AttiR · AttiR · Jul 2, 2026 · Jul 2, 2026 · Jul 2, 2026
diff --git a/.example.env b/.example.env
@@ -9,3 +9,7 @@ LANGFUSE_SECRET_KEY=
 LANGFUSE_HOST=https://cloud.langfuse.com
 EVINCTA_API_KEY=                   # required for write routes in production
 ALLOWED_ORIGINS=http://localhost:5173
+
+# web/ — set in Vercel (same value as EVINCTA_API_KEY for approve/override)
+# VITE_API_URL=http://localhost:8000
+# VITE_EVINCTA_API_KEY=
diff --git a/Dockerfile b/Dockerfile
@@ -10,7 +10,8 @@ RUN pip install --no-cache-dir -r requirements.txt
 
 COPY services/ services/
 COPY api/ api/
-# Demo claims (committed). Mount or bake data/generated/ for the full seeded set (issue #47).
+# Demo claims (committed). Pre-baked index survives Render redeploys.
+COPY data/demo_claims_index.json data/claims_index.json
 COPY data/samples/ data/samples/
 
 # NOTE: no .env, no secrets copied — injected at runtime by Render

diff --git a/README.md b/README.md
@@ -10,7 +10,7 @@ An evidence-backed recommendation — with its reasoning, precedents, and audit
 
 <br>
 
-[![Live demo](https://img.shields.io/badge/▶_Live_demo-Enter_as_demo_adjuster-7c3aed?style=for-the-badge)](https://YOUR-APP.vercel.app)
+[![Live demo](https://img.shields.io/badge/▶_Live_demo-Enter_as_demo_adjuster-7c3aed?style=for-the-badge)]([Evincta](https://evincta.vercel.app))
 &nbsp;
 [![Eval gate](https://img.shields.io/badge/CI-eval_gate_passing-16a34a?style=for-the-badge)](.github/workflows/eval.yml)
 
@@ -31,7 +31,7 @@ An evidence-backed recommendation — with its reasoning, precedents, and audit
 ---
 
 > [!NOTE]
-> **Live demo:** [YOUR-APP.vercel.app](https://YOUR-APP.vercel.app) → click **“Enter as demo adjuster.”**
+> **Live demo:** [https://evincta.vercel.app](https://evincta.vercel.app) → click **“Enter as demo adjuster.”**
 > The API runs on a free-tier container that sleeps when idle, so the **first load may take ~30s to wake** — after that it's fast. Browsing and deciding claims costs nothing (recommendations are pre-computed and cached).
 
 <div align="center">

diff --git a/api/auth.py b/api/auth.py
@@ -5,5 +5,7 @@
 
 
 def require_key(x_api_key: str = Header(None)):
-    if not API_KEY or x_api_key != API_KEY:
+    if not API_KEY:
+        return  # local dev — no key configured
+    if x_api_key != API_KEY:
         raise HTTPException(401, "invalid or missing API key")
diff --git a/api/claims.py b/api/claims.py
@@ -2,6 +2,7 @@
 from fastapi.concurrency import run_in_threadpool
 from pydantic import BaseModel
 from services.agents.run import start_claim, resume_claim
+from services.agents.audit import audit_node
 from services.ingest.extractor import extract_claim
 from fastapi import Depends
 from .auth import require_key
@@ -44,7 +45,8 @@ def _run():
 def list_claims():
     # compact rows for the queue view
     return [{"claim_id": c["claim_id"], "status": c["status"],
-             "decision": c["recommendation"]["decision"],
+             "recommendation": c["recommendation"]["decision"],
+             "human_decision": c.get("human_decision"),
              "fraud_risk": c["recommendation"]["fraud_risk"]}
 
             for c in store.all_claims()]
@@ -77,7 +79,14 @@ async def decide(claim_id: str, body: DecisionIn):
              else c["recommendation"]["decision"])
 
     def _resume():
-        resume_claim(c["thread_id"], human_decision=final, approver=body.approver)
+        try:
+            resume_claim(c["thread_id"], human_decision=final,
+                         approver=body.approver)
+        except Exception:
+            # baked demo index may not have a matching LangGraph checkpoint
+            audit_node({"claim_id": claim_id,
+                        "recommendation": c["recommendation"],
+                        "human_decision": final, "approver": body.approver})
     await run_in_threadpool(_resume)
 
     store.upsert(claim_id, status="decided",

diff --git a/data/demo_claims_index.json b/data/demo_claims_index.json
@@ -0,0 +1,114 @@
+{
+  "claim_0001": {
+    "thread_id": "3de5e4ee-ad20-4576-99e5-c13b37dc8590",
+    "status": "pending",
+    "recommendation": {
+      "decision": "approve",
+      "payout_low_usd": 6500.0,
+      "payout_high_usd": 26000.0,
+      "fraud_risk": "low",
+      "confidence": 0.83,
+      "rationale": "Applying adjudication rules in order: (1) Coverage is confirmed - covered is true, so no denial on coverage grounds. (2) Fraud risk is low with no flags, so no investigation required. (3) Cost midpoint ($17,250) exceeds deductible ($1,000), so no denial on deductible grounds. (4) Therefore, we approve the claim. Payout range is calculated as repair cost estimates ($7,500-$27,000) minus the $1,000 deductible, with the high end capped at the coverage limit of $26,000. Similar precedent claim_0012 involved vandalism and was approved with a comparable payout of $8,750.",
+      "cited_precedents": [
+        "claim_0012",
+        "claim_0013",
+        "claim_0023"
+      ],
+      "policy_basis": "Claim meets all approval criteria: incident is covered under policy, fraud risk is low, and estimated repair costs exceed the deductible amount. Payout is subject to $1,000 deductible and $26,000 coverage limit as specified in the policy terms."
+    },
+    "evidence": {
+      "incident_type": "collision",
+      "damaged_parts": [
+        "front end",
+        "hood",
+        "front driver side",
+        "windshield",
+        "front bumper",
+        "driver door"
+      ],
+      "severity": "severe",
+      "visible_pre_existing_damage": false,
+      "image_inconsistency": true,
+      "policy_number": "AUTO-42953",
+      "deductible_usd": 1000.0,
+      "coverage_limit_usd": 26000.0,
+      "notes_summary": "Insured reported vandalism at junction with severe damage. However, the photo evidence shows extensive collision damage consistent with a high-speed impact rather than vandalism. The front end is severely crushed, hood crumpled, windshield shattered, and driver door damaged - all typical of a collision. Adjuster notes mention 'severe dmg to unknown' suggesting uncertainty about the incident type. Discrepancy exists between claimed vandalism and visible collision damage.",
+      "extraction_confidence": 0.65
+    },
+    "claim_id": "claim_0001"
+  },
+  "claim_0000": {
+    "thread_id": "b9fcefa6-51b0-4e57-9e1c-f5436a4d11f0",
+    "status": "pending",
+    "recommendation": {
+      "decision": "approve",
+      "payout_low_usd": 80.0,
+      "payout_high_usd": 1400.0,
+      "fraud_risk": "low",
+      "confidence": 0.92,
+      "rationale": "Coverage is confirmed and the claim passes all checks. The cost midpoint ($990) exceeds the deductible ($250), satisfying the threshold requirement. Fraud risk is assessed as low with no flags present. All precedent cases with similar image characteristics were approved with no fraud issues. The estimated repair costs fall well within the policy coverage limit of $26,000. Payout range calculated as cost estimates ($330-$1,650) minus the $250 deductible.",
+      "cited_precedents": [
+        "claim_0000",
+        "claim_0014",
+        "claim_0032",
+        "claim_0072",
+        "claim_0062"
+      ],
+      "policy_basis": "Claim is covered under policy with $250 deductible and $26,000 coverage limit. Cost estimate midpoint of $990 exceeds deductible threshold. No fraud indicators or coverage exclusions apply."
+    },
+    "evidence": {
+      "incident_type": "theft",
+      "damaged_parts": [
+        "rear bumper",
+        "quarter panel"
+      ],
+      "severity": "minor",
+      "visible_pre_existing_damage": false,
+      "image_inconsistency": true,
+      "policy_number": "AUTO-74686",
+      "deductible_usd": 250.0,
+      "coverage_limit_usd": 26000.0,
+      "notes_summary": "Reporting party reports theft incident occurred on driveway with minor damage to unknown parts. Image shows blue vehicle with scrapes and damage to rear bumper/quarter panel area.",
+      "extraction_confidence": 0.65
+    },
+    "claim_id": "claim_0000"
+  },
+  "claim_0007": {
+    "thread_id": "af646410-16fa-4592-97df-b65693ad59a2",
+    "status": "pending",
+    "recommendation": {
+      "decision": "approve",
+      "payout_low_usd": 6750.0,
+      "payout_high_usd": 24950.0,
+      "fraud_risk": "low",
+      "confidence": 0.85,
+      "rationale": "Coverage is confirmed with a deductible of $250 and coverage limit of $26,000. The estimated repair cost ranges from $7,000 to $25,200 (midpoint $16,100), which exceeds the deductible threshold. Fraud risk is assessed as low with no flags identified. Following the decision rules in order: (1) coverage is confirmed as covered, (2) fraud risk is low (not medium or high), (3) the midpoint cost of $16,100 exceeds the $250 deductible. Therefore, the claim should be approved. Similar precedents show approved severe damage claims (claim_0059, claim_0006, claim_0012) with comparable severity patterns. Payout calculation: cost range minus $250 deductible = $6,750 to $24,950.",
+      "cited_precedents": [
+        "claim_0059",
+        "claim_0006",
+        "claim_0012",
+        "claim_0061"
+      ],
+      "policy_basis": "Claim meets coverage requirements under the policy. The incident is covered, fraud risk is low, and repair costs significantly exceed the deductible amount. Policy allows for approval with payout calculated as estimated costs less applicable deductible."
+    },
+    "evidence": {
+      "incident_type": "collision",
+      "damaged_parts": [
+        "rear end",
+        "trunk",
+        "rear bumper",
+        "rear quarter panel",
+        "roof"
+      ],
+      "severity": "severe",
+      "visible_pre_existing_damage": false,
+      "image_inconsistency": true,
+      "policy_number": "AUTO-41195",
+      "deductible_usd": 250.0,
+      "coverage_limit_usd": 26000.0,
+      "notes_summary": "Caller reports collision in car park with severe damage. No injuries reported. Image shows catastrophic damage to entire rear section and roof structure with complete collapse - inconsistent with typical car park collision.",
+      "extraction_confidence": 0.65
+    },
+    "claim_id": "claim_0007"
+  }
+}
diff --git a/web/src/ClaimDetail.tsx b/web/src/ClaimDetail.tsx
@@ -28,6 +28,7 @@ export default function ClaimDetail({
 }) {
   const [c, setC] = useState<any>(null);
   const [busy, setBusy] = useState(false);
+  const [err, setErr] = useState<string | null>(null);
 
   useEffect(() => {
     setC(null);
@@ -43,10 +44,13 @@ export default function ClaimDetail({
 
   async function act(decision: string, override_to?: string) {
     setBusy(true);
+    setErr(null);
     try {
       await api.decide(id, { decision, approver: "atti@evincta", override_to });
       await api.get(id).then(setC);
       onDecided();
+    } catch (e) {
+      setErr(e instanceof Error ? e.message : "Request failed");
     } finally {
       setBusy(false);
     }
@@ -62,7 +66,7 @@ export default function ClaimDetail({
           </h1>
           <p className="mt-0.5 text-sm capitalize text-zinc-500">{c.status}</p>
         </div>
-        <Badge decision={r.decision} />
+        <Badge decision={decided ? c.human_decision : "pending"} />
       </div>
 
       {/* recommendation */}
@@ -84,7 +88,7 @@ export default function ClaimDetail({
             </dd>
           </div>
           <div className="px-5 py-4">
-            <dt className={ui.label}>Decision</dt>
+            <dt className={ui.label}>Recommended</dt>
             <dd className="mt-1.5">
               <Badge decision={r.decision} />
             </dd>
@@ -185,21 +189,28 @@ export default function ClaimDetail({
           )}
         </div>
       ) : (
-        <div className="mt-6 flex items-center gap-3">
-          <button
-            disabled={busy}
-            onClick={() => act("approve")}
-            className={ui.btnPrimary}
-          >
-            Approve recommendation
-          </button>
-          <button
-            disabled={busy}
-            onClick={() => act("override", "deny")}
-            className={ui.btnSecondary}
-          >
-            Override → Deny
-          </button>
+        <div className="mt-6 space-y-3">
+          {err && (
+            <p className="rounded-md border border-rose-200 bg-rose-50 px-3 py-2 text-sm text-rose-700">
+              {err}
+            </p>
+          )}
+          <div className="flex items-center gap-3">
+            <button
+              disabled={busy}
+              onClick={() => act("approve")}
+              className={ui.btnPrimary}
+            >
+              Approve recommendation
+            </button>
+            <button
+              disabled={busy}
+              onClick={() => act("override", "deny")}
+              className={ui.btnSecondary}
+            >
+              Override → Deny
+            </button>
+          </div>
         </div>
       )}
     </div>

diff --git a/web/src/ClaimList.tsx b/web/src/ClaimList.tsx
@@ -5,7 +5,8 @@ import Badge from "./Badge";
 type Row = {
   claim_id: string;
   status: string;
-  decision: string;
+  recommendation: string;
+  human_decision?: string;
   fraud_risk: string;
 };
 
@@ -60,7 +61,13 @@ export default function ClaimList({
                 <span className="text-sm font-medium text-zinc-900">
                   {c.claim_id}
                 </span>
-                <Badge decision={c.decision} />
+                <Badge
+                  decision={
+                    c.status === "decided"
+                      ? c.human_decision
+                      : "pending"
+                  }
+                />
               </div>
               <div className="flex items-center gap-2 text-xs text-zinc-500">
                 <span className="capitalize">{c.status}</span>

diff --git a/web/src/WakeApi.tsx b/web/src/WakeApi.tsx
@@ -0,0 +1,76 @@
+import { useEffect, useState } from "react";
+import { api } from "./api";
+
+const INTERVAL_MS = 2_000;
+const MAX_ATTEMPTS = 45; // ~90s — covers Render free-tier cold start
+
+export default function WakeApi({ children }: { children: React.ReactNode }) {
+  const [ready, setReady] = useState(false);
+  const [failed, setFailed] = useState(false);
+  const [attempt, setAttempt] = useState(0);
+
+  useEffect(() => {
+    let cancelled = false;
+
+    async function poll() {
+      for (let i = 0; i < MAX_ATTEMPTS; i++) {
+        if (cancelled) return;
+        setAttempt(i + 1);
+        try {
+          await api.health();
+          setReady(true);
+          return;
+        } catch {
+          await new Promise((r) => setTimeout(r, INTERVAL_MS));
+        }
+      }
+      if (!cancelled) setFailed(true);
+    }
+
+    poll();
+    return () => {
+      cancelled = true;
+    };
+  }, []);
+
+  if (ready) return <>{children}</>;
+
+  return (
+    <div className="grid h-screen place-items-center bg-zinc-50 px-6">
+      <div className="w-80 text-center">
+        <div className="mx-auto mb-5 h-10 w-10 animate-spin rounded-full border-[3px] border-violet-600 border-t-transparent" />
+        {failed ? (
+          <>
+            <p className="text-sm font-medium text-zinc-800">
+              API didn&apos;t wake up in time
+            </p>
+            <p className="mt-2 text-sm text-zinc-500">
+              The demo API may still be starting. Refresh in a moment, or check
+              Render dashboard.
+            </p>
+            <button
+              onClick={() => window.location.reload()}
+              className="mt-5 rounded-lg bg-violet-600 px-4 py-2 text-sm font-medium text-white hover:bg-violet-700"
+            >
+              Retry
+            </button>
+          </>
+        ) : (
+          <>
+            <p className="text-sm font-medium text-zinc-800">
+              Waking demo API…
+            </p>
+            <p className="mt-2 text-sm text-zinc-500">
+              Free-tier container was asleep. First load can take up to ~30s.
+            </p>
+            {attempt > 3 && (
+              <p className="mt-3 text-xs text-zinc-400">
+                Still waiting… ({attempt}/{MAX_ATTEMPTS})
+              </p>
+            )}
+          </>
+        )}
+      </div>
+    </div>
+  );
+}