fix: stable claims paging and a deep link that does not lose its date window

- claims paging ordered by claim_id (unique) instead of service_date: each page
  is a separate Spark query, and with thousands of ties the engine repeated rows
  across pages
- the case's anchor date comes from the review's own ceiling or the lakehouse
  stats view, not from the briefing, which loads seconds later; an
  expiring-appeals link with no anchor is now suppressed rather than opening
  every unfiled denial

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

src/lib/claimsLink.ts

@@ -27,7 +27,10 @@ export function claimsHref(q: ClaimsQuery): string {
 /** The rows behind "claims denied" / "affected". */
 export function affectedClaimsHref(kind: string, slice: Slice | null, anchor: string | null): string | null {
   if (kind === 'expiring_appeals') {
-    return claimsHref({ status: 'denied', appeal: 'not_filed', dl_from: anchor ?? undefined, dl_to: anchor ? addDays(anchor, 14) : undefined });
+    // Without the anchor the 14-day window cannot be expressed, and the link
+    // would open every unfiled denial ever. Show no link instead.
+    if (!anchor) return null;
+    return claimsHref({ status: 'denied', appeal: 'not_filed', dl_from: anchor, dl_to: addDays(anchor, 14) });
   }
   if (kind === 'slow_payer') {
     return slice?.payer ? claimsHref({ payer: slice.payer, status: 'pending' }) : null;

src/screens/CaseScreen.tsx

@@ -17,7 +17,7 @@ import { ActivityForm } from '../components/ActivityForm';
 import { AvaLive } from '../components/AvaLive';
 import { Disclosure, FullStory, Recheck, Recommendations, SqlEvidence, Verdict } from '../components/Findings';
 import { Briefing, BriefingSkeleton } from '../components/Briefing';
-import { anchorDate, loadBriefing, sliceOf } from '../api/charts';
+import { loadBriefing, sliceOf } from '../api/charts';
 import type { Action } from '../api/config';
 import { dateTime, num, str } from '../format';
 
@@ -67,7 +67,14 @@ export function CaseScreen() {
   // Chart rows come from six Databricks views through the record-view API. Loaded
   // once there are findings to picture; the slice keys decide which weekly line
   // and which recovery breakdown are drawn.
-  const asOf = str(row?.review_as_of).slice(0, 10) || null;
+  const asOf = str(row?.review_to || row?.review_as_of).slice(0, 10) || null;
+  // The anchor behind "expiring within 14 days". It is the review's own
+  // ceiling date when the reviewer set one, otherwise the lakehouse's latest
+  // service date — read from the stats view, NOT from the briefing: the charts
+  // take seconds to load and a link rendered without the anchor would silently
+  // drop the deadline window and open thousands of rows instead of hundreds.
+  const lake = useQuery(() => client.rows('dni-rv-lakehouse-tables', { limit: 10 }), [], !asOf && hasFindings);
+  const anchor = asOf ?? str(lake.data?.find((r) => str(r.table_name) === 'claims')?.to_date).slice(0, 10) ?? null;
   const briefing = useQuery(
     () => loadBriefing(client, slice, { kind, asOf }),
     [id, slice?.payer, slice?.department, slice?.code, kind, asOf],
@@ -171,7 +178,7 @@ export function CaseScreen() {
       {/* ---------- The findings: what happened, how much, what to approve ---------- */}
       {hasFindings && (
         <>
-          <Verdict row={row} slice={slice} kind={kind} anchor={anchorDate(briefing.data)} />
+          <Verdict row={row} slice={slice} kind={kind} anchor={anchor || null} />
           {briefing.data ? (
             <Briefing data={briefing.data} slice={slice} kind={kind} />
           ) : (

src/screens/ClaimsScreen.tsx

@@ -69,8 +69,13 @@ export function ClaimsScreen() {
   const filtered = filters.length > 0 || !!search;
 
   const key = params.toString();
+  // ORDER BY A UNIQUE COLUMN. Each page is its own query against Spark, and
+  // `service_date` has thousands of ties — with a non-unique sort key the
+  // engine is free to order tied rows differently per execution, so page 2
+  // repeated rows from page 1. `claim_id` is unique and is assigned in date
+  // order, so descending is still newest-first, and every page is disjoint.
   const q = useQuery(
-    () => client.recordView('dni-rv-claims', { page, limit: PAGE, filters, search, sortBy: 'service_date', sortDir: 'desc' }),
+    () => client.recordView('dni-rv-claims', { page, limit: PAGE, filters, search, sortBy: 'claim_id', sortDir: 'desc' }),
     [key],
   );
   // The platform caps its count at 10,000; the stats view has the exact