social-autoposter 1.3.10 → 1.4.0

package/bin/server.js

@@ -477,7 +477,11 @@ const RUN_MONITOR_PATH = path.join(LOG_DIR, 'run_monitor.log');
 // queries+candidates+above_floor only. Each sub-key is omitted when zero, so
 // `discover=` itself is absent on lines from pipelines that don't emit it.
 // Old log lines without the segment still parse cleanly via the optional `?`.
-const RUN_LINE_RE = /^(\d{4}-\d{2}-\d{2}T\d{2}:\d{2}:\d{2})\s*\|\s*(\S+)\s*\|\s*posted=(\d+)\s+skipped=(\d+)\s+failed=(\d+)(?:\s+replies_refreshed=(\d+))?(?:\s+checked=(\d+)\s+updated=(\d+)\s+removed=(\d+))?(?:\s+unavailable=(\d+))?(?:\s+not_found=(\d+))?(?:\s+salvaged=(\d+))?(?:\s+discover=([^\s|]+))?(?:\s+scan=([^\s|]+))?\s+cost=\$([\d.]+)\s+elapsed=(\d+)s(?:\s+failure_reasons=([^\s|]+))?(?:\s+skip_reasons=([^\s|]+))?/;
+// 2026-05-18 stats-pill relabel: three new optional groups (scanned,
+// changed, views_refreshed) tail the unavailable/not_found block so old log
+// lines still parse via the existing positional regex. Each is independently
+// optional so a partial roll-out (just scanned, just changed) also parses.
+const RUN_LINE_RE = /^(\d{4}-\d{2}-\d{2}T\d{2}:\d{2}:\d{2})\s*\|\s*(\S+)\s*\|\s*posted=(\d+)\s+skipped=(\d+)\s+failed=(\d+)(?:\s+replies_refreshed=(\d+))?(?:\s+checked=(\d+)\s+updated=(\d+)\s+removed=(\d+))?(?:\s+unavailable=(\d+))?(?:\s+not_found=(\d+))?(?:\s+scanned=(\d+))?(?:\s+changed=(\d+))?(?:\s+views_refreshed=(\d+))?(?:\s+salvaged=(\d+))?(?:\s+discover=([^\s|]+))?(?:\s+scan=([^\s|]+))?\s+cost=\$([\d.]+)\s+elapsed=(\d+)s(?:\s+failure_reasons=([^\s|]+))?(?:\s+skip_reasons=([^\s|]+))?/;
 
 // posts.platform is lowercase; UI labels are capitalized.
 const PLATFORM_LABELS = {
@@ -586,7 +590,7 @@ function parseRunMonitorLog(maxLines) {
   for (const line of tail) {
     const m = line.match(RUN_LINE_RE);
     if (!m) continue;
-    const [, ts, script, posted, skipped, failed, repliesRefreshed, checked, updated, removed, unavailable, notFound, salvaged, discoverStr, scanStr, cost, elapsed, failureReasonsStr, skipReasonsStr] = m;
+    const [, ts, script, posted, skipped, failed, repliesRefreshed, checked, updated, removed, unavailable, notFound, scannedRaw, changedRaw, viewsRefreshedRaw, salvaged, discoverStr, scanStr, cost, elapsed, failureReasonsStr, skipReasonsStr] = m;
     if (JOB_HISTORY_HIDDEN_SCRIPTS.has(script)) continue;
     // log_run.py writes naive local-wallclock time (strftime without tz), so
     // `new Date(ts)` in node interprets it as local on the server. That is
@@ -670,6 +674,13 @@ function parseRunMonitorLog(maxLines) {
         removed: removed ? parseInt(removed, 10) : 0,
         unavailable: unavailable ? parseInt(unavailable, 10) : 0,
         not_found: notFound ? parseInt(notFound, 10) : 0,
+        // 2026-05-18 stats-pill relabel: three new fields surface the split
+        // between Step 1 view-scrape leg and Step 2 metric-changed leg.
+        // Absent on pre-relabel log lines (defaults to 0); renderResultPills
+        // falls back to `checked`/`updated` when these are 0.
+        scanned: scannedRaw ? parseInt(scannedRaw, 10) : 0,
+        changed: changedRaw ? parseInt(changedRaw, 10) : 0,
+        views_refreshed: viewsRefreshedRaw ? parseInt(viewsRefreshedRaw, 10) : 0,
         salvaged: salvaged ? parseInt(salvaged, 10) : 0,
         discover, // {} when no `discover=` segment was present on the line
         scan, // {} when no `scan=` segment was present on the line
@@ -1130,7 +1141,8 @@ async function enrichPostCommentsTwitterRuns(runs) {
   // otherwise undercount the per-run queue snapshot at older runs.
   const candidateRows = await pq(
     "SELECT discovered_at, posted_at, t1_checked_at, drafted_at, " +
-    "       (draft_reply_text IS NOT NULL) AS has_draft, status, batch_id " +
+    "       (draft_reply_text IS NOT NULL) AS has_draft, status, batch_id, " +
+    "       matched_project " +
     "FROM twitter_candidates " +
     "WHERE discovered_at >= $1::timestamp OR posted_at >= $1::timestamp OR t1_checked_at >= $1::timestamp OR status='pending'",
     [since]
@@ -1194,6 +1206,7 @@ async function enrichPostCommentsTwitterRuns(runs) {
       exitMs,
       status: r.status,
       batch_id: r.batch_id || '',
+      matched_project: r.matched_project || '',
     };
   });
 
@@ -1235,9 +1248,25 @@ async function enrichPostCommentsTwitterRuns(runs) {
     }
     let candidatesPassed = 0;
     let salvagePosted = 0;
+    // Project labels this cycle actually worked on, surfaced at the end of the
+    // pill row (mirrors enrichPostCommentsRedditRuns). Source is the
+    // twitter_candidates.matched_project values for rows tied to this run's
+    // own batch_id — i.e. tweets Phase 1 scraped + scored on behalf of these
+    // projects. Insertion-order Set preserves first-seen order while deduping.
+    const projectsSeen = new Set();
+    const projectsList = [];
+    const recordProject = (raw) => {
+      if (!raw) return;
+      const proj = raw.trim();
+      if (proj && !projectsSeen.has(proj)) {
+        projectsSeen.add(proj);
+        projectsList.push(proj);
+      }
+    };
     for (const c of candNorm) {
       if (!ownBatchId || c.batch_id !== ownBatchId) continue;
       candidatesPassed++;
+      recordProject(c.matched_project);
       if (c.status === 'posted') {
         posted++;
         // Salvage signature: candidate's discovered_at predates this cycle's
@@ -1270,6 +1299,19 @@ async function enrichPostCommentsTwitterRuns(runs) {
         const body = fs.readFileSync(path.join(LOG_DIR, chosenLog), 'utf8');
         const m = body.match(phase0SalvageRe);
         if (m) salvageAttempted = parseInt(m[1], 10);
+        // Fallback for scan-only cycles where 0 candidates upserted (so
+        // matched_project never landed in twitter_candidates for this batch):
+        // pull project names from the cycle log's "Selected projects:" header.
+        // This surfaces the projects the cycle SCANNED for, mirroring Reddit's
+        // behaviour of capturing project labels even when nothing posted.
+        if (!projectsList.length) {
+          const selM = body.match(/Selected projects:\s*([^\n]+)/);
+          if (selM) {
+            for (const tok of selM[1].split(',')) {
+              recordProject(tok);
+            }
+          }
+        }
       } catch { /* empty */ }
     }
     // Per-run queue delta. ADD = candidates whose discovered_at fell in this
@@ -1373,6 +1415,12 @@ async function enrichPostCommentsTwitterRuns(runs) {
       salvage_attempted: salvageAttempted,
       salvage_posted: salvagePosted,
       own_batch_id: ownBatchId,
+      // Project(s) this cycle actually worked on (Phase 1 scraped + scored
+      // candidates for), parsed from twitter_candidates.matched_project for
+      // rows tied to this run's batch_id. Surfaces at the end of the dashboard
+      // pill row so the operator can see at a glance which projects consumed
+      // the cycle, even when posted=0. Mirrors enrichPostCommentsRedditRuns.
+      projects_worked: projectsList,
       cost_usd: prior.cost_usd || 0,
       failed: prior.failed || 0,
       failure_reasons: Array.isArray(prior.failure_reasons) ? prior.failure_reasons : [],
@@ -2224,7 +2272,7 @@ async function enrichSeoRuns(runs) {
     }
     if (sessionIds.length) {
       const rows = await pq(
-        'SELECT session_id, total_cost_usd, orchestrator_cost_usd FROM claude_sessions WHERE session_id = ANY($1::uuid[])',
+        'SELECT session_id, total_cost_usd, orchestrator_cost_usd, subagent_cost_usd FROM claude_sessions WHERE session_id = ANY($1::uuid[])',
         [sessionIds]
       );
       if (rows && rows.length) {
@@ -2234,6 +2282,9 @@ async function enrichSeoRuns(runs) {
             orchestrator: r.orchestrator_cost_usd != null
               ? Number(r.orchestrator_cost_usd)
               : null,
+            subagent: r.subagent_cost_usd != null
+              ? Number(r.subagent_cost_usd)
+              : null,
           }])
         );
         for (const run of seoRuns) {
@@ -2246,16 +2297,20 @@ async function enrichSeoRuns(runs) {
             // sessions, locked callers that don't pass --orchestrator-cost-usd),
             // keep whatever streamRes value _collectSeoDetails parsed from the
             // .log file.
+            // SDK-only: d.cost_usd is orchestrator_cost_usd alone. Estimate
+            // and subagent are kept on the row for diagnostic tooltips but
+            // not folded into the displayed total.
             if (Number.isFinite(row.orchestrator)) {
               d.cost_usd = row.orchestrator;
               d.cost_usd_orchestrator = row.orchestrator;
-            } else if (Number.isFinite(d.cost_usd)) {
-              d.cost_usd_orchestrator = d.cost_usd;
+            } else {
+              d.cost_usd = null;
+              d.cost_usd_orchestrator = null;
             }
-            // Manual estimate alongside (always populated by log_claude_session.py).
             if (Number.isFinite(row.estimated)) {
               d.cost_usd_estimated = row.estimated;
             }
+            d.cost_usd_subagent = Number.isFinite(row.subagent) ? row.subagent : 0;
           }
         }
       }
@@ -2284,6 +2339,229 @@ async function enrichSeoRuns(runs) {
   }
 }
 
+// Maps a Job History row's `script` (log_run.py canonical name) to the
+// `claude_sessions.script` values its wrapper actually spawns. Without this
+// constraint a 40-min post_reddit run's window catches every concurrent
+// pipeline's sessions (engage-dm-replies, seo_generate_page, etc.) and the
+// "cost" cell sums everything that ran in parallel — meaningless. Lookup is
+// keyed on the underscore form (post_reddit, dm_replies_reddit) that survives
+// classifyScript's normalization. Add new entries here when a new wrapper
+// pipeline ships.
+const _PHASE_FAMILY = {
+  // Reddit
+  post_reddit:          ['post_reddit'],
+  engage_reddit:        ['engage_reddit'],
+  thread_reddit:        ['run-reddit-threads'],
+  dm_outreach_reddit:   ['dm-outreach-reddit'],
+  dm_replies_reddit:    ['engage-dm-replies'],
+  link_edit_reddit:     ['link-edit-reddit'],
+  // Twitter
+  post_twitter:         ['run-twitter-cycle-scan', 'run-twitter-cycle-prep'],
+  engage_twitter:       ['engage-twitter-phaseB'],
+  thread_twitter:       ['run-twitter-threads'],
+  dm_outreach_twitter:  ['dm-outreach-twitter'],
+  dm_replies_twitter:   ['engage-dm-replies'],
+  link_edit_twitter:    ['link-edit-twitter'],
+  // LinkedIn (run-linkedin bare = pre-phase-split legacy tag, kept for old rows)
+  post_linkedin:        ['run-linkedin-phaseA', 'run-linkedin-phaseB', 'run-linkedin'],
+  engage_linkedin:      ['engage-linkedin-phaseA', 'engage-linkedin-phaseB'],
+  dm_replies_linkedin:  ['engage-dm-replies'],
+  dm_outreach_linkedin: ['dm-outreach-linkedin'],
+  link_edit_linkedin:   ['link-edit-linkedin'],
+  // GitHub
+  post_github:          ['post_github'],
+  engage_github:        ['engage_github', 'github-engage', 'run-github-cycle'],
+  link_edit_github:     ['link-edit-github'],
+  // Moltbook
+  post_moltbook:        ['run-moltbook-cycle'],
+  // SEO
+  gsc_seo:              ['seo_generate_page', 'seo_generate_page_retry'],
+  serp_seo:             ['seo_generate_page', 'seo_generate_page_retry'],
+  seo_improve:          ['seo_improve_page'],
+  seo_weekly_roundup:   ['seo_generate_page'],
+  seo_top_pages:        ['seo_generate_page'],
+  seo_top_posts:        ['seo_generate_page'],
+};
+
+function _phaseFamilyFor(runScript) {
+  if (!runScript) return null;
+  const norm = String(runScript).replace(/-/g, '_').toLowerCase();
+  const aliased = SCRIPT_ALIASES[norm] || norm;
+  const family = _PHASE_FAMILY[aliased];
+  return family && family.length ? new Set(family) : null;
+}
+
+// Per-phase Claude cost breakdown for Job History rows. For each completed
+// run, queries `claude_sessions` rows whose `script` belongs to the run's
+// known phase family (see _PHASE_FAMILY) AND whose `started_at` falls inside
+// [run.started_at - slack, run.finished_at + slack]. Groups results by
+// session script (the phase) and attaches:
+//   run.result.cost_breakdown = {
+//     total, orchestrator, subagent, estimated,
+//     phases: [{phase, sessions, total, orch, sub, est}, ...] // desc by total
+//   }
+//   run.result.cost_usd          = total (overrides the shell-log value when
+//                                  we found ≥1 matching session — the wrapper
+//                                  log line often misses sub-phase cost,
+//                                  e.g. engage_twitter logs $0 but phaseB
+//                                  spent real money).
+//   run.result.cost_usd_from_log = original shell-log value (preserved for
+//                                  audit/provenance).
+//   run.result.cost_usd_*        = orch/subagent/estimated lanes for the
+//                                  4-lane tooltip the Cost column renders.
+// Runs whose `script` isn't in _PHASE_FAMILY get NO breakdown (we don't know
+// which sessions to attribute) and keep their shell-log cost.
+//
+// cycle_id disambiguation (2026-05-15): time-window + family matching alone
+// mis-attributes overlapping cycles. Twitter cycles fire every 15 min, so a
+// 30-min run's window catches the previous cycle's late-running prep and the
+// next cycle's scan. Each wrapper invocation exports a unique SA_CYCLE_ID
+// that every child claude_sessions row inherits, so after window+family
+// matching we pick the ONE cycle_id whose earliest session is closest to the
+// run's started_at (that's the run's own scan/first phase) and drop every
+// session belonging to a different cycle_id. Sessions with NULL cycle_id are
+// only used when the run matched no cycle_id at all (pure fallback).
+async function enrichRunsCostBreakdown(runs) {
+  const candidates = (runs || []).filter(r =>
+    !r.running && r.started_at && r.finished_at && _phaseFamilyFor(r.script)
+  );
+  if (!candidates.length) return;
+  let minStart = Infinity, maxEnd = 0;
+  for (const r of candidates) {
+    const s = Date.parse(r.started_at);
+    const e = Date.parse(r.finished_at);
+    if (Number.isFinite(s) && s < minStart) minStart = s;
+    if (Number.isFinite(e) && e > maxEnd) maxEnd = e;
+  }
+  if (minStart === Infinity) return;
+  const slackMs = _RUN_WINDOW_SLACK_MS;
+  const since = new Date(minStart - slackMs).toISOString();
+  const until = new Date(maxEnd + slackMs).toISOString();
+  let rows;
+  try {
+    rows = await pq(
+      // SDK-only mode: orchestrator_cost_usd is the only number we display.
+      // Subagent and transcript-estimate are kept for diagnostic tooltips but
+      // never added into the total. has_sdk lets the UI distinguish "session
+      // happened but SDK didn't capture cost" from "session cost was $0".
+      "SELECT script, started_at, cycle_id, " +
+        "orchestrator_cost_usd IS NOT NULL AS has_sdk, " +
+        "COALESCE(orchestrator_cost_usd, 0)::float8 AS orch, " +
+        "COALESCE(total_cost_usd, 0)::float8 AS est, " +
+        "COALESCE(subagent_cost_usd, 0)::float8 AS sub " +
+      "FROM claude_sessions WHERE started_at BETWEEN $1::timestamp AND $2::timestamp",
+      [since, until]
+    );
+  } catch (e) {
+    console.error('[enrichRunsCostBreakdown] query failed:', e && e.message || e);
+    return;
+  }
+  if (!rows || !rows.length) return;
+  const sessionList = rows.map(r => ({
+    ts: r.started_at instanceof Date ? r.started_at.getTime() : Date.parse(r.started_at),
+    script: r.script || '(unknown)',
+    cycleId: r.cycle_id || null,
+    hasSdk: !!r.has_sdk,
+    orch: Number(r.orch) || 0,
+    est:  Number(r.est)  || 0,
+    sub:  Number(r.sub)  || 0,
+    // SDK-only: total = orch. Sessions without SDK contribute 0 to total
+    // (they're flagged via hasSdk so the tooltip can show coverage).
+    total: Number(r.orch) || 0,
+  }));
+  for (const r of candidates) {
+    const runStartMs = Date.parse(r.started_at);
+    const startMs = runStartMs - slackMs;
+    const endMs   = Date.parse(r.finished_at) + slackMs;
+    const family = _phaseFamilyFor(r.script);
+    // Step 1: window + family match (over-broad — catches overlapping cycles).
+    const windowMatched = sessionList.filter(s =>
+      s.ts >= startMs && s.ts <= endMs && (!family || family.has(s.script))
+    );
+    // Step 2: cycle_id disambiguation. Group matched sessions by cycle_id and
+    // pick the group whose earliest session starts closest to this run's
+    // started_at — that's the run's own cycle. Drop sessions from other
+    // cycle_ids. If no session carries a cycle_id, fall back to the full
+    // window-matched set (older sessions / pipelines that don't set
+    // SA_CYCLE_ID).
+    let attributed;
+    const byCycle = new Map();
+    for (const s of windowMatched) {
+      if (!s.cycleId) continue;
+      let g = byCycle.get(s.cycleId);
+      if (!g) { g = { earliest: s.ts, sessions: [] }; byCycle.set(s.cycleId, g); }
+      if (s.ts < g.earliest) g.earliest = s.ts;
+      g.sessions.push(s);
+    }
+    if (byCycle.size > 0) {
+      let bestCycle = null, bestDelta = Infinity;
+      for (const [cid, g] of byCycle) {
+        const delta = Math.abs(g.earliest - runStartMs);
+        if (delta < bestDelta) { bestDelta = delta; bestCycle = cid; }
+      }
+      attributed = byCycle.get(bestCycle).sessions;
+    } else {
+      attributed = windowMatched;
+    }
+    const byPhase = {};
+    let totalOrch = 0, totalEst = 0, totalSub = 0;
+    let sessionsAll = 0, sessionsWithSdk = 0;
+    for (const s of attributed) {
+      const key = s.script;
+      const cur = byPhase[key] || { phase: key, sessions: 0, sessions_with_sdk: 0, orch: 0, est: 0, sub: 0 };
+      cur.sessions += 1;
+      if (s.hasSdk) cur.sessions_with_sdk += 1;
+      cur.orch += s.orch;
+      cur.est  += s.est;
+      cur.sub  += s.sub;
+      byPhase[key] = cur;
+      sessionsAll += 1;
+      if (s.hasSdk) sessionsWithSdk += 1;
+      totalOrch += s.orch;
+      totalEst  += s.est;
+      totalSub  += s.sub;
+    }
+    if (!sessionsAll) continue;
+    const phases = Object.values(byPhase)
+      .sort((a, b) => b.orch - a.orch || b.sessions - a.sessions)
+      .map(p => ({
+        phase: p.phase,
+        sessions: p.sessions,
+        sessions_with_sdk: p.sessions_with_sdk,
+        sessions_missing_sdk: p.sessions - p.sessions_with_sdk,
+        total: Number(p.orch.toFixed(6)),
+        orch:  Number(p.orch.toFixed(6)),
+        sub:   Number(p.sub.toFixed(6)),
+        est:   Number(p.est.toFixed(6)),
+      }));
+    if (!r.result) r.result = {};
+    // Preserve provenance: the original shell-log cost (parseFloat'd from the
+    // run_monitor line) is kept under cost_usd_from_log so an operator can
+    // see what the wrapper reported vs what we recomputed.
+    if (typeof r.result.cost_usd === 'number') {
+      r.result.cost_usd_from_log = r.result.cost_usd;
+    }
+    // SDK-only displayed total. Sessions missing SDK contribute 0 — the
+    // sessions_missing_sdk counter on each phase row is the only signal that
+    // real spend went unrecorded.
+    r.result.cost_breakdown = {
+      total: Number(totalOrch.toFixed(6)),
+      orchestrator: Number(totalOrch.toFixed(6)),
+      subagent: Number(totalSub.toFixed(6)),
+      estimated: Number(totalEst.toFixed(6)),
+      sessions: sessionsAll,
+      sessions_with_sdk: sessionsWithSdk,
+      sessions_missing_sdk: sessionsAll - sessionsWithSdk,
+      phases,
+    };
+    r.result.cost_usd = r.result.cost_breakdown.total;
+    r.result.cost_usd_orchestrator = r.result.cost_breakdown.orchestrator;
+    r.result.cost_usd_estimated    = r.result.cost_breakdown.estimated;
+    r.result.cost_usd_subagent     = r.result.cost_breakdown.subagent;
+    r.result.cost_sessions_missing_sdk = r.result.cost_breakdown.sessions_missing_sdk;
+  }
+}
+
 // DB-backed enrichment uses run.started_at and run.finished_at to define a
 // window, plus a small slack on each side for clock skew between the shell
 // trap that wrote the run_monitor line and the page row's completed_at.
@@ -3526,6 +3804,7 @@ async function handleApi(req, res) {
       await enrichPostCommentsTwitterRuns(runs);
       await enrichPostCommentsRedditRuns(runs);
       await enrichSeoRuns(runs);
+      await enrichRunsCostBreakdown(runs);
       // Prepend in-progress pipelines so they appear at the top of the table.
       // Always included regardless of the hours window — a long-running job
       // started before the window is still relevant right now.
@@ -3632,28 +3911,37 @@ async function handleApi(req, res) {
       "), session_counts AS (" +
         "SELECT claude_session_id, COUNT(*)::int AS rows_in_session FROM src GROUP BY claude_session_id" +
       "), session_cost AS (" +
+        // SDK-only mode (2026-05-15): per_row_cost = orchestrator_cost_usd
+        // alone, split evenly across activity rows. NULL when the wrapper
+        // didn't pass --orchestrator-cost-usd (e.g. shell wrappers that omit
+        // --output-format json so Claude never emits total_cost_usd). The
+        // transcript estimate and subagent dollars are computed from a local
+        // pricing table — kept in the JSON payload for diagnostics only,
+        // never folded into the displayed total.
         "SELECT cs.session_id, " +
-          "(COALESCE(cs.orchestrator_cost_usd, cs.total_cost_usd) / NULLIF(sc.rows_in_session, 0))::numeric(10,6) AS per_row_cost, " +
+          "(cs.orchestrator_cost_usd / NULLIF(sc.rows_in_session, 0))::numeric(10,6) AS per_row_cost, " +
           "(cs.orchestrator_cost_usd / NULLIF(sc.rows_in_session, 0))::numeric(10,6) AS per_row_cost_orchestrator, " +
-          "(cs.total_cost_usd / NULLIF(sc.rows_in_session, 0))::numeric(10,6) AS per_row_cost_estimated " +
+          "(cs.total_cost_usd / NULLIF(sc.rows_in_session, 0))::numeric(10,6) AS per_row_cost_estimated, " +
+          "(cs.subagent_cost_usd / NULLIF(sc.rows_in_session, 0))::numeric(10,6) AS per_row_cost_subagent " +
         "FROM claude_sessions cs JOIN session_counts sc ON sc.claude_session_id = cs.session_id" +
       ") " +
       "SELECT json_agg(row_to_json(r)) FROM (" +
-      "SELECT * FROM (SELECT posted_at AS occurred_at, CASE WHEN thread_url = our_url AND (thread_author IS NULL OR thread_author = our_account) THEN 'posted_thread' ELSE 'posted_comment' END AS type, platform, our_account AS actor, CASE WHEN thread_url = our_url AND (thread_author IS NULL OR thread_author = our_account) THEN COALESCE(thread_title, LEFT(our_content, 280)) ELSE LEFT(our_content, 280) END AS summary, engagement_style AS detail, our_url AS link, ('p' || posts.id) AS key, project_name AS project, sc.per_row_cost AS cost_usd, sc.per_row_cost_orchestrator AS cost_usd_orchestrator, sc.per_row_cost_estimated AS cost_usd_estimated, c.name AS campaign_name, CASE WHEN thread_url = our_url AND (thread_author IS NULL OR thread_author = our_account) THEN NULL ELSE thread_title END AS context_title, CASE WHEN thread_url = our_url AND (thread_author IS NULL OR thread_author = our_account) THEN NULL ELSE thread_url END AS context_url, LEFT(our_content, 3000) AS body FROM posts LEFT JOIN session_cost sc ON sc.session_id = posts.claude_session_id LEFT JOIN campaigns c ON c.id = posts.campaign_id WHERE posted_at IS NOT NULL AND our_content <> '(mention - no original post)' ORDER BY posted_at DESC LIMIT 150) x1 " +
-      "UNION ALL SELECT * FROM (SELECT r2.replied_at, 'replied', r2.platform, r2.their_author, COALESCE(LEFT(r2.our_reply_content, 280), LEFT(r2.their_content, 280)), CASE WHEN r2.is_recommendation THEN 'rec · ' || COALESCE(r2.engagement_style, '') ELSE r2.engagement_style END, r2.our_reply_url, ('r' || r2.id), p.project_name, sc.per_row_cost, sc.per_row_cost_orchestrator, sc.per_row_cost_estimated, c2.name, p.thread_title, p.thread_url, NULL::text FROM replies r2 LEFT JOIN posts p ON p.id = r2.post_id LEFT JOIN session_cost sc ON sc.session_id = r2.claude_session_id LEFT JOIN campaigns c2 ON c2.id = r2.campaign_id WHERE r2.status='replied' AND r2.replied_at IS NOT NULL ORDER BY r2.replied_at DESC LIMIT 150) x2 " +
-      "UNION ALL SELECT * FROM (SELECT COALESCE(r3.processing_at, r3.discovered_at), 'skipped', r3.platform, r3.their_author, LEFT(r3.their_content, 140), r3.skip_reason, r3.their_comment_url, ('s' || r3.id), p.project_name, sc.per_row_cost, sc.per_row_cost_orchestrator, sc.per_row_cost_estimated, c3.name, p.thread_title, p.thread_url, NULL::text FROM replies r3 LEFT JOIN posts p ON p.id = r3.post_id LEFT JOIN session_cost sc ON sc.session_id = r3.claude_session_id LEFT JOIN campaigns c3 ON c3.id = r3.campaign_id WHERE r3.status='skipped' ORDER BY COALESCE(r3.processing_at, r3.discovered_at) DESC LIMIT 150) x3 " +
-      "UNION ALL SELECT * FROM (SELECT COALESCE(source_timestamp, received_at), 'mention', platform, author, COALESCE(title, LEFT(body, 140)), sentiment, url, ('m' || id), NULL::text, NULL::numeric, NULL::numeric, NULL::numeric, NULL::text, NULL::text, NULL::text, NULL::text FROM octolens_mentions ORDER BY COALESCE(source_timestamp, received_at) DESC LIMIT 150) x4 " +
-      "UNION ALL SELECT * FROM (SELECT sent_at, 'dm_sent', platform, their_author, LEFT(our_dm_content, 140), NULL::text, chat_url, ('d' || dms.id), NULL::text, sc.per_row_cost, sc.per_row_cost_orchestrator, sc.per_row_cost_estimated, NULL::text, NULL::text, NULL::text, NULL::text FROM dms LEFT JOIN session_cost sc ON sc.session_id = dms.claude_session_id WHERE status='sent' AND sent_at IS NOT NULL ORDER BY sent_at DESC LIMIT 150) x5 " +
-      "UNION ALL SELECT * FROM (SELECT m.message_at, 'dm_reply_sent', d.platform, d.their_author, LEFT(m.content, 140), NULL::text, d.chat_url, ('dr' || m.id), NULL::text, sc.per_row_cost, sc.per_row_cost_orchestrator, sc.per_row_cost_estimated, c5.name, NULL::text, NULL::text, NULL::text FROM dm_messages m JOIN dms d ON d.id = m.dm_id LEFT JOIN session_cost sc ON sc.session_id = m.claude_session_id LEFT JOIN campaigns c5 ON c5.id = m.campaign_id WHERE m.direction = 'outbound' AND EXISTS (SELECT 1 FROM dm_messages m2 WHERE m2.dm_id = m.dm_id AND m2.direction = 'inbound' AND m2.message_at < m.message_at) ORDER BY m.message_at DESC LIMIT 150) x5b " +
-      "UNION ALL SELECT * FROM (SELECT completed_at, 'page_published_serp', 'seo', product, keyword, slug, page_url, ('k' || sk.id), product, sc.per_row_cost, sc.per_row_cost_orchestrator, sc.per_row_cost_estimated, NULL::text, NULL::text, NULL::text, NULL::text FROM seo_keywords sk LEFT JOIN session_cost sc ON sc.session_id = sk.claude_session_id WHERE completed_at IS NOT NULL AND page_url IS NOT NULL AND COALESCE(source, '') NOT IN ('reddit', 'top_page', 'top_post', 'roundup') ORDER BY completed_at DESC LIMIT 150) x6 " +
-      "UNION ALL SELECT * FROM (SELECT completed_at, 'page_published_gsc', 'seo', product, query, page_slug, page_url, ('g' || gq.id), product, sc.per_row_cost, sc.per_row_cost_orchestrator, sc.per_row_cost_estimated, NULL::text, NULL::text, NULL::text, NULL::text FROM gsc_queries gq LEFT JOIN session_cost sc ON sc.session_id = gq.claude_session_id WHERE completed_at IS NOT NULL AND page_url IS NOT NULL ORDER BY completed_at DESC LIMIT 150) x7 " +
-      "UNION ALL SELECT * FROM (SELECT completed_at, 'page_published_reddit', 'seo', product, keyword, slug, page_url, ('kr' || sk2.id), product, sc.per_row_cost, sc.per_row_cost_orchestrator, sc.per_row_cost_estimated, NULL::text, NULL::text, NULL::text, NULL::text FROM seo_keywords sk2 LEFT JOIN session_cost sc ON sc.session_id = sk2.claude_session_id WHERE completed_at IS NOT NULL AND page_url IS NOT NULL AND source = 'reddit' ORDER BY completed_at DESC LIMIT 150) x8 " +
-      "UNION ALL SELECT * FROM (SELECT completed_at, 'page_published_top', 'seo', product, keyword, slug, page_url, ('kt' || sk3.id), product, sc.per_row_cost, sc.per_row_cost_orchestrator, sc.per_row_cost_estimated, NULL::text, NULL::text, NULL::text, NULL::text FROM seo_keywords sk3 LEFT JOIN session_cost sc ON sc.session_id = sk3.claude_session_id WHERE completed_at IS NOT NULL AND page_url IS NOT NULL AND source = 'top_page' ORDER BY completed_at DESC LIMIT 150) x8b " +
-      "UNION ALL SELECT * FROM (SELECT completed_at, 'page_published_top_post', 'seo', product, keyword, slug, page_url, ('ktp' || sk5.id), product, sc.per_row_cost, sc.per_row_cost_orchestrator, sc.per_row_cost_estimated, NULL::text, NULL::text, NULL::text, NULL::text FROM seo_keywords sk5 LEFT JOIN session_cost sc ON sc.session_id = sk5.claude_session_id WHERE completed_at IS NOT NULL AND page_url IS NOT NULL AND source = 'top_post' ORDER BY completed_at DESC LIMIT 150) x8tp " +
-      "UNION ALL SELECT * FROM (SELECT completed_at, 'page_published_roundup', 'seo', product, keyword, slug, page_url, ('kru' || sk4.id), product, sc.per_row_cost, sc.per_row_cost_orchestrator, sc.per_row_cost_estimated, NULL::text, NULL::text, NULL::text, NULL::text FROM seo_keywords sk4 LEFT JOIN session_cost sc ON sc.session_id = sk4.claude_session_id WHERE completed_at IS NOT NULL AND page_url IS NOT NULL AND source = 'roundup' ORDER BY completed_at DESC LIMIT 150) x8r " +
-      "UNION ALL SELECT * FROM (SELECT completed_at, 'page_improved', 'seo', product, LEFT(COALESCE(rationale, diff_summary, page_path), 140), page_path, page_url, ('pi' || spi.id), product, sc.per_row_cost, sc.per_row_cost_orchestrator, sc.per_row_cost_estimated, NULL::text, NULL::text, NULL::text, NULL::text FROM seo_page_improvements spi LEFT JOIN session_cost sc ON sc.session_id = spi.claude_session_id WHERE completed_at IS NOT NULL AND status = 'committed' ORDER BY completed_at DESC LIMIT 150) x8c " +
-      "UNION ALL SELECT * FROM (SELECT expired_at, 'page_expired', 'seo', product, regexp_replace(source_path, '^.*/', ''), 'imp=' || impressions_30d || ' clicks=0 age=' || COALESCE(file_age_days::int, 0) || 'd ' || COALESCE(reason,''), page_url, ('xp' || sep.id), product, NULL::numeric, NULL::numeric, NULL::numeric, NULL::text, NULL::text, NULL::text, NULL::text FROM seo_expired_pages sep ORDER BY expired_at DESC LIMIT 150) x8d " +
-      "UNION ALL SELECT * FROM (SELECT resurrected_at AS occurred_at, 'resurrected' AS type, platform, our_account AS actor, CASE WHEN thread_url = our_url AND (thread_author IS NULL OR thread_author = our_account) THEN COALESCE(thread_title, LEFT(our_content, 280)) ELSE LEFT(our_content, 280) END AS summary, NULL::text AS detail, our_url AS link, ('rr' || posts.id) AS key, project_name AS project, sc.per_row_cost AS cost_usd, sc.per_row_cost_orchestrator AS cost_usd_orchestrator, sc.per_row_cost_estimated AS cost_usd_estimated, c9.name AS campaign_name, CASE WHEN thread_url = our_url AND (thread_author IS NULL OR thread_author = our_account) THEN NULL ELSE thread_title END AS context_title, CASE WHEN thread_url = our_url AND (thread_author IS NULL OR thread_author = our_account) THEN NULL ELSE thread_url END AS context_url, LEFT(our_content, 3000) AS body FROM posts LEFT JOIN session_cost sc ON sc.session_id = posts.claude_session_id LEFT JOIN campaigns c9 ON c9.id = posts.campaign_id WHERE resurrected_at IS NOT NULL AND our_content <> '(mention - no original post)' ORDER BY resurrected_at DESC LIMIT 150) x9 " +
+      "SELECT * FROM (SELECT posted_at AS occurred_at, CASE WHEN thread_url = our_url AND (thread_author IS NULL OR thread_author = our_account) THEN 'posted_thread' ELSE 'posted_comment' END AS type, platform, our_account AS actor, CASE WHEN thread_url = our_url AND (thread_author IS NULL OR thread_author = our_account) THEN COALESCE(thread_title, LEFT(our_content, 280)) ELSE LEFT(our_content, 280) END AS summary, engagement_style AS detail, our_url AS link, ('p' || posts.id) AS key, project_name AS project, sc.per_row_cost AS cost_usd, sc.per_row_cost_orchestrator AS cost_usd_orchestrator, sc.per_row_cost_estimated AS cost_usd_estimated, sc.per_row_cost_subagent AS cost_usd_subagent, c.name AS campaign_name, CASE WHEN thread_url = our_url AND (thread_author IS NULL OR thread_author = our_account) THEN NULL ELSE thread_title END AS context_title, CASE WHEN thread_url = our_url AND (thread_author IS NULL OR thread_author = our_account) THEN NULL ELSE thread_url END AS context_url, LEFT(our_content, 3000) AS body FROM posts LEFT JOIN session_cost sc ON sc.session_id = posts.claude_session_id LEFT JOIN campaigns c ON c.id = posts.campaign_id WHERE posted_at IS NOT NULL AND our_content <> '(mention - no original post)' ORDER BY posted_at DESC LIMIT 150) x1 " +
+      "UNION ALL SELECT * FROM (SELECT r2.replied_at, 'replied', r2.platform, r2.their_author, COALESCE(LEFT(r2.our_reply_content, 280), LEFT(r2.their_content, 280)), CASE WHEN r2.is_recommendation THEN 'rec · ' || COALESCE(r2.engagement_style, '') ELSE r2.engagement_style END, r2.our_reply_url, ('r' || r2.id), p.project_name, sc.per_row_cost, sc.per_row_cost_orchestrator, sc.per_row_cost_estimated, sc.per_row_cost_subagent, c2.name, p.thread_title, p.thread_url, NULL::text FROM replies r2 LEFT JOIN posts p ON p.id = r2.post_id LEFT JOIN session_cost sc ON sc.session_id = r2.claude_session_id LEFT JOIN campaigns c2 ON c2.id = r2.campaign_id WHERE r2.status='replied' AND r2.replied_at IS NOT NULL ORDER BY r2.replied_at DESC LIMIT 150) x2 " +
+      "UNION ALL SELECT * FROM (SELECT COALESCE(r3.processing_at, r3.discovered_at), 'skipped', r3.platform, r3.their_author, LEFT(r3.their_content, 140), r3.skip_reason, r3.their_comment_url, ('s' || r3.id), p.project_name, sc.per_row_cost, sc.per_row_cost_orchestrator, sc.per_row_cost_estimated, sc.per_row_cost_subagent, c3.name, p.thread_title, p.thread_url, NULL::text FROM replies r3 LEFT JOIN posts p ON p.id = r3.post_id LEFT JOIN session_cost sc ON sc.session_id = r3.claude_session_id LEFT JOIN campaigns c3 ON c3.id = r3.campaign_id WHERE r3.status='skipped' ORDER BY COALESCE(r3.processing_at, r3.discovered_at) DESC LIMIT 150) x3 " +
+      "UNION ALL SELECT * FROM (SELECT COALESCE(source_timestamp, received_at), 'mention', platform, author, COALESCE(title, LEFT(body, 140)), sentiment, url, ('m' || id), NULL::text, NULL::numeric, NULL::numeric, NULL::numeric, NULL::numeric, NULL::text, NULL::text, NULL::text, NULL::text FROM octolens_mentions ORDER BY COALESCE(source_timestamp, received_at) DESC LIMIT 150) x4 " +
+      "UNION ALL SELECT * FROM (SELECT sent_at, 'dm_sent', platform, their_author, LEFT(our_dm_content, 140), NULL::text, chat_url, ('d' || dms.id), NULL::text, sc.per_row_cost, sc.per_row_cost_orchestrator, sc.per_row_cost_estimated, sc.per_row_cost_subagent, NULL::text, NULL::text, NULL::text, NULL::text FROM dms LEFT JOIN session_cost sc ON sc.session_id = dms.claude_session_id WHERE status='sent' AND sent_at IS NOT NULL ORDER BY sent_at DESC LIMIT 150) x5 " +
+      "UNION ALL SELECT * FROM (SELECT m.message_at, 'dm_reply_sent', d.platform, d.their_author, LEFT(m.content, 140), NULL::text, d.chat_url, ('dr' || m.id), NULL::text, sc.per_row_cost, sc.per_row_cost_orchestrator, sc.per_row_cost_estimated, sc.per_row_cost_subagent, c5.name, NULL::text, NULL::text, NULL::text FROM dm_messages m JOIN dms d ON d.id = m.dm_id LEFT JOIN session_cost sc ON sc.session_id = m.claude_session_id LEFT JOIN campaigns c5 ON c5.id = m.campaign_id WHERE m.direction = 'outbound' AND EXISTS (SELECT 1 FROM dm_messages m2 WHERE m2.dm_id = m.dm_id AND m2.direction = 'inbound' AND m2.message_at < m.message_at) ORDER BY m.message_at DESC LIMIT 150) x5b " +
+      "UNION ALL SELECT * FROM (SELECT completed_at, 'page_published_twitter', 'seo', product, keyword, slug, page_url, ('k' || sk.id), product, sc.per_row_cost, sc.per_row_cost_orchestrator, sc.per_row_cost_estimated, sc.per_row_cost_subagent, NULL::text, NULL::text, NULL::text, NULL::text FROM seo_keywords sk LEFT JOIN session_cost sc ON sc.session_id = sk.claude_session_id WHERE completed_at IS NOT NULL AND page_url IS NOT NULL AND source = 'twitter' ORDER BY completed_at DESC LIMIT 150) x6 " +
+      "UNION ALL SELECT * FROM (SELECT completed_at, 'page_published_misc', 'seo', product, keyword, slug, page_url, ('km' || sk6.id), product, sc.per_row_cost, sc.per_row_cost_orchestrator, sc.per_row_cost_estimated, sc.per_row_cost_subagent, NULL::text, NULL::text, NULL::text, NULL::text FROM seo_keywords sk6 LEFT JOIN session_cost sc ON sc.session_id = sk6.claude_session_id WHERE completed_at IS NOT NULL AND page_url IS NOT NULL AND COALESCE(source, '') NOT IN ('reddit', 'top_page', 'top_post', 'roundup', 'twitter') ORDER BY completed_at DESC LIMIT 150) x6m " +
+      "UNION ALL SELECT * FROM (SELECT completed_at, 'page_published_gsc', 'seo', product, query, page_slug, page_url, ('g' || gq.id), product, sc.per_row_cost, sc.per_row_cost_orchestrator, sc.per_row_cost_estimated, sc.per_row_cost_subagent, NULL::text, NULL::text, NULL::text, NULL::text FROM gsc_queries gq LEFT JOIN session_cost sc ON sc.session_id = gq.claude_session_id WHERE completed_at IS NOT NULL AND page_url IS NOT NULL ORDER BY completed_at DESC LIMIT 150) x7 " +
+      "UNION ALL SELECT * FROM (SELECT completed_at, 'page_published_reddit', 'seo', product, keyword, slug, page_url, ('kr' || sk2.id), product, sc.per_row_cost, sc.per_row_cost_orchestrator, sc.per_row_cost_estimated, sc.per_row_cost_subagent, NULL::text, NULL::text, NULL::text, NULL::text FROM seo_keywords sk2 LEFT JOIN session_cost sc ON sc.session_id = sk2.claude_session_id WHERE completed_at IS NOT NULL AND page_url IS NOT NULL AND source = 'reddit' ORDER BY completed_at DESC LIMIT 150) x8 " +
+      "UNION ALL SELECT * FROM (SELECT completed_at, 'page_published_top', 'seo', product, keyword, slug, page_url, ('kt' || sk3.id), product, sc.per_row_cost, sc.per_row_cost_orchestrator, sc.per_row_cost_estimated, sc.per_row_cost_subagent, NULL::text, NULL::text, NULL::text, NULL::text FROM seo_keywords sk3 LEFT JOIN session_cost sc ON sc.session_id = sk3.claude_session_id WHERE completed_at IS NOT NULL AND page_url IS NOT NULL AND source = 'top_page' ORDER BY completed_at DESC LIMIT 150) x8b " +
+      "UNION ALL SELECT * FROM (SELECT completed_at, 'page_published_top_post', 'seo', product, keyword, slug, page_url, ('ktp' || sk5.id), product, sc.per_row_cost, sc.per_row_cost_orchestrator, sc.per_row_cost_estimated, sc.per_row_cost_subagent, NULL::text, NULL::text, NULL::text, NULL::text FROM seo_keywords sk5 LEFT JOIN session_cost sc ON sc.session_id = sk5.claude_session_id WHERE completed_at IS NOT NULL AND page_url IS NOT NULL AND source = 'top_post' ORDER BY completed_at DESC LIMIT 150) x8tp " +
+      "UNION ALL SELECT * FROM (SELECT completed_at, 'page_published_roundup', 'seo', product, keyword, slug, page_url, ('kru' || sk4.id), product, sc.per_row_cost, sc.per_row_cost_orchestrator, sc.per_row_cost_estimated, sc.per_row_cost_subagent, NULL::text, NULL::text, NULL::text, NULL::text FROM seo_keywords sk4 LEFT JOIN session_cost sc ON sc.session_id = sk4.claude_session_id WHERE completed_at IS NOT NULL AND page_url IS NOT NULL AND source = 'roundup' ORDER BY completed_at DESC LIMIT 150) x8r " +
+      "UNION ALL SELECT * FROM (SELECT completed_at, 'page_improved', 'seo', product, LEFT(COALESCE(rationale, diff_summary, page_path), 140), page_path, page_url, ('pi' || spi.id), product, sc.per_row_cost, sc.per_row_cost_orchestrator, sc.per_row_cost_estimated, sc.per_row_cost_subagent, NULL::text, NULL::text, NULL::text, NULL::text FROM seo_page_improvements spi LEFT JOIN session_cost sc ON sc.session_id = spi.claude_session_id WHERE completed_at IS NOT NULL AND status = 'committed' ORDER BY completed_at DESC LIMIT 150) x8c " +
+      "UNION ALL SELECT * FROM (SELECT expired_at, 'page_expired', 'seo', product, regexp_replace(source_path, '^.*/', ''), 'imp=' || impressions_30d || ' clicks=0 age=' || COALESCE(file_age_days::int, 0) || 'd ' || COALESCE(reason,''), page_url, ('xp' || sep.id), product, NULL::numeric, NULL::numeric, NULL::numeric, NULL::numeric, NULL::text, NULL::text, NULL::text, NULL::text FROM seo_expired_pages sep ORDER BY expired_at DESC LIMIT 150) x8d " +
+      "UNION ALL SELECT * FROM (SELECT resurrected_at AS occurred_at, 'resurrected' AS type, platform, our_account AS actor, CASE WHEN thread_url = our_url AND (thread_author IS NULL OR thread_author = our_account) THEN COALESCE(thread_title, LEFT(our_content, 280)) ELSE LEFT(our_content, 280) END AS summary, NULL::text AS detail, our_url AS link, ('rr' || posts.id) AS key, project_name AS project, sc.per_row_cost AS cost_usd, sc.per_row_cost_orchestrator AS cost_usd_orchestrator, sc.per_row_cost_estimated AS cost_usd_estimated, sc.per_row_cost_subagent AS cost_usd_subagent, c9.name AS campaign_name, CASE WHEN thread_url = our_url AND (thread_author IS NULL OR thread_author = our_account) THEN NULL ELSE thread_title END AS context_title, CASE WHEN thread_url = our_url AND (thread_author IS NULL OR thread_author = our_account) THEN NULL ELSE thread_url END AS context_url, LEFT(our_content, 3000) AS body FROM posts LEFT JOIN session_cost sc ON sc.session_id = posts.claude_session_id LEFT JOIN campaigns c9 ON c9.id = posts.campaign_id WHERE resurrected_at IS NOT NULL AND our_content <> '(mention - no original post)' ORDER BY resurrected_at DESC LIMIT 150) x9 " +
       "ORDER BY 1 DESC LIMIT 500) r";
     return (async () => {
       const rows = await pq(q);
@@ -3668,6 +3956,7 @@ async function handleApi(req, res) {
           delete e.cost_usd;
           delete e.cost_usd_orchestrator;
           delete e.cost_usd_estimated;
+          delete e.cost_usd_subagent;
         });
       }
       return json(res, { events });
@@ -4158,7 +4447,16 @@ async function handleApi(req, res) {
     // started with a prospect."
     parts.push("SELECT 'dm_sent' AS type, d.platform AS pl FROM dms d WHERE EXISTS (SELECT 1 FROM dm_messages m WHERE m.dm_id = d.id AND m.direction='outbound' AND m.message_at >= NOW() - " + win + " AND NOT EXISTS (SELECT 1 FROM dm_messages m2 WHERE m2.dm_id = d.id AND m2.direction='outbound' AND m2.message_at < m.message_at))" + dmsAliasedPc.clause);
     parts.push("SELECT 'dm_reply_sent' AS type, d.platform AS pl FROM dm_messages m JOIN dms d ON d.id = m.dm_id WHERE m.direction='outbound' AND m.message_at >= NOW() - " + win + " AND EXISTS (SELECT 1 FROM dm_messages m2 WHERE m2.dm_id = m.dm_id AND m2.direction='inbound' AND m2.message_at < m.message_at)" + dmsAliasedPc.clause);
-    parts.push("SELECT 'page_published_serp' AS type, 'seo' AS pl FROM seo_keywords WHERE completed_at >= NOW() - " + win + " AND page_url IS NOT NULL AND COALESCE(source, '') NOT IN ('reddit', 'top_page', 'top_post', 'roundup')" + seoProdPc.clause);
+    // Pre-2026-05-16: a single 'page_published_serp' bucket caught every
+    // seo_keywords row whose source was not (reddit, top_page, top_post,
+    // roundup). The real SERP pipeline was unloaded 2026-04-17; the cards
+    // labelled "SERP SEO" since then have actually been Twitter-cycle
+    // page-gen output (twitter_gen_links.py -> generate_page.py --trigger
+    // twitter, source='twitter'). Split into an honest twitter bucket and a
+    // misc catch-all for legacy/edge sources (existing_page, gsc,
+    // suggestion:*, competitor:*, topic_template, etc).
+    parts.push("SELECT 'page_published_twitter' AS type, 'seo' AS pl FROM seo_keywords WHERE completed_at >= NOW() - " + win + " AND page_url IS NOT NULL AND source = 'twitter'" + seoProdPc.clause);
+    parts.push("SELECT 'page_published_misc' AS type, 'seo' AS pl FROM seo_keywords WHERE completed_at >= NOW() - " + win + " AND page_url IS NOT NULL AND COALESCE(source, '') NOT IN ('reddit', 'top_page', 'top_post', 'roundup', 'twitter')" + seoProdPc.clause);
     parts.push("SELECT 'page_published_gsc' AS type, 'seo' AS pl FROM gsc_queries WHERE completed_at >= NOW() - " + win + " AND page_url IS NOT NULL" + seoProdPc.clause);
     parts.push("SELECT 'page_published_reddit' AS type, 'seo' AS pl FROM seo_keywords WHERE completed_at >= NOW() - " + win + " AND page_url IS NOT NULL AND source='reddit'" + seoProdPc.clause);
     parts.push("SELECT 'page_published_top' AS type, 'seo' AS pl FROM seo_keywords WHERE completed_at >= NOW() - " + win + " AND page_url IS NOT NULL AND source='top_page'" + seoProdPc.clause);
@@ -4354,10 +4652,18 @@ async function handleApi(req, res) {
     const projectFilter = project
       ? " AND p.project_name = '" + project.replace(/'/g, "''") + "'"
       : '';
+    // Split posts_made into threads_made (we authored the thread itself) vs
+    // comments_made (we engaged on someone else's thread). Matches the
+    // /api/activity classifier: thread iff thread_url = our_url AND
+    // (thread_author IS NULL OR thread_author = our_account).
+    const threadClause =
+      "p.thread_url = p.our_url AND (p.thread_author IS NULL OR p.thread_author = p.our_account)";
     const q =
       "SELECT json_agg(row_to_json(r)) FROM (" +
         "SELECT to_char((p.posted_at AT TIME ZONE 'UTC')::date, 'YYYY-MM-DD') AS day, " +
-          "COUNT(*)::bigint AS posts_made " +
+          "COUNT(*)::bigint AS posts_made, " +
+          "SUM(CASE WHEN " + threadClause + " THEN 1 ELSE 0 END)::bigint AS threads_made, " +
+          "SUM(CASE WHEN " + threadClause + " THEN 0 ELSE 1 END)::bigint AS comments_made " +
         "FROM posts p " +
         "WHERE p.posted_at IS NOT NULL " +
           "AND p.posted_at >= CURRENT_DATE - INTERVAL '" + days + " days' " +
@@ -4506,7 +4812,8 @@ async function handleApi(req, res) {
     const sumCols =
       "COALESCE(SUM(sc.per_row_cost), 0)::numeric(12,4) AS total_cost_usd, " +
       "COALESCE(SUM(sc.per_row_cost_orchestrator), 0)::numeric(12,4) AS total_cost_usd_orchestrator, " +
-      "COALESCE(SUM(sc.per_row_cost_estimated), 0)::numeric(12,4) AS total_cost_usd_estimated";
+      "COALESCE(SUM(sc.per_row_cost_estimated), 0)::numeric(12,4) AS total_cost_usd_estimated, " +
+      "COALESCE(SUM(sc.per_row_cost_subagent), 0)::numeric(12,4) AS total_cost_usd_subagent";
     if (includeThread) {
       rowQueries.push(
         "SELECT 'thread' AS type, COUNT(*)::int AS count, " + sumCols + " " +
@@ -4545,15 +4852,46 @@ async function handleApi(req, res) {
       "WITH src AS (" + parts.join(' UNION ALL ') + "), " +
       "session_counts AS (SELECT claude_session_id, COUNT(*)::int AS rows_in_session FROM src GROUP BY claude_session_id), " +
       "session_cost AS (SELECT cs.session_id, " +
-          "(COALESCE(cs.orchestrator_cost_usd, cs.total_cost_usd) / NULLIF(sc.rows_in_session, 0))::numeric(12,6) AS per_row_cost, " +
+          "(cs.orchestrator_cost_usd / NULLIF(sc.rows_in_session, 0))::numeric(12,6) AS per_row_cost, " +
           "(cs.orchestrator_cost_usd / NULLIF(sc.rows_in_session, 0))::numeric(12,6) AS per_row_cost_orchestrator, " +
-          "(cs.total_cost_usd / NULLIF(sc.rows_in_session, 0))::numeric(12,6) AS per_row_cost_estimated " +
+          "(cs.total_cost_usd / NULLIF(sc.rows_in_session, 0))::numeric(12,6) AS per_row_cost_estimated, " +
+          "(cs.subagent_cost_usd / NULLIF(sc.rows_in_session, 0))::numeric(12,6) AS per_row_cost_subagent " +
         "FROM claude_sessions cs JOIN session_counts sc ON sc.claude_session_id = cs.session_id) " +
       "SELECT json_agg(row_to_json(r)) FROM (" + rowQueries.join(' UNION ALL ') + ") r";
+    // Per-phase (script) cost rollup over the same window. Groups every
+    // claude_sessions row in the window by its `script` column (e.g.
+    // run-twitter-cycle-scan, post_reddit, seo_generate_page) and surfaces
+    // total + orchestrator + subagent + estimate lanes. Independent of the
+    // activity-type rollup above: a single post_reddit session might produce
+    // 0 or 1 thread row, but its cost still shows up in the per-phase view.
+    // SDK-only per-phase rollup. total_cost_usd = SUM(orchestrator_cost_usd)
+    // — no transcript estimate, no subagent fold-in. Phases with 0% SDK
+    // coverage (wrapper doesn't capture --orchestrator-cost-usd) show
+    // total $0 but a non-zero sessions_missing_sdk count, which is the
+    // signal to investigate. Also surfaces the estimate and subagent
+    // columns as diagnostic-only fields.
+    const phaseQ =
+      "SELECT script AS phase, " +
+        "COUNT(*)::int AS sessions, " +
+        "COUNT(orchestrator_cost_usd)::int AS sessions_with_sdk, " +
+        "(COUNT(*) - COUNT(orchestrator_cost_usd))::int AS sessions_missing_sdk, " +
+        "COALESCE(SUM(orchestrator_cost_usd), 0)::numeric(12,4) AS total_cost_usd, " +
+        "COALESCE(SUM(orchestrator_cost_usd), 0)::numeric(12,4) AS total_cost_usd_orchestrator, " +
+        "COALESCE(SUM(total_cost_usd), 0)::numeric(12,4) AS total_cost_usd_estimated, " +
+        "COALESCE(SUM(subagent_cost_usd), 0)::numeric(12,4) AS total_cost_usd_subagent " +
+      "FROM claude_sessions " +
+      "WHERE started_at >= NOW() - " + win + " " +
+      "GROUP BY script " +
+      "HAVING COUNT(*) > 0 " +
+      "ORDER BY total_cost_usd DESC, sessions DESC " +
+      "LIMIT 50";
     return (async () => {
       const dbRows = await pq(q);
       const value = (dbRows && dbRows.length && dbRows[0].json_agg) ? dbRows[0].json_agg : [];
-      return json(res, { windowHours, platform: plat || 'all', rows: value });
+      let phases = [];
+      try { phases = await pq(phaseQ) || []; }
+      catch (e) { console.error('[cost/stats] phase query failed:', e && e.message || e); }
+      return json(res, { windowHours, platform: plat || 'all', rows: value, phases });
     })().catch(e => json(res, { error: e.message }, 500));
   }
 
@@ -4674,7 +5012,7 @@ async function handleApi(req, res) {
       "WITH src AS (" + parts.join(' UNION ALL ') + "), " +
       "session_counts AS (SELECT claude_session_id, COUNT(*)::int AS rows_in_session FROM src GROUP BY claude_session_id), " +
       "session_cost AS (SELECT cs.session_id, " +
-          "(COALESCE(cs.orchestrator_cost_usd, cs.total_cost_usd) / NULLIF(sc.rows_in_session, 0))::numeric(12,6) AS per_row_cost " +
+          "(cs.orchestrator_cost_usd / NULLIF(sc.rows_in_session, 0))::numeric(12,6) AS per_row_cost " +
         "FROM claude_sessions cs JOIN session_counts sc ON sc.claude_session_id = cs.session_id), " +
       "in_window AS (" + inWindow.join(' UNION ALL ') + ") " +
       "SELECT json_agg(row_to_json(r)) FROM (" +
@@ -5181,10 +5519,47 @@ async function handleApi(req, res) {
     const pc = auth.projectClause(req.user, 'project_name', url.searchParams.get('project'));
     if (!pc.ok) return json(res, { posts: [], window: windowKey, platform: platformFilter || 'all', kind: kindFilter });
     if (pc.clause) whereParts.push(pc.clause.replace(/^\s*AND\s+/, ''));
+    // 2026-05-18 dashboard parity for Reddit DM-rail follow-ups.
+    // Reply-to-reply rows (we replied back to someone who replied to our
+    // top-level comment) live in the `replies` table, not `posts`, so they
+    // were invisible on the Top tab even though they have real engagement
+    // (some routinely break 1000 views). LinkedIn migrated the analogous
+    // data into `posts` on 2026-05-11, but the (platform, thread_url) dedup
+    // on /api/v1/posts blocks the same migration for Reddit (most reply-
+    // to-replies share thread_url with our top-level post in the same
+    // thread). UNION-into-/api/top sidesteps that entirely: the `replies`
+    // table stays the source of truth, and the Top tab surfaces both
+    // surfaces. The activity feed (line ~3919) already filters
+    // r.status='replied' so reply-to-replies appear ONCE there.
+    //
+    // Build the same WHERE clause for the replies branch, but referencing
+    // `r2.replied_at` for the time window and `parent.thread_url` semantics
+    // for the kind filter. Replies are by definition never threads, so
+    // kind='threads' must exclude the whole branch.
+    const repliesPc = auth.projectClause(req.user, 'parent.project_name', url.searchParams.get('project'));
+    const repliesWhere = [
+      "r2.platform = 'reddit'",          // only Reddit has the gap today
+      "r2.status = 'replied'",
+      "r2.our_reply_id IS NOT NULL",
+      "r2.our_reply_url IS NOT NULL",
+      "r2.our_reply_content IS NOT NULL AND LENGTH(r2.our_reply_content) >= 30",
+    ];
+    if (windowHours != null) {
+      repliesWhere.push("r2.replied_at >= NOW() - INTERVAL '" + windowHours + " hours'");
+    }
+    if (platformFilter && platformFilter !== 'reddit') {
+      // Caller filtered to a non-Reddit platform; replies branch yields nothing.
+      repliesWhere.push("FALSE");
+    }
+    // kind: 'threads' excludes replies entirely; 'comments' and 'all' include.
+    if (kindFilter === 'threads') {
+      repliesWhere.push("FALSE");
+    }
+    if (repliesPc.clause) repliesWhere.push(repliesPc.clause.replace(/^\s*AND\s+/, ''));
     // Moltbook and GitHub have no views metric; return NULL for those so the UI can
     // render a dash instead of a misleading 0. Score still uses COALESCE so they
     // rank alongside other platforms based on upvotes + comments only.
-    const q = "SELECT json_agg(row_to_json(r)) FROM (" +
+    const postsBranch =
       "SELECT posts.id, posts.platform, " +
         // Upvotes are reported NET on Reddit/Moltbook (both auto-apply a +1 OP
         // self-upvote on every post). Strip it per row, clamped at 0 so
@@ -5216,7 +5591,8 @@ async function handleApi(req, res) {
         "COALESCE(pl.bot_clicks, 0)::int AS link_bot_clicks, " +
         "COALESCE(pl.backfill_real, 0)::int AS link_backfill_real, " +
         "COALESCE(pl.link_count, 0)::int AS link_count, " +
-        "pl.first_code AS link_code " +
+        "pl.first_code AS link_code, " +
+        "'post'::text AS row_kind " +
       "FROM posts LEFT JOIN campaigns c ON c.id = posts.campaign_id " +
       // pl rollup: legacy `total_clicks` reads the post_links.clicks integer
       // (humans-only after 2026-05-07; pre-existing rows include bots).
@@ -5245,6 +5621,50 @@ async function handleApi(req, res) {
       ") pl ON pl.post_id = posts.id " +
       "WHERE " + whereParts.join(' AND ') + " " +
       "ORDER BY upvotes DESC NULLS LAST, comments_count DESC NULLS LAST, views DESC NULLS LAST " +
+      "LIMIT " + limit;
+    // Replies branch: shape-compatible SELECT against `replies` JOIN `posts` (parent)
+    // for thread context. ID is negated to guarantee uniqueness within the UNION
+    // (posts.id and replies.id are independent sequences and would otherwise
+    // collide for sort/key purposes). The FE only uses `id` as a React key, so
+    // negative integers are fine.
+    const repliesBranch =
+      "SELECT (-r2.id)::int AS id, r2.platform, " +
+        // Reddit-only branch today; strip the OP self-upvote like the posts branch.
+        "GREATEST(0, COALESCE(r2.upvotes, 0) - 1)::int AS upvotes, " +
+        "COALESCE(r2.comments_count, 0)::int AS comments_count, " +
+        "COALESCE(r2.views, 0)::int AS views, " +
+        // Same score formula. Views (Reddit only) contribute /100.
+        "(COALESCE(r2.comments_count,0) * 5 " +
+          "+ GREATEST(0, COALESCE(r2.upvotes,0) - 1) * 5 " +
+          "+ COALESCE(r2.views,0) / 100)::int AS score, " +
+        "FALSE AS is_thread, " +
+        "r2.replied_at AS posted_at, " +
+        "r2.engagement_updated_at, " +
+        "r2.our_reply_content AS our_content, " +
+        "r2.our_reply_url AS our_url, " +
+        "parent.thread_url, parent.thread_title, " +
+        "LEFT(COALESCE(parent.thread_content, ''), 400) AS thread_content, " +
+        "parent.our_account, parent.project_name, " +
+        "r2.engagement_style, r2.is_recommendation, " +
+        "cr.name AS campaign_name, " +
+        // No link tracking on reply-to-replies yet (we don't usually drop a CTA
+        // there). All link counters return 0 so they sort to the bottom of
+        // any link-clicks ordering.
+        "0::int AS link_clicks, 0::int AS link_real_clicks, " +
+        "0::int AS link_bot_clicks, 0::int AS link_backfill_real, " +
+        "0::int AS link_count, NULL::text AS link_code, " +
+        "'reply'::text AS row_kind " +
+      "FROM replies r2 " +
+      "LEFT JOIN posts parent ON parent.id = r2.post_id " +
+      "LEFT JOIN campaigns cr ON cr.id = r2.campaign_id " +
+      "WHERE " + repliesWhere.join(' AND ') + " " +
+      "ORDER BY r2.upvotes DESC NULLS LAST, r2.comments_count DESC NULLS LAST, r2.views DESC NULLS LAST " +
+      "LIMIT " + limit;
+    const q = "SELECT json_agg(row_to_json(r)) FROM (" +
+      "SELECT * FROM (" + postsBranch + ") posts_branch " +
+      "UNION ALL " +
+      "SELECT * FROM (" + repliesBranch + ") replies_branch " +
+      "ORDER BY upvotes DESC NULLS LAST, comments_count DESC NULLS LAST, views DESC NULLS LAST " +
       "LIMIT " + limit +
       ") r";
     return (async () => {
@@ -5254,6 +5674,134 @@ async function handleApi(req, res) {
     })().catch(e => json(res, { error: e.message }, 500));
   }
 
+  // GET /api/top/destinations - post links rolled up by target URL.
+  // One row per unique destination URL (e.g. https://s4l.ai vs
+  // https://s4l.ai/ghostwriting vs https://s4l.ai/t/<slug>), with click totals
+  // aggregated across every short code that pointed at that URL. Used by the
+  // "Links" subtab in the Top tab to answer "where are my posts sending
+  // traffic and how many clicks does each destination get?"
+  if (p === '/api/top/destinations' && req.method === 'GET') {
+    const url = new URL(req.url, 'http://localhost');
+    const WINDOW_HOURS = { '24h': 24, '7d': 24*7, '14d': 24*14, '30d': 24*30, '90d': 24*90, 'all': null };
+    const rawWindow = String(url.searchParams.get('window') || '7d').toLowerCase();
+    const windowKey = Object.prototype.hasOwnProperty.call(WINDOW_HOURS, rawWindow) ? rawWindow : '7d';
+    const windowHours = WINDOW_HOURS[windowKey];
+    const rawPlatform = String(url.searchParams.get('platform') || '').toLowerCase().trim();
+    const ALLOWED_PLATFORMS = new Set(['reddit', 'twitter', 'x', 'linkedin', 'moltbook', 'github']);
+    const platformFilter = ALLOWED_PLATFORMS.has(rawPlatform) ? rawPlatform : '';
+    const pc = auth.projectClause(req.user, 'pl.project_name', url.searchParams.get('project'));
+    if (!pc.ok) return json(res, { destinations: [], window: windowKey, platform: 'all' });
+    const limit = Math.max(50, Math.min(500, parseInt(url.searchParams.get('limit') || '200', 10) || 200));
+    const whereParts = [];
+    if (windowHours != null) {
+      whereParts.push("pl.minted_at >= NOW() - INTERVAL '" + windowHours + " hours'");
+    }
+    if (platformFilter) {
+      whereParts.push("LOWER(pl.platform) = '" + platformFilter + "'");
+    }
+    if (pc.clause) whereParts.push(pc.clause.replace(/^\s*AND\s+/, ''));
+    const whereSql = whereParts.length ? ('WHERE ' + whereParts.join(' AND ')) : '';
+    // Grouping key is the URL with all query params + trailing slash stripped.
+    // UTM params (utm_source / utm_medium / utm_campaign / utm_term /
+    // utm_content) are baked into every target_url at mint time so each
+    // post gets its own uniquely-tracked URL even when posting at the same
+    // destination; without stripping, the same homepage would split into
+    // 50 rows. Path is preserved (so /ghostwriting stays distinct from /).
+    // Project + platform stay in GROUP BY so a multi-project repo (one
+    // website hosting pages for several projects) keeps them on separate
+    // rows.
+    const q = "SELECT json_agg(row_to_json(r)) FROM (" +
+      "SELECT " +
+        "REGEXP_REPLACE(REGEXP_REPLACE(pl.target_url, '\\?.*$', ''), '/$', '') AS target_url, " +
+        "pl.project_name, pl.platform, " +
+        "COUNT(DISTINCT pl.post_id)::int AS posts, " +
+        "COUNT(*)::int AS codes, " +
+        "COALESCE(SUM(pl.clicks), 0)::int AS legacy_clicks, " +
+        "COALESCE(SUM(plc.real_clicks), 0)::int AS real_clicks, " +
+        "COALESCE(SUM(plc.bot_clicks), 0)::int AS bot_clicks, " +
+        "COALESCE(SUM(pl.real_clicks), 0)::int AS backfill_real, " +
+        "MIN(pl.minted_at) AS first_minted_at, " +
+        "MAX(pl.last_click_at) AS last_click_at " +
+      "FROM post_links pl " +
+      "LEFT JOIN (" +
+        "SELECT code, " +
+        "  COUNT(*) FILTER (WHERE is_bot = false)::int AS real_clicks, " +
+        "  COUNT(*) FILTER (WHERE is_bot = true)::int AS bot_clicks " +
+        "FROM post_link_clicks GROUP BY code" +
+      ") plc ON plc.code = pl.code " +
+      whereSql + " " +
+      "GROUP BY 1, pl.project_name, pl.platform " +
+      "ORDER BY real_clicks DESC NULLS LAST, legacy_clicks DESC NULLS LAST, codes DESC " +
+      "LIMIT " + limit +
+      ") r";
+    return (async () => {
+      const rows = await pq(q);
+      const destinations = (rows && rows.length && rows[0].json_agg) ? rows[0].json_agg : [];
+      // Server-side classification of each destination URL into a kind bucket
+      // and (when applicable) the audience-page angle. Reads config.json
+      // once per request to look up each project's website host + audience
+      // pages list; classification is plain hostname / path matching. Done
+      // here so every consumer (UI, future CSV exports) sees the same
+      // canonical label without having to re-implement classify logic.
+      let cfg = null;
+      try {
+        cfg = JSON.parse(fs.readFileSync(CONFIG_FILE, 'utf8'));
+      } catch (_e) { cfg = { projects: [] }; }
+      const projIdx = {};
+      for (const p of (cfg && cfg.projects) || []) {
+        if (!p || !p.name) continue;
+        const host = String(p.website || '').replace(/^https?:\/\//, '').replace(/^www\./, '').replace(/\/.*$/, '').toLowerCase();
+        const audience = ((p.landing_pages || {}).audience_pages) || [];
+        const apIdx = [];
+        for (const a of audience) {
+          try {
+            const u = new URL(a.url);
+            apIdx.push({
+              angle: a.angle,
+              host: (u.hostname || '').toLowerCase().replace(/^www\./, ''),
+              path: (u.pathname || '/').replace(/\/+$/, '') || '/',
+            });
+          } catch (_e) {}
+        }
+        projIdx[p.name] = { website_host: host, audience_pages: apIdx };
+      }
+      const classify = (targetUrl, projectName) => {
+        if (!targetUrl) return { kind: 'other', audience_page_angle: null };
+        let host = '', pathName = '/';
+        try {
+          const u = new URL(targetUrl);
+          host = (u.hostname || '').toLowerCase().replace(/^www\./, '');
+          pathName = (u.pathname || '/').replace(/\/+$/, '') || '/';
+        } catch (_e) { return { kind: 'other', audience_page_angle: null }; }
+        if (/(^|\.)cal\.com$/.test(host) || /(^|\.)calendly\.com$/.test(host)) return { kind: 'booking', audience_page_angle: null };
+        if (host === 'github.com') return { kind: 'github', audience_page_angle: null };
+        const entry = projectName ? projIdx[projectName] : null;
+        // Audience-page exact host+path match wins over generic SUBPAGE.
+        if (entry) {
+          for (const ap of entry.audience_pages || []) {
+            if (ap.host === host && ap.path === pathName) {
+              return { kind: 'audience_page', audience_page_angle: ap.angle || null };
+            }
+          }
+          if (entry.website_host && host === entry.website_host) {
+            if (pathName === '/' || pathName === '') return { kind: 'home', audience_page_angle: null };
+            if (/^\/t\//.test(pathName)) return { kind: 'seo', audience_page_angle: null };
+            return { kind: 'subpage', audience_page_angle: null };
+          }
+        }
+        if (pathName === '/' || pathName === '') return { kind: 'other', audience_page_angle: null };
+        if (/^\/t\//.test(pathName)) return { kind: 'seo', audience_page_angle: null };
+        return { kind: 'external', audience_page_angle: null };
+      };
+      for (const d of destinations) {
+        const c = classify(d.target_url, d.project_name);
+        d.kind = c.kind;
+        d.audience_page_angle = c.audience_page_angle;
+      }
+      return json(res, { destinations, window: windowKey, platform: platformFilter || 'all' });
+    })().catch(e => json(res, { error: e.message }, 500));
+  }
+
   // GET /api/top/links - post short links ranked by click count.
   // Queries post_links joined with posts so the content snippet is available.
   // Returns links with >= 1 click, ordered by clicks desc. Used by the "Links"
@@ -5486,11 +6034,23 @@ async function handleApi(req, res) {
           "AND query IS NOT NULL AND length(trim(query)) > 0 " +
       "), " +
       "cand AS ( " +
-        "SELECT 'twitter'  AS platform, c.search_topic AS query, " +
+        // Twitter: c.search_topic is the SEED concept (e.g. "vibe coding") while
+        // twitter_search_attempts.query is the literal X advanced-search string
+        // (e.g. '("vibe coded" OR ...) min_faves:30 since:... -filter:replies').
+        // The two never line up textually, so we associate each candidate to
+        // its parent attempt via (batch_id, project_name) and project the
+        // attempt's `query` as the join key. When one project ran multiple
+        // queries in the same batch (~1.5% of cases), this attributes each
+        // candidate to all of them — known minor over-attribution, acceptable
+        // until we add a per-attempt seed column.
+        "SELECT 'twitter' AS platform, a.query, " +
                "COALESCE(c.matched_project, '(none)') AS project_name, c.post_id " +
         "FROM twitter_candidates c " +
+        "JOIN twitter_search_attempts a " +
+          "ON a.batch_id = c.batch_id " +
+         "AND COALESCE(a.project_name, '(none)') = COALESCE(c.matched_project, '(none)') " +
         "WHERE c.discovered_at >= NOW() - INTERVAL '" + windowHours + " hours' " +
-          "AND c.search_topic IS NOT NULL " +
+          "AND c.batch_id IS NOT NULL " +
         "UNION ALL " +
         "SELECT 'linkedin', c.search_query, COALESCE(c.matched_project, '(none)'), c.post_id " +
         "FROM linkedin_candidates c " +
@@ -5680,7 +6240,7 @@ async function handleApi(req, res) {
     // suppresses the column. SDK and estimate lanes are surfaced separately
     // so the dashboard tooltip can show both, same UX as cost-stats.
     let costByProject = {};
-    let grandCost = 0, grandCostOrch = 0, grandCostEst = 0;
+    let grandCost = 0, grandCostOrch = 0, grandCostEst = 0, grandCostSub = 0;
     if (req.user && req.user.admin) {
       const costSrcParts = [
         "SELECT claude_session_id FROM posts WHERE claude_session_id IS NOT NULL AND posted_at IS NOT NULL",
@@ -5694,24 +6254,24 @@ async function handleApi(req, res) {
       const costWin = "INTERVAL '" + hours + " hours'";
       const costAttributed = [
         "SELECT COALESCE(posts.project_name, '(none)') AS project, " +
-          "sc.per_row_cost, sc.per_row_cost_orchestrator, sc.per_row_cost_estimated " +
+          "sc.per_row_cost, sc.per_row_cost_orchestrator, sc.per_row_cost_estimated, sc.per_row_cost_subagent " +
           "FROM posts LEFT JOIN session_cost sc ON sc.session_id = posts.claude_session_id " +
           "WHERE posts.posted_at >= NOW() - " + costWin + " " +
           "AND posts.our_content <> '(mention - no original post)'",
         "SELECT COALESCE(replies.project_name, '(none)') AS project, " +
-          "sc.per_row_cost, sc.per_row_cost_orchestrator, sc.per_row_cost_estimated " +
+          "sc.per_row_cost, sc.per_row_cost_orchestrator, sc.per_row_cost_estimated, sc.per_row_cost_subagent " +
           "FROM replies LEFT JOIN session_cost sc ON sc.session_id = replies.claude_session_id " +
           "WHERE replies.status='replied' AND replies.replied_at >= NOW() - " + costWin,
         "SELECT COALESCE(dms.target_project, '(none)') AS project, " +
-          "sc.per_row_cost, sc.per_row_cost_orchestrator, sc.per_row_cost_estimated " +
+          "sc.per_row_cost, sc.per_row_cost_orchestrator, sc.per_row_cost_estimated, sc.per_row_cost_subagent " +
           "FROM dms LEFT JOIN session_cost sc ON sc.session_id = dms.claude_session_id " +
           "WHERE dms.status='sent' AND dms.sent_at >= NOW() - " + costWin,
         "SELECT COALESCE(seo_keywords.product, '(none)') AS project, " +
-          "sc.per_row_cost, sc.per_row_cost_orchestrator, sc.per_row_cost_estimated " +
+          "sc.per_row_cost, sc.per_row_cost_orchestrator, sc.per_row_cost_estimated, sc.per_row_cost_subagent " +
           "FROM seo_keywords LEFT JOIN session_cost sc ON sc.session_id = seo_keywords.claude_session_id " +
           "WHERE seo_keywords.completed_at >= NOW() - " + costWin + " AND seo_keywords.page_url IS NOT NULL",
         "SELECT COALESCE(gsc_queries.product, '(none)') AS project, " +
-          "sc.per_row_cost, sc.per_row_cost_orchestrator, sc.per_row_cost_estimated " +
+          "sc.per_row_cost, sc.per_row_cost_orchestrator, sc.per_row_cost_estimated, sc.per_row_cost_subagent " +
           "FROM gsc_queries LEFT JOIN session_cost sc ON sc.session_id = gsc_queries.claude_session_id " +
           "WHERE gsc_queries.completed_at >= NOW() - " + costWin + " AND gsc_queries.page_url IS NOT NULL",
       ];
@@ -5719,15 +6279,17 @@ async function handleApi(req, res) {
         "WITH src AS (" + costSrcParts.join(' UNION ALL ') + "), " +
         "session_counts AS (SELECT claude_session_id, COUNT(*)::int AS rows_in_session FROM src GROUP BY claude_session_id), " +
         "session_cost AS (SELECT cs.session_id, " +
-            "(COALESCE(cs.orchestrator_cost_usd, cs.total_cost_usd) / NULLIF(sc.rows_in_session, 0))::numeric(12,6) AS per_row_cost, " +
+            "(cs.orchestrator_cost_usd / NULLIF(sc.rows_in_session, 0))::numeric(12,6) AS per_row_cost, " +
             "(cs.orchestrator_cost_usd / NULLIF(sc.rows_in_session, 0))::numeric(12,6) AS per_row_cost_orchestrator, " +
-            "(cs.total_cost_usd / NULLIF(sc.rows_in_session, 0))::numeric(12,6) AS per_row_cost_estimated " +
+            "(cs.total_cost_usd / NULLIF(sc.rows_in_session, 0))::numeric(12,6) AS per_row_cost_estimated, " +
+            "(cs.subagent_cost_usd / NULLIF(sc.rows_in_session, 0))::numeric(12,6) AS per_row_cost_subagent " +
           "FROM claude_sessions cs JOIN session_counts sc ON sc.claude_session_id = cs.session_id), " +
         "attributed AS (" + costAttributed.join(' UNION ALL ') + ") " +
         "SELECT project, " +
           "COALESCE(SUM(per_row_cost), 0)::numeric(12,4) AS cost_usd, " +
           "COALESCE(SUM(per_row_cost_orchestrator), 0)::numeric(12,4) AS cost_usd_orchestrator, " +
-          "COALESCE(SUM(per_row_cost_estimated), 0)::numeric(12,4) AS cost_usd_estimated " +
+          "COALESCE(SUM(per_row_cost_estimated), 0)::numeric(12,4) AS cost_usd_estimated, " +
+          "COALESCE(SUM(per_row_cost_subagent), 0)::numeric(12,4) AS cost_usd_subagent " +
         "FROM attributed GROUP BY project";
       try {
         const costRows = await pq(costQ) || [];
@@ -5736,10 +6298,12 @@ async function handleApi(req, res) {
           const c = Number(r.cost_usd) || 0;
           const co = Number(r.cost_usd_orchestrator) || 0;
           const ce = Number(r.cost_usd_estimated) || 0;
-          costByProject[proj] = { cost_usd: c, cost_usd_orchestrator: co, cost_usd_estimated: ce };
+          const cs = Number(r.cost_usd_subagent) || 0;
+          costByProject[proj] = { cost_usd: c, cost_usd_orchestrator: co, cost_usd_estimated: ce, cost_usd_subagent: cs };
           grandCost += c;
           grandCostOrch += co;
           grandCostEst += ce;
+          grandCostSub += cs;
         });
       } catch (e) {
         // Soft fail: log and continue without cost data. Don't block the
@@ -5753,6 +6317,7 @@ async function handleApi(req, res) {
       r.cost_usd = c ? c.cost_usd : 0;
       r.cost_usd_orchestrator = c ? c.cost_usd_orchestrator : 0;
       r.cost_usd_estimated = c ? c.cost_usd_estimated : 0;
+      r.cost_usd_subagent = c ? c.cost_usd_subagent : 0;
       return r;
     };
     projects.forEach(attachCost);
@@ -5791,6 +6356,7 @@ async function handleApi(req, res) {
       grand_cost_usd: grandCost,
       grand_cost_usd_orchestrator: grandCostOrch,
       grand_cost_usd_estimated: grandCostEst,
+      grand_cost_usd_subagent: grandCostSub,
       cost_available: !!(req.user && req.user.admin),
       can_edit_weight: !auth.CLIENT_MODE && !!(req.user && req.user.admin),
       projects,
@@ -6389,6 +6955,19 @@ const HTML = `<!DOCTYPE html>
   #top-pages-container .style-stats-table th,
   #top-pages-container .style-stats-table td { overflow: hidden; text-overflow: ellipsis; white-space: nowrap; padding: 10px 10px; }
   #top-pages-container .style-stats-table td[data-col-key="path"] { white-space: normal; overflow: visible; text-overflow: clip; word-break: break-all; }
+
+  #top-links-container .style-stats-table { table-layout: fixed; }
+  #top-links-container .style-stats-table th,
+  #top-links-container .style-stats-table td { overflow: hidden; text-overflow: ellipsis; white-space: nowrap; padding: 10px 10px; vertical-align: top; }
+  #top-links-container .style-stats-table td[data-col-key="target_url"] { white-space: normal; overflow: visible; text-overflow: clip; word-break: break-all; }
+  #top-links-container .dest-kind-badge { display: inline-block; padding: 2px 8px; border-radius: 999px; font-size: 11px; font-weight: 600; letter-spacing: 0.02em; text-transform: uppercase; }
+  #top-links-container .dest-kind-home     { background: rgba(16,185,129,0.18); color: #10b981; }
+  #top-links-container .dest-kind-subpage  { background: rgba(59,130,246,0.18); color: #3b82f6; }
+  #top-links-container .dest-kind-seo      { background: rgba(168,85,247,0.18); color: #a855f7; }
+  #top-links-container .dest-kind-booking  { background: rgba(245,158,11,0.18); color: #f59e0b; }
+  #top-links-container .dest-kind-github   { background: rgba(148,163,184,0.18); color: #94a3b8; }
+  #top-links-container .dest-kind-external { background: rgba(239,68,68,0.18); color: #ef4444; }
+  #top-links-container .dest-kind-other    { background: rgba(148,163,184,0.18); color: #94a3b8; }
   /* DMs sub-tab */
   #top-dms-container .style-stats-table { table-layout: fixed; }
   #top-dms-container .style-stats-table th,
@@ -6662,6 +7241,24 @@ const HTML = `<!DOCTYPE html>
   .style-stats-pill:hover { border-color: var(--border-strong); background: var(--bg-hover); }
   .style-stats-pill.active { background: var(--accent-panel-bg); border-color: #3b82f6; color: var(--text); }
 
+  /* In-place loading state for stats containers. Dims the previously-rendered
+     grid/table so it stays visible (no layout jump) but reads clearly as stale.
+     Used by loadActivityStats / loadCohortStats / loadStyleStats while their
+     fetches are in flight. */
+  .is-loading { opacity: 0.5; transition: opacity 0.12s linear; pointer-events: none; }
+
+  /* Disable Stats-tab filter pills while any stats fetch is in flight
+     (body.sa-stats-busy) so users can't queue overlapping requests across
+     rapid pill changes. Covers the three pill rows that drive
+     reloadStatsTabSections: stats window + style-stats platform/project. */
+  body.sa-stats-busy #stats-window-pills .style-stats-pill,
+  body.sa-stats-busy #style-stats-platform-pills .style-stats-pill,
+  body.sa-stats-busy #style-stats-project-pills .style-stats-pill {
+    pointer-events: none;
+    opacity: 0.55;
+    cursor: wait;
+  }
+
   @media (max-width: 600px) { .cards { grid-template-columns: 1fr; } .content { padding: 16px; } }
 
   /* Client-mode auth overlay. Non-admin users see the app with admin-only
@@ -7096,6 +7693,11 @@ const HTML = `<!DOCTYPE html>
         <span class="top-subtab-label">DMs</span>
         <span class="top-subtab-sub">prospect chats</span>
       </span>
+      <span class="top-subtab" data-subtab="links" role="tab" aria-selected="false" title="Destination URLs across all posts, ranked by clicks. Homepage vs audience pages vs SEO pages.">
+        <span class="top-subtab-icon" aria-hidden="true">\ud83d\udd17</span>
+        <span class="top-subtab-label">Links</span>
+        <span class="top-subtab-sub">destinations</span>
+      </span>
     </div>
     <div class="top-controls">
       <input id="top-search" class="top-search" type="search" placeholder="Search posts\u2026" />
@@ -7197,6 +7799,9 @@ const HTML = `<!DOCTYPE html>
   <div id="top-dms-container" class="hidden">
     <div class="style-stats-empty">Loading\u2026</div>
   </div>
+  <div id="top-links-container" class="hidden">
+    <div class="style-stats-empty">Loading\u2026</div>
+  </div>
 </div>
 
 <div class="content hidden" id="tab-logs">
@@ -7855,31 +8460,53 @@ function renderResult(run) {
       'salvaged <span style="color:var(--text);font-weight:600;">' + salvPrimary + '</span>' +
       salvBracket +
       '</span>';
-    const tooltip = 'searches: ' + searches +
-      ' / raw tweets: ' + raw +
-      ' / passed score-time cuts: ' + passed +
-      ' / dropped pre-score (already-posted or age>18h): ' + dropped +
-      ' / expired (delta<1 floor): ' + expired +
-      ' / above review cap (delta>=10, gates POST_LIMIT=3): ' + aboveFloor +
-      ' / posted: ' + posted +
-      ' / Phase 0 salvaged into this cycle: ' + salvAttempted +
-      ' (of which posted: ' + salvPosted + ')' +
-      ' / salvageable now (pool size for next cycle): ' + salvageableLive +
-      ' (+' + salvAdded + ' became salvageable / -' + salvDrained + ' drained this run)' +
-      ' / pending end-of-run: ' + queue +
-      ' (start: ' + queueStart + ', +' + qAdded + ' added, -' + qDrained + ' drained = ' +
-        qDrainedPosted + ' posted + ' + qDrainedExpired + ' expired + ' + qDrainedSkipped + ' skipped)' +
-      ' / pending right now (live): ' + pendingLive;
+    // Tooltip is grouped by cycle phase so the funnel reads chronologically.
+    // String.fromCharCode(10) sidesteps the outer HTML backtick template
+    // that strips literal backslash-n escapes (see feedback_server_js_template_regex).
+    // CSS .sa-tooltip white-space:pre-line turns these into line breaks.
+    const NL = String.fromCharCode(10);
+    const tooltip =
+      'Phase 0 (cleanup):' + NL +
+      '• salvaged ' + salvAttempted + ': orphan rows adopted from prior dead cycles (' + salvPosted + ' posted this cycle)' + NL +
+      '• pool for next cycle: ' + salvageableLive + ' salvageable (+' + salvAdded + ' / -' + salvDrained + ' this run)' + NL +
+      NL +
+      'Phase 1 (scrape):' + NL +
+      '• searches ' + searches + ': queries run' + NL +
+      '• raw ' + raw + ': tweets returned' + NL +
+      '• passed ' + passed + ': after dedup + age>18h cuts (' + dropped + ' dropped)' + NL +
+      NL +
+      'Phase 2a (Δ re-score):' + NL +
+      '• expired ' + expired + ': below Δ<1 likes floor' + NL +
+      NL +
+      'Phase 2b (draft + post):' + NL +
+      '• Δ≥10 ' + aboveFloor + ': crossed POST_LIMIT=3 review cap' + NL +
+      '• posted ' + posted + ': shipped' + NL +
+      '• failed ' + failed + ': post errors' + NL +
+      NL +
+      'Pending end-of-run: ' + queue +
+      ' (start ' + queueStart + ', +' + qAdded + ' / -' + qDrained + ' = ' +
+        qDrainedPosted + ' posted, ' + qDrainedExpired + ' expired, ' + qDrainedSkipped + ' skipped)' + NL +
+      'Pending live: ' + pendingLive;
+    // Pill order mirrors the tooltip story: salvaged (Phase 0 input) leads,
+    // then Phase 1 funnel (searches, raw, passed), Phase 2a drop (expired),
+    // Phase 2b decision and outcome (Δ≥10, posted, failed).
     return (
       '<span title="' + tooltip.replace(/"/g, '&quot;') + '" style="display:inline-block;">' +
+        queuePill +
         pill('searches', searches, searches > 0 ? 'var(--text)' : 'var(--muted)') +
         pill('raw', raw, raw > 0 ? 'var(--text)' : 'var(--muted)') +
         pill('passed', passed, passed > 0 ? '#22c55e' : 'var(--muted)') +
         pill('expired', expired, expired > 0 ? 'var(--text)' : 'var(--muted)') +
         pill('Δ≥10', aboveFloor, aboveFloor > 0 ? '#a78bfa' : 'var(--muted)') +
         pill('posted', posted, posted > 0 ? '#22c55e' : 'var(--muted)') +
-        queuePill +
         renderFailedPill() +
+        (Array.isArray(r.projects_worked) && r.projects_worked.length
+          ? '<span style="display:inline-block;margin-right:10px;font-size:12px;color:var(--muted);">'
+            + 'projects '
+            + '<span style="color:var(--text);font-weight:600;">'
+            + r.projects_worked.join(', ')
+            + '</span></span>'
+          : '') +
       '</span>'
     );
   }
@@ -8153,6 +8780,26 @@ function renderResult(run) {
   // the old "posted=18216" pill was the total active-posts count from the
   // DB, which had nothing to do with what the run did. Render the real
   // per-run counters parsed out of the stats log instead.
+  //
+  // 2026-05-18 relabel: split the misleading single "updated" pill into
+  // four explicit pills (scanned / checked / changed / views) so the
+  // operator can read at a glance what the run actually did:
+  //
+  //   scanned -> total rows considered (= polled + every flavor of skip)
+  //   skipped -> rows we deliberately did NOT poll (covered by Step 1
+  //              scrape or stable cooldown). Saves API calls.
+  //   checked -> rows we actually hit the platform API for this run.
+  //   changed -> subset of "checked" where any tracked metric moved.
+  //              Was the original intent of "updated" but the legacy
+  //              field also summed in the Step 1 view-scrape count.
+  //   views   -> Step 1 scrape leg only: rows where the cheap profile-
+  //              page scrape wrote a fresh view count (Reddit) or
+  //              fxtwitter view (Twitter). Distinct from "changed".
+  //   replies -> per-reply rows refreshed (DM-rail follow-ups we made
+  //              on someone else thread, live in the "replies" table).
+  //
+  // Each pill carries data-tooltip so hovering surfaces the meaning
+  // line-by-line via the global .sa-tooltip handler.
   if (run.job_type === 'stats') {
     const checked = r.checked || 0;
     const updated = r.updated || 0;
@@ -8162,19 +8809,71 @@ function renderResult(run) {
     const skipped = r.skipped || 0;
     const failed = r.failed || 0;
     const repliesRefreshed = r.replies_refreshed || 0;
+    // New 2026-05-18 fields; fall back to derived values when the log line
+    // pre-dates the relabel pass (so historical rows still render sanely).
+    const scanned = r.scanned || (checked + skipped) || 0;
+    const changed = r.changed || 0;
+    const viewsRefreshed = r.views_refreshed || 0;
     if (!checked && !updated && !removed && !unavailable && !notFound &&
-        !skipped && !failed && !repliesRefreshed) {
+        !skipped && !failed && !repliesRefreshed &&
+        !scanned && !changed && !viewsRefreshed) {
       return '<span style="color:var(--muted);font-size:12px;">—</span>';
     }
+    // Inline helper: pill with a data-tooltip attribute for per-pill hover
+    // explanations. Plain pill() (above) has no tooltip slot; this is
+    // local to stats-job rendering only.
+    const tipPill = (label, n, color, tip) => {
+      const tipEsc = (tip || '').replace(/"/g, '&quot;');
+      return '<span data-tooltip="' + tipEsc + '" style="display:inline-block;' +
+        'margin-right:10px;font-size:12px;color:var(--muted);cursor:help;">' +
+        label + ' <span style="color:' + (color || 'var(--text)') +
+        ';font-weight:600;">' + n + '</span></span>';
+    };
     return (
-      pill('checked', checked, 'var(--text)') +
-      pill('updated', updated, '#22c55e') +
-      (removed ? pill('removed', removed, '#eab308') : '') +
-      (unavailable ? pill('unavail', unavailable, '#eab308') : '') +
-      (notFound ? pill('not found', notFound, 'var(--muted)') : '') +
-      (skipped ? pill('skipped', skipped, 'var(--muted)') : '') +
-      (repliesRefreshed ? pill('replies', repliesRefreshed, '#3b82f6') : '') +
-      (failed ? pill('failed', failed, '#ef4444') : '')
+      tipPill('scanned', scanned, 'var(--text)',
+        'Total rows the run considered (every active row in the relevant ' +
+        'platform tables). = checked + skipped + bypassed-as-fresh.') +
+      (skipped ? tipPill('skipped', skipped, 'var(--muted)',
+        'Rows we deliberately did NOT poll this run. Two reasons: ' +
+        '(1) already refreshed by the cheap scrape leg within the last 4h, ' +
+        '(2) stable cooldown (2+ scans with no metric change AND older than ' +
+        '3 days). Saves API calls; data is still current.') : '') +
+      tipPill('checked', checked, 'var(--text)',
+        'Rows we actually hit the platform API for this run ' +
+        '(Reddit old.reddit.com JSON / fxtwitter / LinkedIn activity feed). ' +
+        'Includes both successful polls and the ones that errored mid-fetch.') +
+      tipPill('changed', changed, '#22c55e',
+        'Subset of CHECKED where any tracked metric (upvotes, ' +
+        'comments_count, views) actually moved since the prior scan. ' +
+        'The real-activity signal; everything else is no-op polling.') +
+      (viewsRefreshed ? tipPill('views', viewsRefreshed, '#06b6d4',
+        'Rows where the cheap view-scrape leg wrote a fresh view count ' +
+        'this run. Reddit: Step 1 profile-page scrape (sees every ' +
+        'comment + thread on /user/<name>/). Twitter: built-in to the ' +
+        'fxtwitter call. Separate from CHANGED because views can ' +
+        'tick up without upvotes/comments moving.') : '') +
+      (repliesRefreshed ? tipPill('replies', repliesRefreshed, '#3b82f6',
+        'Per-reply rows refreshed: comments we authored AS replies to ' +
+        'someone else reply to our original comment (the DM-rail ' +
+        'follow-up). Live in the replies table, not posts. ' +
+        'Reddit refreshes upvotes + reply-count via batch JSON API. ' +
+        'Twitter also refreshes views via fxtwitter.') : '') +
+      (removed ? tipPill('removed', removed, '#eab308',
+        'Posts newly flagged deleted/removed this run. Reddit: comment ' +
+        'gone from thread JSON for 2+ consecutive scans (deletion_detect_' +
+        'count threshold). LinkedIn: post returned "Post unavailable".') : '') +
+      (unavailable ? tipPill('unavail', unavailable, '#eab308',
+        'LinkedIn only: post explicitly returned a Post Unavailable ' +
+        'string. Subset of REMOVED; rendered as its own pill so an ' +
+        'operator can tell hard-deletion from rate-limit or network noise.') : '') +
+      (notFound ? tipPill('not found', notFound, 'var(--muted)',
+        'LinkedIn only: post is still active on LinkedIn but our specific ' +
+        'comment couldn\'t be located on the activity feed (may have ' +
+        'aged off our visible recent-activity window).') : '') +
+      (failed ? tipPill('failed', failed, '#ef4444',
+        'API errors during the run, broken down by category: 404 not_found, ' +
+        'rate-limited (429), empty / malformed response, or other / network. ' +
+        'Includes step-exit failures from the shell pipeline as well.') : '')
     );
   }
   // seo_expire (delete dead-weight pages): repurposes posted/skipped from the
@@ -8332,7 +9031,7 @@ function buildSeoDetailRows(run) {
   if (!details.length) return '';
   const subRows = details.map(d => {
     const cost = (typeof d.cost_usd === 'number' && d.cost_usd > 0)
-      ? fmtCostCell(d.cost_usd, d.cost_usd_orchestrator, d.cost_usd_estimated)
+      ? fmtCostCell(d.cost_usd, d.cost_usd_orchestrator, d.cost_usd_estimated, d.cost_usd_subagent)
       : '<span style="color:var(--muted);">—</span>';
     const turns = (typeof d.num_turns === 'number' && d.num_turns > 0)
       ? d.num_turns
@@ -8378,6 +9077,85 @@ function buildSeoDetailRows(run) {
   );
 }
 
+// Cost cell for a Job History row. SDK-only mode (2026-05-15): the headline
+// total is the sum of orchestrator_cost_usd across the phases of this run.
+// Sessions whose wrapper didn't capture SDK cost contribute 0; the per-phase
+// breakdown surfaces a "missing SDK" count so the operator can spot pipelines
+// where real spend went unrecorded (the cost cell shows the smaller real
+// number, not an inflated estimate).
+function _jobHistoryCostCell(result) {
+  const fmtLane = (v) => {
+    if (v == null) return 'n/a';
+    const n = Number(v);
+    if (!isFinite(n)) return 'n/a';
+    if (n === 0) return '$0';
+    if (n < 0.01) return '$' + n.toFixed(4);
+    return '$' + n.toFixed(4);
+  };
+  const bd    = result.cost_breakdown;
+  const orch  = result.cost_usd_orchestrator != null ? Number(result.cost_usd_orchestrator) : null;
+  const sub   = result.cost_usd_subagent != null ? Number(result.cost_usd_subagent) : null;
+  const est   = result.cost_usd_estimated != null ? Number(result.cost_usd_estimated) : null;
+  const sessionsAll = bd ? Number(bd.sessions) || 0 : 0;
+  const sessionsWithSdk = bd ? Number(bd.sessions_with_sdk) || 0 : 0;
+  const sessionsMissing = Math.max(0, sessionsAll - sessionsWithSdk);
+  const totalForDisplay = orch != null ? orch : 0;
+  // Header value: "n/a" when no SDK data captured for any of the run's
+  // sessions, otherwise the orchestrator sum (with a "(N missing SDK)"
+  // hint inline when partial).
+  let headerHtml;
+  if (sessionsAll === 0) {
+    headerHtml = '<span style="color:var(--text-very-faint);">n/a</span>';
+  } else if (sessionsWithSdk === 0) {
+    headerHtml = '<span style="color:var(--text-very-faint);">n/a</span>';
+  } else if (sessionsMissing > 0) {
+    headerHtml = fmtCost(totalForDisplay) +
+      ' <span style="color:#eab308;font-size:11px;">(' + sessionsMissing + ' missing)</span>';
+  } else {
+    headerHtml = fmtCost(totalForDisplay);
+  }
+  const lines = [
+    'Cost (SDK orchestrator): ' + (sessionsWithSdk > 0 ? fmtLane(orch) : 'n/a'),
+  ];
+  if (sessionsAll > 0) {
+    lines.push('  Sessions: ' + sessionsAll +
+      '  ·  with SDK data: ' + sessionsWithSdk +
+      '  ·  missing SDK: ' + sessionsMissing);
+  }
+  if (bd && Array.isArray(bd.phases) && bd.phases.length) {
+    lines.push('');
+    lines.push('Per-phase (claude_sessions.script grouping):');
+    const shown = bd.phases.slice(0, 10);
+    for (const p of shown) {
+      const missing = (p.sessions_missing_sdk && p.sessions_missing_sdk > 0)
+        ? ('  [' + p.sessions_missing_sdk + ' missing SDK]')
+        : '';
+      const orchVal = (p.sessions_with_sdk && p.sessions_with_sdk > 0)
+        ? fmtLane(p.orch)
+        : 'n/a';
+      lines.push('  ' + (p.phase || '(unknown)') + '  x' + p.sessions +
+        '  ' + orchVal + missing);
+    }
+    if (bd.phases.length > shown.length) {
+      lines.push('  …(' + (bd.phases.length - shown.length) + ' more)');
+    }
+  }
+  if (typeof result.cost_usd_from_log === 'number') {
+    lines.push('');
+    lines.push('Wrapper shell-log value: ' + fmtLane(result.cost_usd_from_log));
+  }
+  lines.push('');
+  lines.push('Diagnostic-only (local pricing estimate, not actual billing):');
+  lines.push('  Transcript estimate: ' + fmtLane(est));
+  lines.push('  Subagent (est): ' + fmtLane(sub));
+  lines.push('');
+  lines.push('SDK-only mode: shows orchestrator_cost_usd captured by the SDK result event. "missing SDK" = wrapper script didn\\'t pass --output-format json to claude, so no result event = no cost data recorded. Patch the wrapper to fix.');
+  const tip = lines.join('\\n');
+  return '<span data-tooltip="' + escapeHtml(tip) +
+    '" style="cursor:help;border-bottom:1px dotted var(--text-muted);">' +
+    headerHtml + '</span>';
+}
+
 // Stable identity for a job-history row across polls. (script, started_at)
 // is unique in practice; pid is appended as a tiebreaker for the rare case
 // where two parallel fires of the same script land in the same second.
@@ -8406,8 +9184,16 @@ function _jobHistoryRowSig(r) {
   ].join('|');
 }
 function _buildJobHistoryRowGroup(r, idx) {
-  const cost = r.result && r.result.cost_usd;
-  const costCell = cost ? fmtCost(cost) : '<span style="color:var(--muted);">—</span>';
+  // SDK-only mode: render the cost cell whenever the run produced any
+  // claude_sessions rows (cost_breakdown is attached), even if the SDK lane
+  // is NULL across all of them — the tooltip surfaces "missing SDK" so the
+  // operator sees pipelines whose wrappers haven't been patched yet.
+  const hasBreakdown = r.result && r.result.cost_breakdown
+    && Number(r.result.cost_breakdown.sessions) > 0;
+  const hasLogCost = r.result && typeof r.result.cost_usd === 'number';
+  const costCell = (hasBreakdown || hasLogCost)
+    ? _jobHistoryCostCell(r.result)
+    : '<span style="color:var(--muted);">—</span>';
   const hasDetails = Array.isArray(r.details) && r.details.length;
   const caret = hasDetails
     ? '<span class="sa-job-caret" style="display:inline-block;width:12px;color:var(--muted);cursor:pointer;user-select:none;transition:transform 0.15s ease;">&#9656;</span> '
@@ -8972,8 +9758,12 @@ async function saveSettings() {
 }
 
 // Activity tab
-const EVENT_TYPES = ['posted_thread', 'posted_comment', 'replied', 'skipped', 'mention', 'dm_sent', 'dm_reply_sent', 'page_published_serp', 'page_published_gsc', 'page_published_reddit', 'page_published_top', 'page_published_top_post', 'page_published_roundup', 'page_improved', 'page_expired', 'resurrected'];
-const EVENT_LABELS = { posted_thread: 'thread posted', posted_comment: 'comment posted', replied: 'engage replied', skipped: 'engage skipped', mention: 'mention', dm_sent: 'dm sent', dm_reply_sent: 'dm reply', page_published_serp: 'page (serp)', page_published_gsc: 'page (gsc)', page_published_reddit: 'page (reddit)', page_published_top: 'page (top)', page_published_top_post: 'page (top post)', page_published_roundup: 'page (roundup)', page_improved: 'page (improved)', page_expired: 'page expired', resurrected: 'resurrected' };
+// Page-generation event types. Folded into a single 'pages_generated' card
+// in renderActivityStats (with per-pipeline breakdown in the body + tooltip).
+// SQL still emits each subtype so the breakdown is faithful.
+const PAGE_GEN_EVENT_TYPES = ['page_published_twitter', 'page_published_gsc', 'page_published_reddit', 'page_published_top', 'page_published_top_post', 'page_published_roundup', 'page_published_misc', 'page_improved'];
+const EVENT_TYPES = ['posted_thread', 'posted_comment', 'replied', 'skipped', 'mention', 'dm_sent', 'dm_reply_sent', ...PAGE_GEN_EVENT_TYPES, 'page_expired', 'resurrected'];
+const EVENT_LABELS = { posted_thread: 'thread posted', posted_comment: 'comment posted', replied: 'engage replied', skipped: 'engage skipped', mention: 'mention', dm_sent: 'dm sent', dm_reply_sent: 'dm reply', page_published_twitter: 'page (twitter)', page_published_gsc: 'page (gsc)', page_published_reddit: 'page (reddit)', page_published_top: 'page (top)', page_published_top_post: 'page (top post)', page_published_roundup: 'page (roundup)', page_published_misc: 'page (misc)', page_improved: 'page (improved)', page_expired: 'page expired', resurrected: 'resurrected', pages_generated: 'pages generated' };
 const EVENT_DESCRIPTIONS = {
   posted_thread: 'Original thread the bot published (Post Threads job): a new top-level Reddit submission, tweet/X post, LinkedIn post, etc. Identified by thread_url = our_url AND thread_author is empty or matches our_account.',
   posted_comment: 'Comment the bot left on someone else’s thread (Post Comments job): a Reddit/Twitter/LinkedIn/Moltbook/GitHub reply where thread_url ≠ our_url, or where thread_url = our_url but thread_author is someone else (LinkedIn comment job stores the parent URL in both columns).',
@@ -8982,13 +9772,15 @@ const EVENT_DESCRIPTIONS = {
   mention: 'Someone mentioned one of our products on a tracked platform. Detection only, no engagement action.',
   dm_sent: 'New direct-message conversation the bot started with a prospect.',
   dm_reply_sent: 'Follow-up message sent inside an existing DM conversation.',
-  page_published_serp: 'SEO landing page generated from the SERP pipeline (based on ranked search results for target keywords).',
+  pages_generated: 'Total SEO/landing pages produced across every pipeline in this window. Hover the breakdown chips for the per-pipeline counts. The standalone SERP pipeline was unloaded 2026-04-17; any page tagged "twitter" came from the Twitter cycle page-gen A/B lane (twitter_gen_links.py), not a dedicated SERP run.',
+  page_published_twitter: 'SEO landing page generated by the Twitter cycle page-gen A/B lane (twitter_gen_links.py -> generate_page.py --trigger twitter). Source rows have seo_keywords.source=twitter. This is what the legacy "SERP SEO" pill was actually counting after the standalone SERP pipeline was unloaded on 2026-04-17.',
   page_published_gsc: 'SEO page generated from a Google Search Console query the site already gets impressions for.',
   page_published_reddit: 'SEO page generated from a high-intent Reddit thread.',
-  page_published_top: 'SEO page generated for a top-of-funnel ranking opportunity.',
+  page_published_top: 'SEO page generated for a top-of-funnel ranking opportunity (top_pages pipeline).',
   page_published_top_post: 'SEO page generated retroactively for a viral social-autoposter post (>=10k views, last 7d) whose link still points at the homepage. Pipeline also mounts a NewsStrip on the homepage routing existing organic clickthroughs to the new /t/ page.',
   page_published_roundup: 'Roundup or list-style SEO page (comparisons, best-of, alternatives).',
-  page_improved: 'Existing SEO page that was updated or rewritten to improve rankings.',
+  page_published_misc: 'SEO page from a legacy or edge source (existing_page, suggestion:*, competitor:*, topic_template, feature_template, client_blog, etc). Mostly historical; near-zero new volume in 2026-05.',
+  page_improved: 'Existing SEO page that was updated or rewritten to improve rankings (seo_page_improvements table).',
   page_expired: 'SEO page deleted by the daily expire pipeline because it had zero clicks in the last 30 days. The on-disk source file was removed; Next.js now returns 404 for the URL. Logged for audit/revert in seo_expired_pages.',
   resurrected: 'Previously archived or unavailable item brought back into rotation (e.g., a removed post restored after reappearing).',
 };
@@ -9000,7 +9792,10 @@ const ACTIVITY_CAMPAIGN_ORGANIC = '(organic)';
 let _activitySeen = new Set();
 let _activityFirstLoad = true;
 // Activity-tab filters/sort/search are persisted across reloads.
-let _activityTypeFilter = saLoadSet('sa.activity.typeFilter.v2', EVENT_TYPES);
+// Bumped v2 -> v3 on 2026-05-16 when page_published_serp was split into
+// page_published_twitter + page_published_misc; old saved Sets did not
+// include the new types, so they'd silently disappear from the feed.
+let _activityTypeFilter = saLoadSet('sa.activity.typeFilter.v3', EVENT_TYPES);
 let _activityPlatformFilter = saLoadSet('sa.activity.platformFilter.v1', ACTIVITY_PLATFORMS);
 let _activityProjectFilter = saLoadSet('sa.activity.projectFilter.v1', []);
 let _activityKnownProjects = saLoad('sa.activity.knownProjects.v1', []);
@@ -9098,7 +9893,7 @@ function buildActivityFilters() {
     var added;
     if (_activityTypeFilter.has(t)) { _activityTypeFilter.delete(t); el.classList.remove('active'); added = false; }
     else { _activityTypeFilter.add(t); el.classList.add('active'); added = true; }
-    saSaveSet('sa.activity.typeFilter.v2', _activityTypeFilter);
+    saSaveSet('sa.activity.typeFilter.v3', _activityTypeFilter);
     try { window.posthog && window.posthog.capture('filter_toggle', { table: 'activity', dimension: 'type', value: t, action: added ? 'add' : 'remove' }); } catch (er) {}
     _activityPage = 0;
     renderActivity(_lastActivityEvents || []);
@@ -9151,11 +9946,11 @@ function buildActivityFilters() {
       if (a === 'type-all') {
         _activityTypeFilter = new Set(EVENT_TYPES);
         tEl.querySelectorAll('[data-type]').forEach(c => c.classList.add('active'));
-        saSaveSet('sa.activity.typeFilter.v2', _activityTypeFilter);
+        saSaveSet('sa.activity.typeFilter.v3', _activityTypeFilter);
       } else if (a === 'type-none') {
         _activityTypeFilter = new Set();
         tEl.querySelectorAll('[data-type]').forEach(c => c.classList.remove('active'));
-        saSaveSet('sa.activity.typeFilter.v2', _activityTypeFilter);
+        saSaveSet('sa.activity.typeFilter.v3', _activityTypeFilter);
       } else if (a === 'platform-all') {
         _activityPlatformFilter = new Set(ACTIVITY_PLATFORMS);
         pEl.querySelectorAll('[data-platform]').forEach(c => c.classList.add('active'));
@@ -9257,19 +10052,21 @@ function fmtCost(c) {
 //
 // Args (no backticks anywhere; this whole helper sits inside the dashboard
 // HTML template literal, see feedback_server_js_template_regex memory):
-//   displayed     value rendered in the cell. Already prefers SDK, falls
-//                 back to estimate. Source of truth for the text.
+//   displayed     value rendered in the cell. Total = COALESCE(orch,
+//                 estimate) + subagent. Source of truth for the text.
 //   orchestrator  native SDK orchestrator cost (claude_sessions.
 //                 orchestrator_cost_usd, captured from streamRes.
 //                 total_cost_usd). Authoritative for orchestrator billing
 //                 but EXCLUDES Task subagent costs (anthropics/claude-code
-//                 issue #43945).
-//   estimated     manual transcript-derived estimate using local pricing
-//                 tables (claude_sessions.total_cost_usd, written by
+//                 issue #43945). Subagent is now folded in via the 4th arg.
+//   estimated     manual transcript-derived estimate of orchestrator turns
+//                 only (claude_sessions.total_cost_usd, written by
 //                 log_claude_session.py).
-function fmtCostCell(displayed, orchestrator, estimated) {
-  const text = fmtCost(displayed);
-  if (text === '') return '';
+//   subagent      Task/Agent subagent cost from sidechain transcripts +
+//                 sibling subagents/*.jsonl files (claude_sessions.
+//                 subagent_cost_usd). Added on top of the orch/estimate
+//                 lane to form the displayed total.
+function fmtCostCell(displayed, orchestrator, estimated, subagent) {
   const fmtLane = (v) => {
     if (v == null) return 'n/a';
     const n = Number(v);
@@ -9278,13 +10075,23 @@ function fmtCostCell(displayed, orchestrator, estimated) {
     if (n < 0.01) return '$' + n.toFixed(4);
     return '$' + n.toFixed(4);
   };
+  // SDK-only mode (2026-05-15): the displayed value is orchestrator_cost_usd
+  // alone. When that's NULL we render "n/a" — not $0, since the session DID
+  // spend money but the wrapper didn't capture --orchestrator-cost-usd.
+  // Estimate and subagent are diagnostic-only (computed from local pricing
+  // table; not actual billing data).
+  const hasOrch = orchestrator != null && Number.isFinite(Number(orchestrator));
+  const text = hasOrch
+    ? fmtCost(Number(orchestrator))
+    : '<span style="color:var(--text-very-faint);">n/a</span>';
   const lines = [
-    'Orchestrator (SDK): ' + fmtLane(orchestrator),
-    'Estimated (transcript): ' + fmtLane(estimated),
+    'Cost (SDK orchestrator): ' + fmtLane(orchestrator),
     '',
-    'Displayed value prefers the SDK orchestrator cost (native streamRes.total_cost_usd, matches Anthropic billing for the orchestrator session) and falls back to the manual transcript-derived estimate when the SDK value is unavailable.',
+    'Diagnostic-only (not actual billing):',
+    '  Transcript estimate: ' + fmtLane(estimated),
+    '  Subagent (est): ' + fmtLane(subagent),
     '',
-    'Note: orchestrator cost EXCLUDES Task subagent spend (anthropics/claude-code #43945). The estimate uses our local pricing table over the parent transcript only and has the same exclusion.',
+    'Displays Anthropic-reported orchestrator_cost_usd only. "n/a" means the wrapper didn\\'t capture the SDK cost (no --output-format json on the claude call). Transcript estimate and subagent figures are computed locally from a hardcoded pricing table — informational, not billing-accurate on subscription plans.',
   ];
   const tip = lines.join('\\n');
   return '<span data-tooltip="' + escapeHtml(tip) +
@@ -9581,22 +10388,31 @@ function currentStatsProject() {
   const row = document.getElementById('style-stats-project-pills');
   return (row && row.dataset.selected) || 'all';
 }
-function reloadStatsTabSections() {
-  loadActivityStats();
-  loadCohortStats();
-  loadStyleStats();
+// Sets body.sa-stats-busy while the batch of stats fetches kicked off by a
+// single filter change is in flight. CSS uses the class to disable the four
+// pill rows (status window, stats window, style-stats platform/project) so
+// users can't queue overlapping reloads across rapid pill clicks.
+async function reloadStatsTabSections() {
+  document.body.classList.add('sa-stats-busy');
+  const pending = [
+    loadActivityStats(),
+    loadCohortStats(),
+    loadStyleStats(),
+  ];
   // daily-metrics chart now lives on its own Trends tab with its own filter
   // bar; intentionally NOT reloaded on stats-tab window/platform/project
   // of the filter bar.
   const funnelEl = document.getElementById('funnel-stats');
   if (funnelEl && funnelEl.open) {
     if (_lastFunnelPayload) renderFunnelStats(_lastFunnelPayload);
-    else loadFunnelStats(true);
+    else pending.push(loadFunnelStats(true));
   }
   const dmEl = document.getElementById('dm-stats');
-  if (dmEl && dmEl.open) loadDmStats(true);
+  if (dmEl && dmEl.open) pending.push(loadDmStats(true));
   const sqEl = document.getElementById('search-queries-stats');
-  if (sqEl && sqEl.open) loadSearchQueriesStats(true);
+  if (sqEl && sqEl.open) pending.push(loadSearchQueriesStats(true));
+  try { await Promise.allSettled(pending); }
+  finally { document.body.classList.remove('sa-stats-busy'); }
 }
 function syncStatsHeadings() {
   const win = currentStatsWindow();
@@ -9641,7 +10457,63 @@ function renderActivityStats(payload) {
     grandTotal += n;
   });
   if (totalEl) totalEl.textContent = grandTotal + ' events in ' + currentStatsWindow().labelLong;
-  grid.innerHTML = EVENT_TYPES.map(t => {
+  // Fold all page-generation subtypes into a single synthetic 'pages_generated'
+  // card so the grid stops looking like six near-empty SEO cells. The breakdown
+  // row inside the card lists each pipeline + count chip; the hover tooltip on
+  // each chip shows the long description for that pipeline. Card-level "i"
+  // icon shows the umbrella tooltip.
+  const pagesBucket = { total: 0, subtypes: {} };
+  PAGE_GEN_EVENT_TYPES.forEach(t => {
+    const b = byType[t];
+    if (!b) return;
+    pagesBucket.total += b.total;
+    if (b.total > 0) pagesBucket.subtypes[t] = b.total;
+  });
+  const PAGE_GEN_CHIP_LABELS = {
+    page_published_twitter: 'twitter',
+    page_published_gsc: 'gsc',
+    page_published_reddit: 'reddit',
+    page_published_top: 'top',
+    page_published_top_post: 'top post',
+    page_published_roundup: 'roundup',
+    page_published_misc: 'misc',
+    page_improved: 'improved',
+  };
+  // Render order: keep PAGE_GEN_EVENT_TYPES order, then drop the non-page tail
+  // cards (page_expired, resurrected) at the end. Suppress individual page_*
+  // cards since we render the umbrella card instead.
+  const renderOrder = [];
+  EVENT_TYPES.forEach(t => {
+    if (PAGE_GEN_EVENT_TYPES.includes(t)) return; // folded into umbrella
+    renderOrder.push(t);
+  });
+  // Insert the umbrella where the first page card used to live (after dm_reply_sent).
+  const dmIdx = renderOrder.indexOf('dm_reply_sent');
+  if (dmIdx >= 0) renderOrder.splice(dmIdx + 1, 0, '__pages_generated__');
+  else renderOrder.push('__pages_generated__');
+
+  grid.innerHTML = renderOrder.map(t => {
+    if (t === '__pages_generated__') {
+      const total = pagesBucket.total;
+      const subtypes = Object.keys(pagesBucket.subtypes).sort((a, b) => pagesBucket.subtypes[b] - pagesBucket.subtypes[a]);
+      const chips = subtypes.length
+        ? subtypes.map(st => {
+            const lab = PAGE_GEN_CHIP_LABELS[st] || st.replace(/^page_published_/, '');
+            const desc = EVENT_DESCRIPTIONS[st] || '';
+            const titleAttr = desc ? ' data-tooltip="' + escapeHtml(desc) + '"' : '';
+            return '<span class="stat-plat"' + titleAttr + '><span class="stat-plat-text">' + escapeHtml(lab) + '</span><span class="stat-plat-count">' + pagesBucket.subtypes[st] + '</span></span>';
+          }).join('')
+        : '<span style="color:var(--text-very-faint);">\u2014</span>';
+      const umbrellaDesc = EVENT_DESCRIPTIONS.pages_generated || '';
+      const infoIcon = '<span class="stat-card-info" data-tooltip="' + escapeHtml(umbrellaDesc) + '" aria-label="' + escapeHtml(umbrellaDesc) + '">i</span>';
+      return '<div class="stat-card ev-pages-generated' + (total === 0 ? ' zero' : '') + '">' +
+        '<div class="stat-card-head">' +
+          '<span class="stat-card-label">pages generated' + infoIcon + '</span>' +
+          '<span class="stat-card-count">' + total + '</span>' +
+        '</div>' +
+        '<div class="stat-card-breakdown">' + chips + '</div>' +
+      '</div>';
+    }
     const bucket = byType[t];
     const total = bucket.total;
     const plats = Object.keys(bucket.platforms).sort((a, b) => bucket.platforms[b] - bucket.platforms[a]);
@@ -9668,6 +10540,13 @@ function renderActivityStats(payload) {
 }
 
 async function loadActivityStats() {
+  // Immediate visual feedback on filter change. Without this the previously
+  // rendered grid sits frozen until the 9-way UNION returns; on a cold cache
+  // miss that's a couple seconds with zero indication anything is happening.
+  const grid = document.getElementById('stats-grid');
+  const totalEl = document.getElementById('stats-total');
+  if (grid) grid.classList.add('is-loading');
+  if (totalEl) totalEl.textContent = 'loading…';
   try {
     const hours = currentStatsWindow().hours;
     const plat = currentStatsPlatform();
@@ -9678,7 +10557,9 @@ async function loadActivityStats() {
     const res = await fetch('/api/activity/stats?' + params.join('&'));
     const data = await res.json();
     renderActivityStats(data);
-  } catch {}
+  } catch {} finally {
+    if (grid) grid.classList.remove('is-loading');
+  }
 }
 
 // Combined daily-metrics line chart (Trends tab). Fetches 4 endpoints (2
@@ -9698,6 +10579,14 @@ async function loadActivityStats() {
 // to a capture day; expect those lines to sit at 0 until at least two
 // consecutive days of snapshots have accumulated per post.
 let DAILY_METRICS = [
+  // Output volume: posts we made per day, split by type. 'threads' counts
+  // posts where we authored the thread itself; 'comments_made' counts posts
+  // where we engaged on someone else's thread. Both come from the same
+  // /api/posts/per-day endpoint (server returns threads_made + comments_made
+  // alongside posts_made). 'comments_made' is intentionally distinct from
+  // the 'comments' pill below, which counts comments EARNED on our posts.
+  { id: 'threads',         label: 'Threads',           color: '#a855f7', endpoint: '/api/posts/per-day',    valueKey: 'threads_made',     platformAware: true },
+  { id: 'comments_made',   label: 'Comments Made',     color: '#d946ef', endpoint: '/api/posts/per-day',    valueKey: 'comments_made',    platformAware: true },
   { id: 'views',           label: 'Views',             color: '#6366f1', endpoint: '/api/views/per-day',    valueKey: 'views_gained',     platformAware: true },
   { id: 'upvotes',         label: 'Upvotes',           color: '#f97316', endpoint: '/api/upvotes/per-day',  valueKey: 'upvotes_gained',   platformAware: true },
   { id: 'comments',        label: 'Comments',          color: '#14b8a6', endpoint: '/api/comments/per-day', valueKey: 'comments_gained',  platformAware: true },
@@ -10452,6 +11341,8 @@ async function loadDailyMetrics() {
     intoSeries('bookings', bookings.rows, 'bookings_gained');
     intoSeries('cost',     cost.rows,     'cost_usd');
     intoSeries('posts',    posts.rows,    'posts_made');
+    intoSeries('threads',       posts.rows, 'threads_made');
+    intoSeries('comments_made', posts.rows, 'comments_made');
     DAILY_METRICS.filter(m => m.funnel).forEach(m => {
       intoSeries(m.id, funnel.rows, m.valueKey);
     });
@@ -10971,6 +11862,10 @@ function getStyleMeta() {
 }
 
 async function loadStyleStats() {
+  const body = document.getElementById('style-stats-body');
+  const totalEl = document.getElementById('style-stats-total');
+  if (body) body.classList.add('is-loading');
+  if (totalEl) totalEl.textContent = 'loading…';
   try {
     const platformRow = document.getElementById('style-stats-platform-pills');
     const projectRow  = document.getElementById('style-stats-project-pills');
@@ -10985,7 +11880,9 @@ async function loadStyleStats() {
       getStyleMeta(),
     ]);
     renderStyleStats(statsRes, meta);
-  } catch {}
+  } catch {} finally {
+    if (body) body.classList.remove('is-loading');
+  }
 }
 
 // Score-cohort distribution. Buckets posts in the trailing window into
@@ -11088,6 +11985,10 @@ function renderCohortStats(payload) {
 }
 
 async function loadCohortStats() {
+  const body = document.getElementById('cohort-stats-body');
+  const totalEl = document.getElementById('cohort-stats-total');
+  if (body) body.classList.add('is-loading');
+  if (totalEl) totalEl.textContent = 'loading…';
   try {
     const platformRow = document.getElementById('style-stats-platform-pills');
     const projectRow  = document.getElementById('style-stats-project-pills');
@@ -11101,8 +12002,9 @@ async function loadCohortStats() {
     const data = await res.json();
     renderCohortStats(data);
   } catch (e) {
-    const body = document.getElementById('cohort-stats-body');
     if (body) body.innerHTML = '<div class="style-stats-empty">Failed to load cohort stats.</div>';
+  } finally {
+    if (body) body.classList.remove('is-loading');
   }
 }
 
@@ -11637,34 +12539,40 @@ function renderCostStats(payload) {
   const byType = {};
   rows.forEach(r => { byType[r.type] = r; });
   const merged = COST_TYPE_ORDER.map(t => {
-    const r = byType[t] || { count: 0, total_cost_usd: 0, total_cost_usd_orchestrator: 0, total_cost_usd_estimated: 0 };
+    const r = byType[t] || { count: 0, total_cost_usd: 0, total_cost_usd_orchestrator: 0, total_cost_usd_estimated: 0, total_cost_usd_subagent: 0 };
     const count = Number(r.count) || 0;
-    const total = Number(r.total_cost_usd) || 0;
+    // SDK-only: total = orchestrator. Estimate/subagent kept for diagnostic
+    // tooltips. /api/cost/stats SQL emits total_cost_usd = SUM(per_row_cost)
+    // which itself is SUM(orchestrator/rows), so it equals total_cost_usd_
+    // orchestrator by construction in SDK-only mode.
     const totalOrch = r.total_cost_usd_orchestrator != null ? Number(r.total_cost_usd_orchestrator) : null;
     const totalEst  = r.total_cost_usd_estimated != null ? Number(r.total_cost_usd_estimated) : null;
+    const totalSub  = r.total_cost_usd_subagent != null ? Number(r.total_cost_usd_subagent) : null;
+    const total = totalOrch != null ? totalOrch : 0;
     return {
       type: t, label: COST_TYPE_LABELS[t], count: count,
-      total: total, totalOrch: totalOrch, totalEst: totalEst,
+      total: total, totalOrch: totalOrch, totalEst: totalEst, totalSub: totalSub,
       avg: count > 0 ? total / count : 0,
       avgOrch: count > 0 && totalOrch != null ? totalOrch / count : null,
       avgEst:  count > 0 && totalEst  != null ? totalEst  / count : null,
+      avgSub:  count > 0 && totalSub  != null ? totalSub  / count : null,
     };
   });
   const totalCount = merged.reduce(function (a, r) { return a + r.count; }, 0);
   const totalCost  = merged.reduce(function (a, r) { return a + r.total; }, 0);
   const totalOrch  = merged.reduce(function (a, r) { return a + (r.totalOrch || 0); }, 0);
   const totalEst   = merged.reduce(function (a, r) { return a + (r.totalEst  || 0); }, 0);
+  const totalSub   = merged.reduce(function (a, r) { return a + (r.totalSub  || 0); }, 0);
   if (totalEl) {
     totalEl.textContent = '$' + totalCost.toFixed(2) + ' · ' + totalCount.toLocaleString() + ' activit' + (totalCount === 1 ? 'y' : 'ies');
-    // Tooltip on the header pill so users can see both lanes for the
-    // headline figure without expanding the table.
     const tipLines = [
-      'Orchestrator (SDK): $' + totalOrch.toFixed(4),
-      'Estimated (transcript): $' + totalEst.toFixed(4),
+      'Cost (SDK orchestrator): $' + totalOrch.toFixed(4),
       '',
-      'Displayed total prefers the SDK orchestrator cost (native streamRes.total_cost_usd) and falls back to the manual transcript estimate where the SDK value is missing.',
+      'Diagnostic-only (local pricing estimate, not actual billing):',
+      '  Transcript estimate: $' + totalEst.toFixed(4),
+      '  Subagent (est): $' + totalSub.toFixed(4),
       '',
-      'Note: orchestrator cost EXCLUDES Task subagent spend (anthropics/claude-code #43945).',
+      'Anthropic SDK-reported orchestrator_cost_usd. Pipelines whose wrappers don\\'t capture --orchestrator-cost-usd contribute $0 — see the per-phase table for which scripts are missing coverage.',
     ];
     totalEl.setAttribute('data-tooltip', tipLines.join('\\n'));
     totalEl.style.cursor = 'help';
@@ -11677,56 +12585,110 @@ function renderCostStats(payload) {
     return '$' + n.toFixed(2);
   }
   function fmtCount(v) { return (Number(v) || 0).toLocaleString(); }
-  // Wraps a money cell in a span that exposes both cost lanes via tooltip.
-  function moneyCell(displayed, orch, est) {
+  function moneyCell(displayed, orch, est, sub) {
     const tip = [
-      'Orchestrator (SDK): ' + (orch != null ? fmtMoney(orch) : 'n/a'),
-      'Estimated (transcript): ' + (est != null ? fmtMoney(est) : 'n/a'),
+      'Cost (SDK orchestrator): ' + (orch != null ? fmtMoney(orch) : 'n/a'),
       '',
-      'Displayed value prefers SDK; falls back to transcript estimate. Subagent costs excluded (anthropics/claude-code #43945).',
+      'Diagnostic-only (local pricing estimate):',
+      '  Transcript estimate: ' + (est != null ? fmtMoney(est) : 'n/a'),
+      '  Subagent (est): ' + (sub != null ? fmtMoney(sub) : 'n/a'),
     ].join('\\n');
     return '<span data-tooltip="' + escapeHtml(tip) +
       '" style="cursor:help;border-bottom:1px dotted var(--text-muted);">' +
       fmtMoney(displayed) + '</span>';
   }
   const rowsHtml = merged.map(function (r) {
-    const totalCellHtml = moneyCell(r.total, r.totalOrch, r.totalEst);
+    const totalCellHtml = moneyCell(r.total, r.totalOrch, r.totalEst, r.totalSub);
+    const subCellHtml   = r.totalSub != null && r.totalSub > 0 ? fmtMoney(r.totalSub) : '<span style="color:var(--text-very-faint);">$0</span>';
     const avgCellHtml = r.count > 0
-      ? moneyCell(r.avg, r.avgOrch, r.avgEst)
+      ? moneyCell(r.avg, r.avgOrch, r.avgEst, r.avgSub)
       : '&mdash;';
     return '<tr>' +
       '<td>' + escapeHtml(r.label) + '</td>' +
       '<td style="text-align:right;font-variant-numeric:tabular-nums;">' + fmtCount(r.count) + '</td>' +
       '<td style="text-align:right;font-variant-numeric:tabular-nums;">' + totalCellHtml + '</td>' +
+      '<td style="text-align:right;font-variant-numeric:tabular-nums;color:var(--text-muted);">' + subCellHtml + '</td>' +
       '<td style="text-align:right;font-variant-numeric:tabular-nums;color:var(--text-muted);">' + avgCellHtml + '</td>' +
     '</tr>';
   }).join('');
-  const footerTotalHtml = moneyCell(totalCost, totalOrch, totalEst);
+  const footerTotalHtml = moneyCell(totalCost, totalOrch, totalEst, totalSub);
+  const footerSubHtml   = totalSub > 0 ? fmtMoney(totalSub) : '<span style="color:var(--text-very-faint);">$0</span>';
   const footerAvgHtml = totalCount > 0
     ? moneyCell(totalCost / totalCount,
                 totalOrch / totalCount,
-                totalEst  / totalCount)
+                totalEst  / totalCount,
+                totalSub  / totalCount)
     : '&mdash;';
   const footerHtml =
     '<tr style="border-top:2px solid var(--border);font-weight:600;background:var(--bg-subtle);">' +
       '<td>Total</td>' +
       '<td style="text-align:right;font-variant-numeric:tabular-nums;">' + fmtCount(totalCount) + '</td>' +
       '<td style="text-align:right;font-variant-numeric:tabular-nums;">' + footerTotalHtml + '</td>' +
+      '<td style="text-align:right;font-variant-numeric:tabular-nums;">' + footerSubHtml + '</td>' +
       '<td style="text-align:right;font-variant-numeric:tabular-nums;">' + footerAvgHtml + '</td>' +
     '</tr>';
+  // Per-phase (script) breakdown. Same window, separate query — gives the
+  // operator an answer to "which phase of which pipeline is burning cash?"
+  // independent of the activity-type rollup above. A row like
+  // run-twitter-cycle-scan dominating the spend is the signal to investigate.
+  const phases = (payload && payload.phases) || [];
+  let phaseTableHtml = '';
+  if (phases.length) {
+    const phaseRowsHtml = phases.map(function (p) {
+      const orch  = p.total_cost_usd_orchestrator != null ? Number(p.total_cost_usd_orchestrator) : 0;
+      const est   = p.total_cost_usd_estimated != null ? Number(p.total_cost_usd_estimated) : null;
+      const sub   = p.total_cost_usd_subagent != null ? Number(p.total_cost_usd_subagent) : null;
+      const sessions = Number(p.sessions) || 0;
+      const withSdk = Number(p.sessions_with_sdk) || 0;
+      const missing = Math.max(0, sessions - withSdk);
+      // SDK-only: per-phase total = orchestrator sum. Phases with 0% SDK
+      // coverage show $0 with a "(N/N missing)" hint so it's obvious the
+      // wrapper needs patching.
+      const totalCellInner = withSdk > 0
+        ? moneyCell(orch, orch, est, sub)
+        : '<span style="color:var(--text-very-faint);">n/a</span>';
+      const coverageCell = missing === 0
+        ? ('<span style="color:#15803d;">' + sessions + '/' + sessions + '</span>')
+        : ('<span style="color:#b91c1c;" title="' + missing + ' sessions missing SDK cost (wrapper script needs --output-format json on claude call)">' +
+            withSdk + '/' + sessions + '</span>');
+      const perSession = withSdk > 0
+        ? moneyCell(orch / withSdk, orch / withSdk, est != null ? est / sessions : null, sub != null ? sub / sessions : null)
+        : '&mdash;';
+      return '<tr>' +
+        '<td style="font-family:ui-monospace,monospace;font-size:12px;">' + escapeHtml(p.phase || '(unknown)') + '</td>' +
+        '<td style="text-align:right;font-variant-numeric:tabular-nums;">' + fmtCount(sessions) + '</td>' +
+        '<td style="text-align:right;font-variant-numeric:tabular-nums;font-size:11px;">' + coverageCell + '</td>' +
+        '<td style="text-align:right;font-variant-numeric:tabular-nums;">' + totalCellInner + '</td>' +
+        '<td style="text-align:right;font-variant-numeric:tabular-nums;color:var(--text-muted);">' + perSession + '</td>' +
+      '</tr>';
+    }).join('');
+    phaseTableHtml =
+      '<div style="font-size:12px;font-weight:600;padding:12px 2px 4px;color:var(--text-secondary);">Cost per Phase (Claude session script)</div>' +
+      '<table class="style-stats-table">' +
+        '<thead><tr>' +
+          '<th style="text-align:left;">Phase</th>' +
+          '<th style="text-align:right;">Sessions</th>' +
+          '<th style="text-align:right;" title="Sessions with SDK cost captured / total sessions. Red = wrapper needs patching.">SDK coverage</th>' +
+          '<th style="text-align:right;">Cost (SDK)</th>' +
+          '<th style="text-align:right;">Cost per Session</th>' +
+        '</tr></thead>' +
+        '<tbody>' + phaseRowsHtml + '</tbody>' +
+      '</table>';
+  }
   body.innerHTML =
     '<table class="style-stats-table">' +
       '<thead><tr>' +
         '<th style="text-align:left;">Type</th>' +
         '<th style="text-align:right;">Activities</th>' +
         '<th style="text-align:right;">Total Cost</th>' +
+        '<th style="text-align:right;">Subagent</th>' +
         '<th style="text-align:right;">Cost per Activity</th>' +
       '</tr></thead>' +
       '<tbody>' + rowsHtml + footerHtml + '</tbody>' +
     '</table>' +
+    phaseTableHtml +
     '<div style="font-size:11px;color:var(--text-muted);padding:8px 2px 2px;">' +
-      'Cost is Claude session spend split evenly across the activity rows each session produced. ' +
-      'Totals here exclude skipped replies, resurrected posts, DM replies, and mentions.' +
+      'SDK-only mode (2026-05-15): cost = Anthropic\\'s orchestrator_cost_usd from the SDK result event. Pipelines whose wrappers don\\'t pass --output-format json to claude show $0 with a red SDK-coverage cell — that\\'s a missing-data signal, not a real $0. Transcript estimate and Task-subagent figures (tooltip-only) come from a local pricing table and don\\'t reflect subscription billing.' +
     '</div>';
 }
 
@@ -11779,6 +12741,10 @@ let _topDmsTableState = { sortField: 'rank', sortDir: 'asc', filters: {} };
 let _topDmsLoaded = false;
 let _topDmsLoading = false;
 let _topDmsPayload = null;
+let _topLinksTableState = { sortField: 'real_clicks', sortDir: 'desc', filters: {} };
+let _topLinksLoaded = false;
+let _topLinksLoading = false;
+let _topLinksPayload = null;
 let _topDmDir = saLoad('sa.top.dmDir.v1', 'all');
 let _topDmInterest = saLoad('sa.top.dmInterest.v1', 'all');
 let _topDmMode = saLoad('sa.top.dmMode.v1', 'all');
@@ -12311,6 +13277,7 @@ const TOP_SUBTAB_HELP = {
   comments: 'Top comments your accounts have left under other people’s threads, ranked by reach and reactions.',
   pages: 'Top landing/SEO pages on your sites this period, ranked by pageviews.',
   dms: 'Direct message conversations with prospects, ranked by recent activity.',
+  links: 'Destination URLs across all posts, ranked by clicks. One row per unique target URL (homepage vs audience pages vs SEO pages vs booking).',
 };
 function syncTopSubtabHelp() {
   const el = document.getElementById('top-subtab-help');
@@ -12363,12 +13330,14 @@ function initTopFilters() {
     saveDashboardWindow(_topWindow);
     if (_topSubtab === 'pages') loadTopPages(true);
     else if (_topSubtab === 'dms') { _topDmOffset = 0; loadTopDms(true); }
+    else if (_topSubtab === 'links') loadTopLinks(true);
     else loadTopPosts(true);
   });
   wireTopPillRow('top-platform-pills', (v) => {
     _topPlatform = v || 'all';
     saSave('sa.top.platform.v1', _topPlatform);
     if (_topSubtab === 'dms') { _topDmOffset = 0; loadTopDms(true); }
+    else if (_topSubtab === 'links') loadTopLinks(true);
     else loadTopPosts(true);
   });
   wireTopPillRow('top-project-pills', (v) => {
@@ -12376,6 +13345,7 @@ function initTopFilters() {
     saSave('sa.top.project.v1', _topProject);
     if (_topSubtab === 'pages') renderTopPagesFromCache();
     else if (_topSubtab === 'dms') { if (_topDmsPayload) renderTopDms(_topDmsPayload); }
+    else if (_topSubtab === 'links') loadTopLinks(true);
     else loadTopPosts(true); // refetch so the SQL LIMIT applies AFTER project filter
   });
   wireTopPillRow('top-campaign-pills', (v) => {
@@ -12494,6 +13464,7 @@ function applyTopSubtabState(sub, loadData) {
   const pagesC = document.getElementById('top-pages-container');
   const pagesUnknownC = document.getElementById('top-pages-unknown-container');
   const dmsC   = document.getElementById('top-dms-container');
+  const linksC = document.getElementById('top-links-container');
   const platRowEl = document.getElementById('top-platform-pills');
   const projRowEl = document.getElementById('top-project-pills');
   const campRowEl = document.getElementById('top-campaign-pills');
@@ -12520,6 +13491,7 @@ function applyTopSubtabState(sub, loadData) {
   if (sub === 'pages') {
     if (postsC) postsC.classList.add('hidden');
     if (dmsC) dmsC.classList.add('hidden');
+    if (linksC) linksC.classList.add('hidden');
     if (pagesC) pagesC.classList.remove('hidden');
     if (pagesUnknownC) pagesUnknownC.classList.remove('hidden');
     if (platRowEl) platRowEl.classList.add('hidden');
@@ -12533,6 +13505,7 @@ function applyTopSubtabState(sub, loadData) {
     if (postsC) postsC.classList.add('hidden');
     if (pagesC) pagesC.classList.add('hidden');
     if (pagesUnknownC) pagesUnknownC.classList.add('hidden');
+    if (linksC) linksC.classList.add('hidden');
     if (dmsC) dmsC.classList.remove('hidden');
     if (platRowEl) platRowEl.classList.remove('hidden');
     if (srcRowEl) srcRowEl.classList.add('hidden');
@@ -12546,10 +13519,29 @@ function applyTopSubtabState(sub, loadData) {
       searchElDm.value = _topDmSearch || '';
     }
     if (loadData) loadTopDms(true);
+  } else if (sub === 'links') {
+    if (postsC) postsC.classList.add('hidden');
+    if (pagesC) pagesC.classList.add('hidden');
+    if (pagesUnknownC) pagesUnknownC.classList.add('hidden');
+    if (dmsC) dmsC.classList.add('hidden');
+    if (linksC) linksC.classList.remove('hidden');
+    if (platRowEl) platRowEl.classList.remove('hidden');
+    if (srcRowEl) srcRowEl.classList.add('hidden');
+    if (campRowEl) campRowEl.classList.add('hidden');
+    setDmRowsHidden(true);
+    setLinkPillHidden(true);
+    if (totalEl) totalEl.textContent = '';
+    const searchElLinks = document.getElementById('top-search');
+    if (searchElLinks) {
+      searchElLinks.placeholder = 'Filter destinations by URL\u2026';
+      searchElLinks.value = (_topLinksTableState && _topLinksTableState.globalQuery) || '';
+    }
+    if (loadData) loadTopLinks(true);
   } else {
     if (pagesC) pagesC.classList.add('hidden');
     if (pagesUnknownC) pagesUnknownC.classList.add('hidden');
     if (dmsC) dmsC.classList.add('hidden');
+    if (linksC) linksC.classList.add('hidden');
     if (postsC) postsC.classList.remove('hidden');
     if (platRowEl) platRowEl.classList.remove('hidden');
     if (srcRowEl) srcRowEl.classList.add('hidden');
@@ -12729,6 +13721,151 @@ async function loadTopPages(force) {
   }
 }
 
+// Render the colored kind badge for a destination row. The server's
+// /api/top/destinations endpoint classifies each row into one of seven kind
+// buckets (home / subpage / audience_page / seo / booking / github /
+// external / other) by reading config.json, so the client just looks up
+// the label and color here.
+function destinationKindBadge(kind, audienceAngle) {
+  const map = {
+    home: { cls: 'dest-kind-home', label: 'HOME' },
+    subpage: { cls: 'dest-kind-subpage', label: 'SUBPAGE' },
+    audience_page: { cls: 'dest-kind-subpage', label: 'AUDIENCE' },
+    seo: { cls: 'dest-kind-seo', label: 'SEO' },
+    booking: { cls: 'dest-kind-booking', label: 'BOOKING' },
+    github: { cls: 'dest-kind-github', label: 'GITHUB' },
+    external: { cls: 'dest-kind-external', label: 'EXT' },
+    other: { cls: 'dest-kind-other', label: 'OTHER' },
+  };
+  const m = map[kind] || map.other;
+  let label = m.label;
+  if (kind === 'audience_page' && audienceAngle) {
+    label = 'AUDIENCE: ' + audienceAngle;
+  }
+  return '<span class="dest-kind-badge ' + m.cls + '" title="' + escapeHtml(kind) + '">' + escapeHtml(label) + '</span>';
+}
+
+async function loadTopLinks(force) {
+  if (_topLinksLoading) return;
+  const container = document.getElementById('top-links-container');
+  if (!_topLinksPayload && container) {
+    container.innerHTML = '<div class="style-stats-empty">Loading\u2026</div>';
+  }
+  _topLinksLoading = true;
+  try {
+    const params = new URLSearchParams();
+    if (_topWindow) params.set('window', _topWindow);
+    if (_topPlatform && _topPlatform !== 'all') params.set('platform', _topPlatform);
+    if (_topProject && _topProject !== 'all') params.set('project', _topProject);
+    const res = await fetch('/api/top/destinations?' + params.toString());
+    const data = await res.json();
+    _topLinksPayload = data;
+    renderTopLinks(data);
+    _topLinksLoaded = true;
+  } catch (e) {
+    if (container) container.innerHTML = '<div class="style-stats-empty">Failed to load.</div>';
+  } finally {
+    _topLinksLoading = false;
+  }
+}
+
+function renderTopLinks(payload) {
+  const container = document.getElementById('top-links-container');
+  if (!container) return;
+  const totalEl = document.getElementById('top-total');
+  const fmt = n => (Number(n) || 0).toLocaleString();
+  const dests = Array.isArray(payload && payload.destinations) ? payload.destinations : [];
+  if (!dests.length) {
+    container.innerHTML = '<div class="style-stats-empty">No destinations in this window yet. Posts with linked URLs will show up here once they accrue clicks.</div>';
+    if (totalEl) totalEl.textContent = '';
+    return;
+  }
+  // Roll up real_clicks vs the legacy/backfill columns: prefer plc.real_clicks
+  // (post-2026-05-07 per-hit log), fall back to pl.real_clicks (PostHog
+  // backfill for older rows), final fallback pl.clicks (legacy counter).
+  const rows = dests.map(d => {
+    const kind = d.kind || 'other';
+    const realClicks = Number(d.real_clicks || 0);
+    const backfillReal = Number(d.backfill_real || 0);
+    const legacyClicks = Number(d.legacy_clicks || 0);
+    const botClicks = Number(d.bot_clicks || 0);
+    const effectiveClicks = realClicks > 0 ? realClicks : (backfillReal > 0 ? backfillReal : legacyClicks);
+    return {
+      target_url: d.target_url || '',
+      project_name: d.project_name || '',
+      platform: d.platform || '',
+      kind,
+      audience_page_angle: d.audience_page_angle || null,
+      kind_label: kind,
+      posts: Number(d.posts || 0),
+      codes: Number(d.codes || 0),
+      real_clicks: realClicks,
+      backfill_real: backfillReal,
+      legacy_clicks: legacyClicks,
+      bot_clicks: botClicks,
+      effective_clicks: effectiveClicks,
+      first_minted_at: d.first_minted_at || null,
+      last_click_at: d.last_click_at || null,
+    };
+  });
+  if (totalEl) {
+    const totalClicks = rows.reduce((a, r) => a + r.effective_clicks, 0);
+    totalEl.textContent = rows.length + ' destination' + (rows.length === 1 ? '' : 's') + ' \u00b7 ' + fmt(totalClicks) + ' clicks';
+  }
+  const fmtUrl = (_v, r) => {
+    const safe = escapeHtml(r.target_url);
+    return '<a href="' + safe + '" target="_blank" rel="noopener" class="top-post-link">'
+      + destinationKindBadge(r.kind, r.audience_page_angle)
+      + ' <span style="word-break:break-all">' + safe + '</span>'
+      + '</a>';
+  };
+  const fmtAgo = (v) => {
+    if (!v) return '\u2014';
+    try {
+      const d = new Date(v);
+      const diff = Date.now() - d.getTime();
+      const days = Math.floor(diff / 86400000);
+      if (days < 1) {
+        const hours = Math.floor(diff / 3600000);
+        if (hours < 1) {
+          const mins = Math.max(0, Math.floor(diff / 60000));
+          return mins + 'm ago';
+        }
+        return hours + 'h ago';
+      }
+      return days + 'd ago';
+    } catch (_e) { return '\u2014'; }
+  };
+  const fmtClicks = (v, r) => {
+    const main = fmt(r.effective_clicks);
+    const bits = [];
+    if (r.real_clicks > 0) bits.push(r.real_clicks + ' real');
+    else if (r.backfill_real > 0) bits.push(r.backfill_real + ' backfill');
+    else if (r.legacy_clicks > 0) bits.push(r.legacy_clicks + ' legacy');
+    if (r.bot_clicks > 0) bits.push(r.bot_clicks + ' bot');
+    const sub = bits.length ? '<div style="font-size:11px;color:var(--text-secondary)">' + escapeHtml(bits.join(' \u00b7 ')) + '</div>' : '';
+    return '<div style="font-weight:600">' + main + '</div>' + sub;
+  };
+  const columns = [
+    { key: 'target_url',    label: 'Destination', type: 'text',    align: 'left',  widthPct: 48, formatter: fmtUrl },
+    { key: 'project_name',  label: 'Project',     type: 'text',    align: 'left',  widthPct: 10, formatter: v => escapeHtml(v) },
+    { key: 'platform',      label: 'Platform',    type: 'text',    align: 'left',  widthPct: 8,  formatter: v => escapeHtml(v) },
+    { key: 'posts',         label: 'Posts',       type: 'numeric', align: 'right', widthPct: 7,  formatter: fmt },
+    { key: 'codes',         label: 'Codes',       type: 'numeric', align: 'right', widthPct: 7,  formatter: fmt },
+    { key: 'effective_clicks', label: 'Clicks',   type: 'numeric', align: 'right', widthPct: 12, formatter: fmtClicks },
+    { key: 'last_click_at', label: 'Last click',  type: 'date',    align: 'right', widthPct: 8,  formatter: fmtAgo },
+  ];
+  container.innerHTML = '';
+  mountSortableTable({
+    containerId: 'top-links-container',
+    rows,
+    state: _topLinksTableState,
+    storageKey: 'sa.topLinksTable.v1',
+    columns,
+    emptyMessage: 'No destinations in this window yet.',
+  });
+}
+
 function dmClassBadge(dm) {
   const status = String(dm.conversation_status || '').toLowerCase();
   const interest = String(dm.interest_level || '').toLowerCase();
@@ -13960,6 +15097,7 @@ function renderProjectStatus(data, opts) {
   const grandCost = Number(data && data.grand_cost_usd) || 0;
   const grandCostOrch = Number(data && data.grand_cost_usd_orchestrator) || 0;
   const grandCostEst = Number(data && data.grand_cost_usd_estimated) || 0;
+  const grandCostSub = Number(data && data.grand_cost_usd_subagent) || 0;
   // Money formatter mirrors fmtCost: $0, $0.0042, $12.34.
   const fmtMoney = (v) => {
     const n = Number(v) || 0;
@@ -13969,12 +15107,17 @@ function renderProjectStatus(data, opts) {
   };
   // Money cell with tooltip exposing SDK + estimate lanes, same UX as
   // moneyCell in renderCostStats so operators see consistent numbers.
-  const costCell = (displayed, orch, est, opts) => {
+  const costCell = (displayed, orch, est, sub, opts) => {
+    // SDK-only mode: displayed value comes from orchestrator_cost_usd; the
+    // estimate and subagent are diagnostic-only (local pricing table).
     const tip = [
-      'Orchestrator (SDK): ' + (orch != null ? fmtMoney(orch) : 'n/a'),
-      'Estimated (transcript): ' + (est != null ? fmtMoney(est) : 'n/a'),
+      'Cost (SDK orchestrator): ' + (orch != null ? fmtMoney(orch) : 'n/a'),
+      '',
+      'Diagnostic-only (local pricing estimate, not actual billing):',
+      '  Transcript estimate: ' + (est != null ? fmtMoney(est) : 'n/a'),
+      '  Subagent (est): ' + (sub != null ? fmtMoney(sub) : 'n/a'),
       '',
-      'Displayed value prefers SDK; falls back to transcript estimate. Subagent costs excluded (anthropics/claude-code #43945).',
+      'Anthropic SDK-reported cost only. "n/a" or $0 means the wrapper didn\\'t capture --orchestrator-cost-usd (no --output-format json on the claude call) for one or more sessions in this window. Subagent and transcript estimates are computed from a local pricing table and are not billing-accurate on subscription plans.',
     ].join('\\n');
     const style = 'text-align:right;font-variant-numeric:tabular-nums;' + (opts && opts.extra || '');
     const inner = '<span data-tooltip="' + escapeHtml(tip) +
@@ -13990,10 +15133,13 @@ function renderProjectStatus(data, opts) {
       : base;
     if (costAvailable) {
       const tipLines = [
-        'Orchestrator (SDK): ' + fmtMoney(grandCostOrch),
-        'Estimated (transcript): ' + fmtMoney(grandCostEst),
+        'Cost (SDK orchestrator): ' + fmtMoney(grandCostOrch),
         '',
-        'Total Claude session cost across all activity rows (posts, comments, DMs, SEO pages) attributed to projects in this window. Same attribution model as Cost per Activity.',
+        'Diagnostic-only (local pricing estimate, not actual billing):',
+        '  Transcript estimate: ' + fmtMoney(grandCostEst),
+        '  Subagent (est): ' + fmtMoney(grandCostSub),
+        '',
+        'Anthropic SDK-reported orchestrator_cost_usd across all activity rows in this window. Per-session attribution is the session\\'s cost split evenly across its activity rows. Pipelines whose wrappers don\\'t pass --output-format json to claude contribute $0.',
       ];
       totalEl.setAttribute('data-tooltip', tipLines.join('\\n'));
       totalEl.style.cursor = 'help';
@@ -14072,7 +15218,7 @@ function renderProjectStatus(data, opts) {
       : nameCell;
     const totalCell = cellWithShare(r.total, grandTotal, targetShare, { extra: 'font-weight:600;', showZeroShare: true });
     const costCellHtml = costAvailable
-      ? costCell(Number(r.cost_usd) || 0, Number(r.cost_usd_orchestrator) || 0, Number(r.cost_usd_estimated) || 0, { extra: 'color:var(--text-secondary);' })
+      ? costCell(Number(r.cost_usd) || 0, Number(r.cost_usd_orchestrator) || 0, Number(r.cost_usd_estimated) || 0, Number(r.cost_usd_subagent) || 0, { extra: 'color:var(--text-secondary);' })
       : '';
     const weightVal = Number(r.weight) || 0;
     const editable = canEditWeight && (!r.unassigned || r.configured);
@@ -14108,7 +15254,7 @@ function renderProjectStatus(data, opts) {
     '<td style="text-align:right;font-variant-numeric:tabular-nums;">' + (Number(totals[p]) || 0) + '</td>'
   ).join('');
   const footerCostCell = costAvailable
-    ? costCell(grandCost, grandCostOrch, grandCostEst, { extra: 'font-weight:600;' })
+    ? costCell(grandCost, grandCostOrch, grandCostEst, grandCostSub, { extra: 'font-weight:600;' })
     : '';
   const footerHtml =
     '<tr style="border-top:2px solid var(--border);font-weight:600;background:var(--bg-subtle);">' +
@@ -14451,7 +15597,7 @@ function renderActivity(events) {
         '</div>' +
       '</td>' +
       '<td class="activity-summary">' + summaryHtml + '</td>' +
-      '<td class="sa-admin-only" style="text-align:right;font-variant-numeric:tabular-nums;color:var(--text-secondary);">' + fmtCostCell(e.cost_usd, e.cost_usd_orchestrator, e.cost_usd_estimated) + '</td>' +
+      '<td class="sa-admin-only" style="text-align:right;font-variant-numeric:tabular-nums;color:var(--text-secondary);">' + fmtCostCell(e.cost_usd, e.cost_usd_orchestrator, e.cost_usd_estimated, e.cost_usd_subagent) + '</td>' +
       '<td style="text-align:center;">' + renderDeleteBtnHtml(e) + '</td>' +
     '</tr>';
   }).join('');
@@ -15013,6 +16159,25 @@ function renderHtml() {
     .replace('__SA_POSTHOG_CONFIG_PLACEHOLDER__', JSON.stringify(posthogWebConfig()));
 }
 
+function renderTikTokOauthCallback(rawUrl) {
+  const u = new URL(rawUrl, 'http://localhost');
+  const code = u.searchParams.get('code') || '';
+  const state = u.searchParams.get('state') || '';
+  const scopes = u.searchParams.get('scopes') || '';
+  const err = u.searchParams.get('error') || '';
+  const errDesc = u.searchParams.get('error_description') || '';
+  const esc = (s) => String(s).replace(/[&<>"']/g, (c) => ({'&':'&amp;','<':'&lt;','>':'&gt;','"':'&quot;',"'":'&#39;'}[c]));
+  const ok = !!code && !err;
+  const banner = ok
+    ? '<div style="background:#ecfdf5;color:#065f46;padding:12px 16px;border-radius:8px;margin-bottom:24px">Authorization successful. Copy the code below and paste it into the local oauth_helper.py script.</div>'
+    : `<div style="background:#fef2f2;color:#991b1b;padding:12px 16px;border-radius:8px;margin-bottom:24px">Authorization error: ${esc(err || 'no code in callback')}${errDesc ? ` &mdash; ${esc(errDesc)}` : ''}</div>`;
+  const codeBox = ok ? `<div><label style="display:block;font-size:13px;color:#52525b;margin-bottom:6px">Authorization code</label><textarea id="tt-code" readonly style="width:100%;font-family:ui-monospace,SFMono-Regular,monospace;font-size:14px;padding:10px;border:1px solid #d4d4d8;border-radius:8px;background:#fafafa" rows="3" onclick="this.select()">${esc(code)}</textarea><button id="tt-copy" style="margin-top:8px;padding:8px 14px;background:#18181b;color:#fff;border:0;border-radius:6px;cursor:pointer;font-size:13px">Copy</button></div><script>(function(){var b=document.getElementById('tt-copy'),t=document.getElementById('tt-code');if(b&&t){b.addEventListener('click',function(){navigator.clipboard.writeText(t.value).then(function(){b.textContent='Copied';setTimeout(function(){b.textContent='Copy'},1500)})})}})();</script>` : '';
+  const next = ok ? `<details style="margin-top:24px"><summary style="cursor:pointer;color:#27272a;font-weight:500">Next steps</summary><pre style="background:#0a0a0a;color:#fafafa;padding:16px;border-radius:8px;overflow:auto;font-size:13px;line-height:1.6;margin-top:12px">cd ~/social-autoposter
+python3 scripts/tiktok/oauth_helper.py exchange '${esc(code)}'</pre><p style="color:#52525b;font-size:13px;margin-top:12px">This exchanges the code for an access token + refresh token via the TikTok API using the client secret stored in your local macOS keychain, then writes the tokens to <code>~/tiktok-content-api/.env</code>.</p></details>` : '';
+  const meta = ok ? `<dl style="display:grid;grid-template-columns:max-content 1fr;gap:6px 16px;margin-top:24px;font-size:13px;color:#52525b"><dt>state</dt><dd style="font-family:ui-monospace,SFMono-Regular,monospace">${esc(state) || '<i>(none)</i>'}</dd><dt>scopes</dt><dd style="font-family:ui-monospace,SFMono-Regular,monospace">${esc(scopes) || '<i>(none)</i>'}</dd></dl>` : '';
+  return `<!doctype html><html><head><meta charset="utf-8"><title>TikTok OAuth callback &middot; Meditation Fellow Studio</title><meta name="robots" content="noindex"><meta name="viewport" content="width=device-width,initial-scale=1"><style>body{font-family:-apple-system,BlinkMacSystemFont,Inter,system-ui,sans-serif;max-width:680px;margin:48px auto;padding:0 24px;color:#18181b;background:#fff;line-height:1.5}h1{font-size:24px;font-weight:600;margin:0 0 24px;letter-spacing:-0.01em}code{font-family:ui-monospace,SFMono-Regular,monospace;background:#f4f4f5;padding:1px 6px;border-radius:4px;font-size:0.9em}</style></head><body><h1>TikTok OAuth callback</h1>${banner}${codeBox}${meta}${next}</body></html>`;
+}
+
 // --- Server ---
 
 const server = http.createServer((req, res) => {
@@ -15038,6 +16203,14 @@ const server = http.createServer((req, res) => {
     Promise.resolve(handleApi(req, res)).catch(e => {
       try { json(res, { error: e.message || String(e) }, 500); } catch {}
     });
+  } else if (pathname === '/oauth/tiktok/callback') {
+    // Minimal TikTok OAuth landing page. We do not exchange the code here
+    // because the dashboard Cloud Run service does not carry the TikTok
+    // client_secret. The operator pastes the displayed code into the local
+    // scripts/tiktok/oauth_helper.py script which does the exchange via
+    // keychain-stored credentials and writes ~/tiktok-content-api/.env.
+    res.writeHead(200, { 'Content-Type': 'text/html; charset=utf-8', 'Cache-Control': 'no-store' });
+    res.end(renderTikTokOauthCallback(req.url));
   } else {
     res.writeHead(404);
     res.end('Not found');