ * The evidence-card record (TASK-consult-evidence-card-20260927 D7): exactly
+ * six fields, numbers / enums / card field names only. No question, answer,
+ * birth data, name, email, user or session id, and never the matched text.
+ */
+export const EVIDENCE_CARD_TELEMETRY_FIELDS = [
+  "domains",
+  "cardVersion",
+  "cardChars",
+  "cardTokenEstimate",
+  "citedFieldIds",
+  "feedback",
+] as const;
+
+export const evidenceCardTelemetrySchema = z.object({
+  domains: z.array(consultationDomainSchema).min(1).max(2),
+  cardVersion: z.literal("evidence-card-v1"),
+  cardChars: countSchema,
+  cardTokenEstimate: countSchema,
+  citedFieldIds: z.array(z.string().min(1).max(120).regex(/^[A-Za-z][A-Za-z0-9._-]*$/, "invalid card field id")).max(64),
+  // Thumbs are client state only today (not stored), so a turn is logged
+  // with `none`; storing thumbs per turn needs a table and is a separate task.
+  feedback: z.enum(["up", "down", "none"]),
+}).strict().readonly();
+
+export type EvidenceCardTelemetry = z.infer<typeof evidenceCardTelemetrySchema>;
+
 export const agentObservabilityEventSchema = z.object({
   runId: opaqueIdSchema.optional(),
   requestId: opaqueIdSchema.optional(),
@@ -155,6 +182,7 @@ export const agentObservabilityEventSchema = z.object({
   reportJobDurationMs: durationMsSchema.optional(),
   reportJobPeakMemoryBytes: z.number().int().min(0).max(Number.MAX_SAFE_INTEGER).optional(),
   billingSettlementResult: z.enum(billingSettlementResults).optional(),
+  evidenceCard: evidenceCardTelemetrySchema.optional(),
 }).strict().refine(
   (event) => Boolean(event.runId || event.requestId || event.sessionId || event.caseId),
   { message: "at least one controlled identifier is required" },

frontend/src/mastra/consultation-tools.ts

@@ -14,12 +14,14 @@ import { runV9RangeReading } from "../lib/rectification-agentic/v9/engine-client
 import { createConsultationPlan, type ConsultationPlan } from "../lib/consultation-plan.ts";
 import type { TechniqueAuditRow, WorkflowReceipt } from "../lib/consultation-agent-events.ts";
 import { normalizeTechniqueAuditRows } from "../lib/consultation-technique-audit.ts";
-import type { AgentModelFinishReason } from "../lib/agent-observability.ts";
+import type { AgentModelFinishReason, EvidenceCardTelemetry } from "../lib/agent-observability.ts";
 import { agentGenerationSettings, AGENT_SLICE_ANSWER_OUTPUT_TOKENS, AGENT_SLICE_THINKING_OUTPUT_TOKENS } from "../lib/agent-generation-settings.ts";
 import { chartCalculationProgressLabel, evidenceLookupActivityLabel } from "../lib/consultation-activity-labels.ts";
 import {
   buildEvidenceCard,
+  citedEvidenceCardFields,
   EVIDENCE_LOOKUP_SECTIONS,
+  evidenceCardTokenEstimate,
   type EvidenceCard,
   type EvidenceLookupSection,
 } from "../lib/consultation-evidence-card.ts";
@@ -338,6 +340,27 @@ export function consultationModelStepTelemetry(state: ConsultationRuntimeState)
   };
 }
 
+/**
+ * The six-field evidence-card record for the observability log (D7). The
+ * answer is read only to find which card fields it quoted (ISO date, degree,
+ * planet-in-sign phrase); neither the answer nor the matched text is kept.
+ */
+export function consultationEvidenceCardTelemetry(
+  state: ConsultationRuntimeState,
+  answer: string,
+): EvidenceCardTelemetry | undefined {
+  const card = state.evidenceCard;
+  if (!card || state.evidenceCardChars === undefined) return undefined;
+  return {
+    domains: card.domains.slice(0, 2),
+    cardVersion: card.card_version,
+    cardChars: state.evidenceCardChars,
+    cardTokenEstimate: evidenceCardTokenEstimate(state.evidenceCardChars),
+    citedFieldIds: citedEvidenceCardFields(card, answer),
+    feedback: "none",
+  };
+}
+
 /**
  * Public receipts carry only the fields the client contract allows. Building
  * the list from an explicit allowlist keeps internal diagnostics, such as the

frontend/tests/consult-evidence-card-telemetry-20260927.test.ts

@@ -0,0 +1,91 @@
+// TASK-consult-evidence-card-20260927 T6 (D7): per-turn evidence-card record
+// in [agent-observability], six fields only, no user text.
+import assert from "node:assert/strict";
+import { readFileSync } from "node:fs";
+import test from "node:test";
+
+import {
+  createAgentObservabilityLogger,
+  EVIDENCE_CARD_TELEMETRY_FIELDS,
+  evidenceCardTelemetrySchema,
+  type AgentObservabilityEvent,
+} from "../src/lib/agent-observability.ts";
+import {
+  consultationEvidenceCardTelemetry,
+  createConsultationRuntimeState,
+  createConsultationTools,
+} from "../src/mastra/consultation-tools.ts";
+
+type Json = Record<string, unknown>;
+const golden = JSON.parse(readFileSync(
+  new URL("./fixtures/consult-evidence-card-golden.json", import.meta.url),
+  "utf8",
+)) as { charts: Array<{ id: string; workflow: Json }> };
+const workflow = golden.charts[0]!.workflow;
+const route = readFileSync(new URL("../src/app/api/consult/route.ts", import.meta.url), "utf8");
+
+async function stateAfterCalculation() {
+  const state = createConsultationRuntimeState();
+  const tools = createConsultationTools({
+    userId: "user-telemetry", sessionId: "session-telemetry", requestId: "req-telemetry", consultationMode: "verified_chart",
+    serverChart: { name: "public", toolInput: { year: 2000, month: 1, day: 1, hour: 0, minute: 0, city: "fixture", lat: 0, lon: 0, tz: 0, ayanamsa: "raman" }, truth: { birthTimeSource: "reported" } } as never,
+    state,
+    runWorkflow: async () => structuredClone(workflow) as never,
+  });
+  await tools["run-jyotish-consultation"].execute!({ question: "q", domains: ["parents"] } as never, { writer: { custom: async () => {} } } as never);
+  return state;
+}
+
+test("the record has exactly the six approved fields", async () => {
+  assert.deepEqual([...EVIDENCE_CARD_TELEMETRY_FIELDS].sort(), Object.keys(evidenceCardTelemetrySchema.unwrap().shape).sort());
+  assert.deepEqual([...EVIDENCE_CARD_TELEMETRY_FIELDS], ["domains", "cardVersion", "cardChars", "cardTokenEstimate", "citedFieldIds", "feedback"]);
+  const state = await stateAfterCalculation();
+  const record = consultationEvidenceCardTelemetry(state, "");
+  assert.ok(record);
+  assert.deepEqual(Object.keys(record).sort(), [...EVIDENCE_CARD_TELEMETRY_FIELDS].sort());
+  assert.deepEqual(record.domains, ["parents"]);
+  assert.equal(record.cardVersion, "evidence-card-v1");
+  assert.equal(record.cardChars, state.evidenceCardChars);
+  assert.equal(record.cardTokenEstimate, Math.round(record.cardChars / 3.5));
+  assert.deepEqual(record.citedFieldIds, []);
+  assert.equal(record.feedback, "none");
+});
+
+test("the schema is strict: no extra field, no free text in field ids", () => {
+  const valid = { domains: ["parents"], cardVersion: "evidence-card-v1", cardChars: 10, cardTokenEstimate: 3, citedFieldIds: ["base.natal.ascendant.sign"], feedback: "up" };
+  assert.equal(evidenceCardTelemetrySchema.safeParse(valid).success, true);
+  for (const extra of ["question", "answer", "sessionId", "userId", "birthDate", "matched"]) {
+    assert.equal(evidenceCardTelemetrySchema.safeParse({ ...valid, [extra]: "x" }).success, false, extra);
+  }
+  assert.equal(evidenceCardTelemetrySchema.safeParse({ ...valid, citedFieldIds: ["我和父母关系如何"] }).success, false);
+  assert.equal(evidenceCardTelemetrySchema.safeParse({ ...valid, feedback: "great" }).success, false);
+});
+
+test("the logged line keeps field ids and drops every word of the user's text", async () => {
+  const state = await stateAfterCalculation();
+  const card = state.evidenceCard!;
+  const pdStart = String((card.base.timing.vimshottari.pratyantardasha as Json).start);
+  // Fictional sensitive values, not a real person.
+  const secrets = ["虚构名字王小明", "fictional.person@example.invalid", "1990-01-01", "我和父母关系如何"];
+  const answer = `${secrets[0]}（${secrets[1]}，生日 ${secrets[2]}）问：${secrets[3]}。子运从 ${pdStart} 开始。`;
+  const lines: AgentObservabilityEvent[] = [];
+  const log = createAgentObservabilityLogger((event) => lines.push(event));
+  log({ runId: "req-telemetry", agentVersion: "consultation-evidence-card-v1", evidenceCard: consultationEvidenceCardTelemetry(state, answer) });
+  const text = JSON.stringify(lines);
+  for (const secret of secrets) assert.equal(text.includes(secret), false, secret);
+  assert.equal(text.includes(pdStart), false, "the matched date is not kept");
+  assert.equal(text.includes("session-telemetry"), false);
+  assert.equal(text.includes("user-telemetry"), false);
+  assert.ok(lines[0]?.evidenceCard?.citedFieldIds.includes("base.timing.vimshottari.pratyantardasha.start"));
+});
+
+test("the consult route logs the record once per natal turn, without session or user id", () => {
+  assert.match(route, /const evidenceCard = consultationEvidenceCardTelemetry\(state, answerForTelemetry\);/);
+  const block = route.slice(route.indexOf("const evidenceCard = consultationEvidenceCardTelemetry"), route.indexOf("const settleRun = async"));
+  assert.match(block, /agentVersion: "consultation-evidence-card-v1"/);
+  assert.doesNotMatch(block, /sessionId|userId/);
+  const natal = route.slice(route.indexOf("const agent = getJyotishAgent(selectedModel, agentContext);"));
+  assert.match(natal, /answerForTelemetry = output;[\s\S]*return settleRun\(\(\) => completeResponse\(/);
+  // Only a finished answer counts citations; a failed run logs the card size with no citations.
+  assert.doesNotMatch(natal, /onError: \(error, _emitted, output\)/);
+});