@m13v/s4l 1.6.197-rc.10

package/scripts/project_stats_json.py

@@ -0,0 +1,1563 @@
+#!/usr/bin/env python3
+"""JSON wrapper around project_stats.py for the dashboard /api/funnel/stats endpoint.
+
+Emits a single JSON object on stdout: { generated_at, days, projects: [ ... ], overall }.
+Keeps project_stats.py untouched (it is chflags uchg-locked).
+"""
+
+import json
+import os
+import re
+import socket
+import sys
+import time
+import urllib.error
+import urllib.parse
+import urllib.request
+from concurrent.futures import ThreadPoolExecutor
+from datetime import datetime, timedelta, timezone
+
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+
+import project_stats as ps
+from project_slugs import bookings_require_utm as _bookings_require_utm
+
+
+_PAGE_FILENAMES = ("page.tsx", "page.ts", "page.jsx", "page.js", "page.mdx", "page.md")
+
+
+def _normalize_platform(p):
+    """Lowercase + alias 'x' -> 'twitter'. Empty / 'all' / None -> '' (no filter).
+
+    Matches the same normalization used by /api/style/stats so the
+    project final stats table speaks the same vocabulary as the
+    engagement-style table when the dashboard's platform pill is set.
+    """
+    if not p:
+        return ""
+    v = str(p).strip().lower()
+    if v in ("", "all"):
+        return ""
+    return "twitter" if v == "x" else v
+
+
+def _platform_sql_clause(platform, table_alias=""):
+    """Return an SQL fragment (string, no placeholders) that:
+
+    1. Filters to the given platform when one is provided (empty = no filter).
+
+    Mentions live in the dedicated `mentions` table now (2026-05-23 cutover);
+    no posts-level filter needed. Previously this clause excluded placeholder
+    `posts` rows where our_content = '(mention - no original post)', which is
+    no longer present after migrate_mentions_out_of_posts.py --commit-delete.
+
+    Folds the 'x' -> 'twitter' alias inside the SQL so reddit/linkedin/twitter
+    all just work. Caller is responsible for placement inside the WHERE.
+    """
+    if not platform:
+        return ""
+    prefix = (table_alias + ".") if table_alias else ""
+    # Safe: platform has already passed the [a-z0-9_]{1,32} regex in the caller.
+    return (
+        " AND LOWER(CASE WHEN LOWER(" + prefix + "platform)='x' "
+        "THEN 'twitter' ELSE " + prefix + "platform END) = '" + platform + "'"
+    )
+
+
+# Synthetic project name for rows in `posts` where project_name IS NULL.
+# Keeps off-config / un-tagged posts visible on the dashboard without
+# polluting real project rows. Chosen to be unambiguous and distinct from
+# any historical 'General' rows the table once had.
+SYNTHETIC_NO_PROJECT_NAME = "(no project)"
+
+
+def _project_filter_sql(proj_name, table_alias="p"):
+    """Return (clause, params) for a per-project WHERE filter.
+
+    Real projects -> "<alias>.project_name = %s" with (proj_name,).
+    Synthetic "(no project)" bucket -> "<alias>.project_name IS NULL" with ().
+    Centralizes the NULL-vs-equality choice so every per-project SQL site
+    handles the synthetic bucket the same way.
+    """
+    prefix = (table_alias + ".") if table_alias else ""
+    if proj_name == SYNTHETIC_NO_PROJECT_NAME:
+        return (prefix + "project_name IS NULL", ())
+    return (prefix + "project_name = %s", (proj_name,))
+
+
+def _bridge_per_project_posthog_keys_from_keychain(config, env):
+    import subprocess
+    seen = set()
+    for proj in config.get("projects", []) or []:
+        name_env = ((proj.get("posthog") or {}).get("api_key_env") or "").strip()
+        if not name_env or name_env in seen or name_env == "POSTHOG_PERSONAL_API_KEY":
+            continue
+        seen.add(name_env)
+        if env.get(name_env):
+            continue
+        try:
+            v = subprocess.check_output(
+                ["security", "find-generic-password", "-s", name_env, "-w"],
+                stderr=subprocess.DEVNULL,
+            ).decode().strip()
+        except subprocess.CalledProcessError:
+            continue
+        if v:
+            env[name_env] = v
+
+
+def _scan_repo_pages(repo_path):
+    """Walk a Next.js app-router repo and return URL paths we ship as static files.
+
+    Skips dynamic segments ([slug], [...rest]), route groups ((group)), private
+    folders (_foo), and parallel-route slots (@slot) per Next.js conventions.
+    Route groups collapse to nothing; dynamic segments exclude the whole branch.
+    """
+    out = set()
+    if not repo_path:
+        return out
+    repo = os.path.expanduser(repo_path)
+    app_roots = [
+        os.path.join(repo, "src", "app"),
+        os.path.join(repo, "app"),
+    ]
+    for root in app_roots:
+        if not os.path.isdir(root):
+            continue
+        for dirpath, dirnames, filenames in os.walk(root):
+            rel = os.path.relpath(dirpath, root)
+            segs = [] if rel == "." else rel.split(os.sep)
+            if any(s.startswith(("[", "_", "@")) for s in segs):
+                dirnames[:] = []
+                continue
+            dirnames[:] = [d for d in dirnames if not d.startswith(("[", "_", "@", "."))
+                           and d not in ("node_modules",)]
+            has_page = any(f in _PAGE_FILENAMES for f in filenames)
+            if has_page:
+                url_segs = [s for s in segs if not (s.startswith("(") and s.endswith(")"))]
+                path = "/" + "/".join(url_segs) if url_segs else "/"
+                out.add(path)
+    return out
+
+
+def _db_created_pages(conn, product_name, days=None):
+    """Return {domain: set(paths)} for pages this project published via the SEO
+    pipelines (seo_keywords) or GSC-driven page generation (gsc_queries).
+
+    When `days` is set, restrict to pages whose `completed_at` falls inside the
+    window. The seo_keywords / gsc_queries rows get `completed_at` stamped when
+    the page is actually generated, so this matches "pages created in the last
+    N days" as used by the dashboard's period selector.
+    """
+    out = {}
+    window_sql = ""
+    if days is not None:
+        window_sql = f" AND completed_at >= NOW() - INTERVAL '{int(days)} days'"
+    for sql in (
+        "SELECT page_url FROM seo_keywords WHERE product = %s AND page_url IS NOT NULL" + window_sql,
+        "SELECT page_url FROM gsc_queries WHERE product = %s AND page_url IS NOT NULL" + window_sql,
+    ):
+        try:
+            cur = conn.execute(sql, (product_name,))
+            for row in cur.fetchall():
+                url = row[0]
+                if not url:
+                    continue
+                try:
+                    parsed = urllib.parse.urlparse(url)
+                except Exception:
+                    continue
+                host = (parsed.netloc or "").lower()
+                path = parsed.path or "/"
+                while len(path) > 1 and path.endswith("/"):
+                    path = path[:-1]
+                if not host:
+                    continue
+                out.setdefault(host, set()).add(path)
+        except Exception as e:
+            print(f"  _db_created_pages query error: {e}", file=sys.stderr)
+    return out
+
+
+def _created_paths_for_project(conn, proj, days=None):
+    """Return {domain: set(paths)} of pages we created for this project.
+
+    Source-of-truth union: filesystem scan of the project's landing-pages repo
+    (applies to every domain the project owns) plus any URLs logged in
+    seo_keywords / gsc_queries (keyed by their own host).
+
+    When `days` is set, the filesystem scan is skipped entirely — static page
+    files on disk carry no creation timestamp we can trust, so a window-scoped
+    "pages created in the last N days" answer has to come from the DB alone.
+    """
+    by_domain = {}
+    domains = ps.get_project_domains(proj) or []
+    if days is None:
+        lp = proj.get("landing_pages") or {}
+        repo_path = lp.get("repo") if isinstance(lp, dict) else None
+        fs_paths = _scan_repo_pages(repo_path) if repo_path else set()
+        for d in domains:
+            by_domain.setdefault(d.lower(), set()).update(fs_paths)
+    for host, paths in _db_created_pages(conn, proj.get("name") or "", days=days).items():
+        by_domain.setdefault(host, set()).update(paths)
+    return by_domain
+
+
+def _norm_path(p):
+    """Match the frontend `normPath` in bin/server.js so PostHog pathnames
+    (`properties.$pathname`) and DB-derived created paths compare cleanly.
+    """
+    s = str(p or "/")
+    if not s.startswith("/"):
+        s = "/" + s
+    while len(s) > 1 and s.endswith("/"):
+        s = s[:-1]
+    return s
+
+
+# HogQL-based PostHog query layer.
+#
+# project_stats.py uses the events LIST endpoint with limit=1000 and no
+# pagination, so any (domain, event) that exceeds 1000 occurrences in the
+# window silently caps at 1000 and misreports the funnel. We swap that out
+# for HogQL aggregate queries (COUNT/GROUP BY), which return the true
+# totals in a single call per query.
+_SAFE_DOMAIN_RE = re.compile(r"^[A-Za-z0-9._-]+$")
+
+
+class HogqlError(Exception):
+    """Raised when a HogQL query fails after all retries.
+
+    Caller is expected to surface this as an error on the affected rows
+    instead of silently rendering zeros.
+    """
+
+
+_RETRY_BACKOFF_S = (2.0, 5.0, 12.0)
+_RETRY_AFTER_CAP_S = 30.0
+
+
+def _hogql(api_key, project_id, query, timeout=120, max_attempts=4):
+    """Run a HogQL query against /api/projects/{pid}/query/.
+
+    Retries on 429 (throttled), 5xx, and socket read timeouts. Honors
+    `Retry-After` up to `_RETRY_AFTER_CAP_S`; otherwise uses
+    `_RETRY_BACKOFF_S`. Raises `HogqlError` on permanent failure so
+    callers can mark rows as errored rather than zero.
+
+    NOTE: the batched-by-$host queries cover many domains in one scan, so a
+    single query for a large shared PostHog bucket (e.g. pid 330744 with
+    ~18 projects) can run >60s on a cold cache. A socket read timeout
+    surfaces as `socket.timeout`/`TimeoutError`, which is a sibling of
+    `urllib.error.URLError` (both subclass OSError), so it must be caught
+    explicitly; otherwise it escapes this retry loop and the caller marks
+    the entire bucket as errored on the very first slow query, rendering
+    'err' for every project sharing that bucket.
+    """
+    url = f"https://us.posthog.com/api/projects/{project_id}/query/"
+    body = json.dumps({"query": {"kind": "HogQLQuery", "query": query}}).encode("utf-8")
+    last_err = None
+    for attempt in range(max_attempts):
+        req = urllib.request.Request(url, data=body, method="POST", headers={
+            "Authorization": f"Bearer {api_key}",
+            "Content-Type": "application/json",
+        })
+        try:
+            with urllib.request.urlopen(req, timeout=timeout) as resp:
+                data = json.loads(resp.read())
+                return data.get("results", []) or []
+        except urllib.error.HTTPError as e:
+            try:
+                detail = e.read().decode("utf-8", errors="replace")[:300]
+            except Exception:
+                detail = ""
+            last_err = f"HTTP {e.code}: {detail}"
+            retryable = (e.code == 429) or (500 <= e.code < 600)
+            if not retryable or attempt == max_attempts - 1:
+                print(f"  HogQL HTTPError {e.code}: {detail} | query={query[:120]}", file=sys.stderr)
+                break
+            wait = _RETRY_BACKOFF_S[min(attempt, len(_RETRY_BACKOFF_S) - 1)]
+            try:
+                ra = e.headers.get("Retry-After") if e.headers else None
+                if ra is not None:
+                    wait = min(_RETRY_AFTER_CAP_S, max(wait, float(ra)))
+            except Exception:
+                pass
+            print(f"  HogQL {e.code} retry {attempt + 1}/{max_attempts - 1} in {wait:.1f}s | query={query[:80]}", file=sys.stderr)
+            time.sleep(wait)
+            continue
+        except (socket.timeout, TimeoutError) as e:
+            # Read timeout on a heavy batched query. Retryable: a retry
+            # often hits a warm cache and returns in time. Caught before
+            # URLError because TimeoutError is NOT a URLError subclass.
+            last_err = f"read timeout after {timeout}s: {e}"
+            if attempt == max_attempts - 1:
+                print(f"  HogQL timeout (>{timeout}s): {e} | query={query[:120]}", file=sys.stderr)
+                break
+            wait = _RETRY_BACKOFF_S[min(attempt, len(_RETRY_BACKOFF_S) - 1)]
+            print(f"  HogQL timeout retry {attempt + 1}/{max_attempts - 1} in {wait:.1f}s | query={query[:80]}", file=sys.stderr)
+            time.sleep(wait)
+            continue
+        except urllib.error.URLError as e:
+            # A URLError can also wrap a socket.timeout (e.reason). Treat
+            # those as the retryable timeout case above.
+            last_err = f"URLError: {e}"
+            if attempt == max_attempts - 1:
+                print(f"  HogQL URLError: {e} | query={query[:120]}", file=sys.stderr)
+                break
+            wait = _RETRY_BACKOFF_S[min(attempt, len(_RETRY_BACKOFF_S) - 1)]
+            print(f"  HogQL URLError retry {attempt + 1}/{max_attempts - 1} in {wait:.1f}s: {e}", file=sys.stderr)
+            time.sleep(wait)
+            continue
+    raise HogqlError(last_err or "unknown HogQL failure")
+
+
+def _empty_domain_stats(domain, error=None):
+    """Zero'd per-domain stats. If `error` is set, treat the zeros as
+    UNKNOWN (not truly 0) so the dashboard can render an error cell
+    instead of silently misreporting."""
+    out = {
+        "pageviews": 0,
+        "cta_clicks": 0,
+        "email_signups": 0,
+        "schedule_clicks": 0,
+        "get_started_clicks": 0,
+        "cross_product_clicks": 0,
+        "pageview_details": {domain: {
+            "total": 0,
+            "top_pages": {},
+            "top_pages_signups": {},
+            "top_pages_schedule": {},
+            "top_pages_get_started": {},
+        }},
+        "cta_details": [],
+    }
+    if error:
+        out["error"] = error
+    return out
+
+
+# Legacy + canonical event names for the "get started" click.  Fazm fires
+# `download_click`, Assrt fires `cta_get_started_clicked`, new sites fire
+# `get_started_click`.  Collapsed back to a single name once both old sites
+# migrate to trackGetStartedClick.
+_GET_STARTED_EVENTS = "('get_started_click', 'download_click', 'cta_get_started_clicked')"
+
+
+def _ph_batch_counts(api_key, project_id, domains, after_iso):
+    """Fetch per-domain PostHog aggregates for every `domain` in one batched
+    pass against a single (api_key, project_id) bucket.
+
+    The previous implementation fired ~10 HogQL queries per domain, which
+    fanned out to 100+ concurrent requests and tripped PostHog's rate
+    limiter; throttled calls silently returned 0, misreporting every
+    project except the one with its own dedicated API key.
+
+    This version groups each aggregate by `properties.$host`, so one query
+    covers every domain in the bucket. Returns `{domain: stats_dict}` in
+    the same shape the old per-domain function produced. On permanent
+    HogQL failure, raises `HogqlError` so the caller can mark rows as
+    errored rather than rendering a misleading zero.
+    """
+    result = {d: _empty_domain_stats(d) for d in domains}
+    safe_domains = []
+    for d in domains:
+        if _SAFE_DOMAIN_RE.match(d or ""):
+            safe_domains.append(d)
+        else:
+            print(f"  skip unsafe domain: {d!r}", file=sys.stderr)
+            result[d]["error"] = "unsafe domain"
+    if not safe_domains:
+        return result
+
+    after_str = (after_iso or "").replace("T", " ")
+    if not after_str:
+        return result
+
+    in_list = ", ".join(f"'{d}'" for d in safe_domains)
+
+    def _count_by_host(event_clause, distinct_key=None):
+        # Pass `distinct_key` (e.g. "properties.email") to dedupe across
+        # double-fired events for the same conversion. Used for email
+        # signups where both `newsletter_subscribed` (client) and
+        # `newsletter_subscribed_server` (server) fire for one submission.
+        count_expr = (
+            f"count(DISTINCT {distinct_key}) AS c"
+            if distinct_key
+            else "count() AS c"
+        )
+        q = (
+            f"SELECT properties.$host AS host, {count_expr} FROM events "
+            f"WHERE {event_clause} "
+            f"AND properties.$host IN ({in_list}) "
+            f"AND timestamp >= toDateTime('{after_str}') "
+            "GROUP BY host"
+        )
+        rows = _hogql(api_key, project_id, q)
+        return {r[0]: int(r[1]) for r in (rows or []) if r and r[0]}
+
+    def _top_pages_by_host(event_clause, row_cap=5000, distinct_key="distinct_id"):
+        # All per-page breakdowns count unique users (distinct_id) rather than
+        # raw events. A visitor that views the same /pricing twice or rage-
+        # clicks the same CTA still counts as 1. Pass `distinct_key=None` to
+        # opt back into raw count() for legacy callers.
+        count_expr = (
+            f"count(DISTINCT {distinct_key}) AS c"
+            if distinct_key
+            else "count() AS c"
+        )
+        q = (
+            f"SELECT properties.$host AS host, properties.$pathname AS path, {count_expr} FROM events "
+            f"WHERE {event_clause} "
+            f"AND properties.$host IN ({in_list}) "
+            f"AND timestamp >= toDateTime('{after_str}') "
+            f"GROUP BY host, path ORDER BY c DESC LIMIT {int(row_cap)}"
+        )
+        rows = _hogql(api_key, project_id, q)
+        out = {d: {} for d in safe_domains}
+        for r in (rows or []):
+            host = r[0] if len(r) > 0 else None
+            path = r[1] if len(r) > 1 and r[1] else "/"
+            cnt = int(r[2]) if len(r) > 2 else 0
+            if host in out:
+                out[host][path] = cnt
+        return out
+
+    # Email signups: client `newsletter_subscribed` is ad-blocker-lossy
+    # (~57% capture). Server-side `newsletter_subscribed_server` (added in
+    # @m13v/seo-components v0.38) is ground truth. Count both with DISTINCT
+    # email so a client + server pair for the same submission collapses to one.
+    _SIGNUP_CLAUSE = (
+        "event IN ('newsletter_subscribed', 'newsletter_subscribed_server')"
+    )
+
+    # Visitors, not raw pageviews. Globally consistent with every other
+    # column in this batch (cta_clicks, schedule_clicks, get_started_clicks,
+    # cross_product_clicks, email_signups all count unique users). A visitor
+    # bouncing between /pricing and /docs still counts as 1.
+    #
+    # PERF/RATE-LIMIT: PostHog throttles the (shared) personal API key
+    # org-wide on the /query endpoint (429 "throttled"). The five
+    # unique-visitor counts that all dedupe on distinct_id are folded into
+    # ONE grouped query (bucketed by event via multiIf) instead of five
+    # separate HogQL calls. Email signups stay on their own query because
+    # they dedupe on coalesce(email, distinct_id), a different key. This
+    # halves the per-bucket request count, which is what was blowing the
+    # rate limit when all windows/buckets fired at once.
+    def _multi_count_by_host():
+        bucket_expr = (
+            "multiIf("
+            "event = '$pageview', 'pv', "
+            "event = 'cta_click', 'cta', "
+            "event = 'schedule_click', 'sched', "
+            f"event IN {_GET_STARTED_EVENTS}, 'gs', "
+            "'cross')"
+        )
+        q = (
+            f"SELECT properties.$host AS host, {bucket_expr} AS bkt, "
+            "count(DISTINCT distinct_id) AS c FROM events "
+            "WHERE event IN ('$pageview', 'cta_click', 'schedule_click', "
+            "'cross_product_click', 'get_started_click', 'download_click', "
+            "'cta_get_started_clicked') "
+            f"AND properties.$host IN ({in_list}) "
+            f"AND timestamp >= toDateTime('{after_str}') "
+            "GROUP BY host, bkt"
+        )
+        rows = _hogql(api_key, project_id, q)
+        out = {"pv": {}, "cta": {}, "sched": {}, "gs": {}, "cross": {}}
+        for r in (rows or []):
+            host = r[0] if len(r) > 0 else None
+            bkt = r[1] if len(r) > 1 else None
+            cnt = int(r[2]) if len(r) > 2 else 0
+            if host and bkt in out:
+                out[bkt][host] = cnt
+        return out
+
+    _counts = _multi_count_by_host()
+    pv_total = _counts["pv"]
+    cta_total = _counts["cta"]
+    sched_total = _counts["sched"]
+    # Get Started = unique users who took the conversion action, not raw clicks.
+    get_started_total = _counts["gs"]
+    cross_product_total = _counts["cross"]
+    # Email signups: client `newsletter_subscribed` is ad-blocker-lossy
+    # (~57% capture). Server-side `newsletter_subscribed_server` is ground
+    # truth. Count both with DISTINCT email so a client + server pair for the
+    # same submission collapses to one. Kept as its own query (distinct key
+    # differs from the distinct_id batch above).
+    signup_total = _count_by_host(
+        _SIGNUP_CLAUSE,
+        distinct_key="coalesce(properties.email, distinct_id)",
+    )
+
+    # Per-page breakdowns. The big $pageview scan keeps its own large cap;
+    # the three low-volume conversion breakdowns (signup/sched/get_started,
+    # all distinct_id) fold into one grouped-by-event query for the same
+    # rate-limit reason as the counts above.
+    top_pv = _top_pages_by_host("event = '$pageview'", row_cap=5000)
+
+    def _multi_top_pages_small():
+        bucket_expr = (
+            "multiIf("
+            "event = 'schedule_click', 'sched', "
+            f"event IN {_GET_STARTED_EVENTS}, 'gs', "
+            "'signup')"
+        )
+        q = (
+            f"SELECT properties.$host AS host, properties.$pathname AS path, "
+            f"{bucket_expr} AS bkt, count(DISTINCT distinct_id) AS c FROM events "
+            "WHERE event IN ('schedule_click', 'newsletter_subscribed', "
+            "'newsletter_subscribed_server', 'get_started_click', "
+            "'download_click', 'cta_get_started_clicked') "
+            f"AND properties.$host IN ({in_list}) "
+            f"AND timestamp >= toDateTime('{after_str}') "
+            "GROUP BY host, path, bkt ORDER BY c DESC LIMIT 1500"
+        )
+        rows = _hogql(api_key, project_id, q)
+        out = {
+            "signup": {d: {} for d in safe_domains},
+            "sched": {d: {} for d in safe_domains},
+            "gs": {d: {} for d in safe_domains},
+        }
+        for r in (rows or []):
+            host = r[0] if len(r) > 0 else None
+            path = r[1] if len(r) > 1 and r[1] else "/"
+            bkt = r[2] if len(r) > 2 else None
+            cnt = int(r[3]) if len(r) > 3 else 0
+            if bkt in out and host in out[bkt]:
+                out[bkt][host][path] = cnt
+        return out
+
+    _tp = _multi_top_pages_small()
+    top_signup = _tp["signup"]
+    top_sched = _tp["sched"]
+    top_get_started = _tp["gs"]
+
+    cta_details_by_host = {d: [] for d in safe_domains}
+    if any(v > 0 for v in cta_total.values()):
+        cta_detail_q = (
+            "SELECT properties.$host AS host, properties.$el_text, properties.text, properties.section, timestamp "
+            "FROM events "
+            "WHERE event = 'cta_click' "
+            f"AND properties.$host IN ({in_list}) "
+            f"AND timestamp >= toDateTime('{after_str}') "
+            "ORDER BY timestamp DESC LIMIT 200"
+        )
+        rows = _hogql(api_key, project_id, cta_detail_q)
+        for r in (rows or []):
+            host = r[0] if len(r) > 0 else None
+            el_text = r[1] if len(r) > 1 else None
+            text = r[2] if len(r) > 2 else None
+            section = r[3] if len(r) > 3 else None
+            ts = r[4] if len(r) > 4 else None
+            bucket = cta_details_by_host.get(host)
+            if bucket is None or len(bucket) >= 10:
+                continue
+            bucket.append({
+                "text": el_text or text or "?",
+                "section": section or "?",
+                "time": (str(ts)[:16] if ts else "?"),
+            })
+
+    # Autocapture fallback: only domains with zero `cta_click` get the
+    # "$autocapture clicks whose text contains 'book'" treatment. Batched
+    # like everything else so we don't fan out.
+    fallback_hosts = [d for d in safe_domains if cta_total.get(d, 0) == 0]
+    if fallback_hosts:
+        fb_in = ", ".join(f"'{d}'" for d in fallback_hosts)
+        ac_total_q = (
+            "SELECT properties.$host AS host, count(DISTINCT distinct_id) AS c FROM events "
+            "WHERE event = '$autocapture' "
+            f"AND properties.$host IN ({fb_in}) "
+            f"AND timestamp >= toDateTime('{after_str}') "
+            "AND lower(properties.$el_text) LIKE '%book%' "
+            "GROUP BY host"
+        )
+        ac_rows = _hogql(api_key, project_id, ac_total_q)
+        ac_total = {r[0]: int(r[1]) for r in (ac_rows or []) if r and r[0]}
+        hosts_with_ac = [d for d in fallback_hosts if ac_total.get(d, 0) > 0]
+        if hosts_with_ac:
+            ac_in = ", ".join(f"'{d}'" for d in hosts_with_ac)
+            ac_detail_q = (
+                "SELECT properties.$host AS host, properties.$el_text, properties.text, properties.section, timestamp "
+                "FROM events "
+                "WHERE event = '$autocapture' "
+                f"AND properties.$host IN ({ac_in}) "
+                f"AND timestamp >= toDateTime('{after_str}') "
+                "AND lower(properties.$el_text) LIKE '%book%' "
+                "ORDER BY timestamp DESC LIMIT 200"
+            )
+            rows = _hogql(api_key, project_id, ac_detail_q)
+            for r in (rows or []):
+                host = r[0] if len(r) > 0 else None
+                el_text = r[1] if len(r) > 1 else None
+                text = r[2] if len(r) > 2 else None
+                section = r[3] if len(r) > 3 else None
+                ts = r[4] if len(r) > 4 else None
+                bucket = cta_details_by_host.get(host)
+                if bucket is None or len(bucket) >= 10:
+                    continue
+                bucket.append({
+                    "text": el_text or text or "?",
+                    "section": section or "?",
+                    "time": (str(ts)[:16] if ts else "?"),
+                })
+        # Roll autocapture counts into cta_total so the funnel "cta_clicks"
+        # column matches the detail list for fallback domains.
+        for h, c in ac_total.items():
+            cta_total[h] = max(cta_total.get(h, 0), c)
+
+    for d in safe_domains:
+        pv = pv_total.get(d, 0)
+        result[d] = {
+            "pageviews": pv,
+            "cta_clicks": cta_total.get(d, 0),
+            "email_signups": signup_total.get(d, 0),
+            "schedule_clicks": sched_total.get(d, 0),
+            "get_started_clicks": get_started_total.get(d, 0),
+            "cross_product_clicks": cross_product_total.get(d, 0),
+            "pageview_details": {d: {
+                "total": pv,
+                "top_pages": top_pv.get(d, {}),
+                "top_pages_signups": top_signup.get(d, {}),
+                "top_pages_schedule": top_sched.get(d, {}),
+                "top_pages_get_started": top_get_started.get(d, {}),
+            }},
+            "cta_details": cta_details_by_host.get(d, []),
+        }
+    return result
+
+
+def _ph_combine(per_domain):
+    out = {
+        "pageviews": 0,
+        "cta_clicks": 0,
+        "email_signups": 0,
+        "schedule_clicks": 0,
+        "get_started_clicks": 0,
+        "cross_product_clicks": 0,
+        "pageview_details": {},
+        "cta_details": [],
+    }
+    for s in per_domain:
+        out["pageviews"] += s.get("pageviews", 0)
+        out["cta_clicks"] += s.get("cta_clicks", 0)
+        out["email_signups"] += s.get("email_signups", 0)
+        out["schedule_clicks"] += s.get("schedule_clicks", 0)
+        out["get_started_clicks"] += s.get("get_started_clicks", 0)
+        out["cross_product_clicks"] += s.get("cross_product_clicks", 0)
+        out["pageview_details"].update(s.get("pageview_details", {}))
+        out["cta_details"].extend(s.get("cta_details", []))
+    return out
+
+
+def _bookings_shared(bookings_conn, client_slug, days, table="cal_bookings", require_utm=False):
+    """Same output shape as ps.get_booking_stats, but reuses a shared psycopg2
+    connection instead of opening a fresh one per project.
+    `table` is `cal_bookings` (Cal.com) or `calendly_bookings` (Calendly).
+    `require_utm` gates `real_bookings` on `utm_source IS NOT NULL` for
+    projects whose booking destination is shared with non-marketing inbound
+    (set in config.json via `bookings_require_utm`)."""
+    if not bookings_conn or not client_slug:
+        return None
+    try:
+        if table not in {"cal_bookings", "calendly_bookings"}:
+            raise ValueError(f"unsupported booking table: {table}")
+        utm_clause = " AND utm_source IS NOT NULL" if require_utm else ""
+        cur = bookings_conn.cursor()
+        cur.execute(
+            "SELECT COUNT(*), "
+            "COUNT(*) FILTER (WHERE status = 'created'), "
+            "COUNT(*) FILTER (WHERE status = 'cancelled'), "
+            "COUNT(*) FILTER (WHERE status = 'rescheduled'), "
+            "COUNT(*) FILTER (WHERE attendee_email NOT ILIKE '%%test%%' "
+            "AND attendee_email NOT ILIKE '%%example%%' "
+            "AND attendee_email NOT ILIKE '%%+%%verify%%' "
+            "AND attendee_name NOT ILIKE '%%test%%' "
+            "AND attendee_name NOT ILIKE '%%verification%%' "
+            "AND attendee_name NOT ILIKE '%%delete-me%%' "
+            "AND attendee_name NOT ILIKE '%%john doe%%'"
+            + utm_clause + ") "
+            "FROM " + table + " WHERE client_slug = %s "
+            "AND created_at >= NOW() - INTERVAL '" + str(days) + " days'",
+            (client_slug,),
+        )
+        row = cur.fetchone()
+        cols = ["total", "booked", "cancelled", "rescheduled", "real_bookings"]
+        result = dict(zip(cols, row)) if row else {}
+
+        cur.execute(
+            "SELECT attendee_name, attendee_email, status, start_time, created_at "
+            "FROM " + table + " WHERE client_slug = %s "
+            "AND created_at >= NOW() - INTERVAL '" + str(days) + " days' "
+            "ORDER BY created_at DESC LIMIT 5",
+            (client_slug,),
+        )
+        result["recent"] = [
+            {"name": r[0], "email": r[1], "status": r[2],
+             "start": str(r[3])[:16] if r[3] else "?",
+             "created": str(r[4])[:16] if r[4] else "?"}
+            for r in cur.fetchall()
+        ]
+        cur.close()
+        return result
+    except Exception as e:
+        print(f"  Bookings DB error for {client_slug}: {e}", file=sys.stderr)
+        return None
+
+
+def _dm_short_link_stats(conn, name, days):
+    """Per-project DM short-link click attribution.
+
+    `dm_clicks`: SUM(dm_links.clicks) JOIN dms d for DMs that reference this
+    project (target_project OR membership in target_projects[]) and were last
+    touched in the window. Captures every DM click — booking, github, website,
+    or kind=other — bumped at the resolver. Multi-link, multi-turn safe.
+    """
+    if not name or name == SYNTHETIC_NO_PROJECT_NAME:
+        return 0
+    try:
+        cur = conn.execute(
+            "SELECT COALESCE(SUM(l.clicks), 0)::int "
+            "FROM dm_links l "
+            "JOIN dms d ON d.id = l.dm_id "
+            "WHERE (COALESCE(d.target_project, d.project_name) = %s "
+            "       OR %s = ANY(d.target_projects)) "
+            "AND COALESCE(d.last_message_at, d.discovered_at) >= NOW() - INTERVAL '" + str(int(days)) + " days'",
+            (name, name),
+        )
+        return int((cur.fetchone() or (0,))[0])
+    except Exception as e:
+        print(f"  dm_short_link_stats error for {name}: {e}", file=sys.stderr)
+        return 0
+
+
+def _dm_booking_count(conn, bookings_conn, name, days):
+    """Count cal_bookings within the window whose metadata.utm_content
+    (`dm_<id>`) maps to a DM targeting this project.
+
+    The webhook stores the entire Cal.com payload under cal_bookings.metadata,
+    and the original UTM lives at metadata.payload.metadata.utm_content. We
+    parse the dm_id out of `dm_<n>`, then join against dms.target_project /
+    project_name in the main DB to scope by project.
+    """
+    if not bookings_conn or not name or name == SYNTHETIC_NO_PROJECT_NAME:
+        return 0
+    try:
+        cur = bookings_conn.cursor()
+        cur.execute(
+            "SELECT metadata#>>'{payload,metadata,utm_content}' AS utm_content "
+            "FROM cal_bookings "
+            "WHERE metadata#>>'{payload,metadata,utm_content}' LIKE 'dm_%%' "
+            "AND created_at >= NOW() - INTERVAL '" + str(int(days)) + " days' "
+            "AND COALESCE(attendee_email, '') NOT ILIKE '%%test%%'"
+        )
+        dm_ids = []
+        for (utm,) in cur.fetchall():
+            if utm and utm.startswith('dm_'):
+                try:
+                    dm_ids.append(int(utm.split('_', 1)[1]))
+                except (ValueError, IndexError):
+                    pass
+        cur.close()
+        if not dm_ids:
+            return 0
+        cur2 = conn.execute(
+            "SELECT COUNT(*)::int FROM dms WHERE id = ANY(%s) "
+            "AND COALESCE(target_project, project_name) = %s",
+            (dm_ids, name),
+        )
+        return int((cur2.fetchone() or (0,))[0])
+    except Exception as e:
+        print(f"  dm_booking_count error for {name}: {e}", file=sys.stderr)
+        return 0
+
+
+def _period_total_engagement(conn, name, days, platform=None):
+    """Total engagement *gained during the window* across ALL posts, regardless
+    of when each post was created.
+
+    Used to populate the "(total)" bracketed value on the project panel.
+    Logic per post:
+      gain = latest_snapshot_in_window - latest_snapshot_before_window
+    with the "before" leg treated as 0 when the post did not exist before
+    the window (new posts contribute their full current value, which is
+    why this differs from the Trends-tab LAG() approach: that one excludes
+    every post's first snapshot and therefore undercounts fresh activity).
+
+    Same platform filter as the Trends tab: excludes moltbook / github /
+    github_issues. Same project filter via posts.project_name.
+
+    For post_clicks: COUNT of post_link_clicks rows with is_bot=FALSE in the
+    window, joined post_links -> posts so we can apply the project filter.
+    Pre-2026-05-07 click rows do not exist (is_bot logging started then), so
+    the count returns 0 for older days rather than mixing inflated counters.
+    """
+    # Period total = engagement gained during the last N days, summed from
+    # two complementary branches that always together produce a value
+    # >= the panel's scoped column:
+    #
+    #   (1) new_posts_branch — posts CREATED in the window. Their full
+    #       live posts.* counters are credited as in-window gain (all of
+    #       it was earned during the window since the post didn't exist
+    #       before). No reddit/moltbook -1 OP self-vote discount here
+    #       (the scoped column applies that discount, so the un-discounted
+    #       sum here is guaranteed >= scoped).
+    #
+    #   (2) old_posts_branch — posts created BEFORE the window. Uses the
+    #       Trends-tab LAG approach over post_views_daily, summing daily
+    #       gains across snapshots inside the window. NULL values
+    #       (Reddit posts don't write upvotes/comments to post_views_daily
+    #       at all) are excluded by the IS NOT NULL FILTER, so old
+    #       Reddit posts contribute 0 here — that's a known limitation
+    #       of the snapshot pipeline and matches the Trends chart.
+    # Per-metric platform filter matches the SCOPED column's filter so the
+    # bracket is always >= scoped for the same metric:
+    #   upvotes:  no platform filter (scoped sums all platforms with reddit/
+    #             moltbook -1 OP self-vote discount; bracket uses raw values).
+    #   comments: no platform filter (scoped sums all platforms).
+    #   views:    excludes moltbook/github/github_issues (matches scoped's
+    #             FILTER clause in _windowed_post_engagement).
+    days_sql = "INTERVAL '" + str(int(days)) + " days'"
+    views_excl = "LOWER(p.platform) NOT IN ('moltbook', 'github', 'github_issues')"
+    # When platform is set, also apply the mention-row exclusion so this
+    # function lines up with the /api/style/stats view that the dashboard
+    # shows above the Project Final Stats table.
+    plat_clause = _platform_sql_clause(platform, "p")
+    proj_clause, proj_params = _project_filter_sql(name, "p")
+    cur = conn.execute(
+        # Branch 1: posts CREATED in the window. Full live posts.* values
+        # are credited as in-window gain. No -1 OP discount on upvotes, so
+        # bracket >= scoped on reddit/moltbook by exactly #posts_in_window.
+        "WITH new_posts AS ("
+          "SELECT "
+            "COALESCE(SUM(p.upvotes),        0)::bigint AS upvotes, "
+            "COALESCE(SUM(p.comments_count), 0)::bigint AS comments, "
+            "COALESCE(SUM(p.views) FILTER (WHERE " + views_excl + "), 0)::bigint AS views "
+          "FROM posts p "
+          "WHERE " + proj_clause + " "
+            "AND p.posted_at >= NOW() - " + days_sql + plat_clause +
+        "), "
+        # Branch 2: posts created BEFORE the window. LAG over snapshots
+        # inside the window. Reddit/moltbook post_views_daily rows carry
+        # NULL upvotes/comments by design (those stats pipelines only
+        # write views), so the IS NOT NULL FILTER drops them: old Reddit
+        # upvotes/comments gain is structurally invisible here, matching
+        # the Trends chart.
+        "old_post_daily AS ("
+          "SELECT pvd.post_id, p.platform, "
+            "pvd.upvotes,  LAG(pvd.upvotes)  OVER w AS prev_upvotes, "
+            "pvd.comments, LAG(pvd.comments) OVER w AS prev_comments, "
+            "pvd.views,    LAG(pvd.views)    OVER w AS prev_views "
+          "FROM post_views_daily pvd "
+          "JOIN posts p ON p.id = pvd.post_id "
+          "WHERE pvd.day >= CURRENT_DATE - " + days_sql + " "
+            "AND " + proj_clause + " "
+            "AND p.posted_at < NOW() - " + days_sql + plat_clause + " "
+          "WINDOW w AS (PARTITION BY pvd.post_id ORDER BY pvd.day)"
+        "), "
+        "old_posts AS ("
+          "SELECT "
+            "COALESCE(SUM(GREATEST(upvotes  - prev_upvotes,  0)) "
+              "FILTER (WHERE prev_upvotes  IS NOT NULL AND upvotes  IS NOT NULL), 0)::bigint AS upvotes, "
+            "COALESCE(SUM(GREATEST(comments - prev_comments, 0)) "
+              "FILTER (WHERE prev_comments IS NOT NULL AND comments IS NOT NULL), 0)::bigint AS comments, "
+            "COALESCE(SUM(GREATEST(views    - prev_views,    0)) "
+              "FILTER (WHERE prev_views    IS NOT NULL AND views    IS NOT NULL "
+                "AND LOWER(platform) NOT IN ('moltbook', 'github', 'github_issues')), 0)::bigint AS views "
+          "FROM old_post_daily"
+        ") "
+        "SELECT "
+          "n.upvotes  + o.upvotes, "
+          "n.comments + o.comments, "
+          "n.views    + o.views "
+        "FROM new_posts n CROSS JOIN old_posts o",
+        proj_params + proj_params,
+    )
+    row = cur.fetchone() or (0, 0, 0)
+    upvotes_total = int(row[0] or 0)
+    comments_total = int(row[1] or 0)
+    views_total = int(row[2] or 0)
+
+    # post_clicks bracket = scoped (post_links.clicks SUM for new posts in
+    # window) + COUNT of post_link_clicks events on OLD posts during the
+    # window. The "new posts" leg matches the scoped column exactly so
+    # bracket >= scoped is guaranteed; the "old posts" leg captures
+    # click traffic that hit pre-existing posts during the period.
+    cur2 = conn.execute(
+        "WITH new_clicks AS ("
+          "SELECT COALESCE(SUM(pl.total_clicks), 0)::bigint AS clicks "
+          "FROM posts p "
+          "LEFT JOIN ("
+            "SELECT post_id, SUM(clicks)::int AS total_clicks "
+            "FROM post_links WHERE post_id IS NOT NULL GROUP BY post_id"
+          ") pl ON pl.post_id = p.id "
+          "WHERE " + proj_clause + " "
+            "AND p.posted_at >= NOW() - " + days_sql + plat_clause +
+        "), "
+        "old_event_clicks AS ("
+          "SELECT COALESCE(COUNT(*), 0)::bigint AS clicks "
+          "FROM post_link_clicks plc "
+          "JOIN post_links pl ON pl.code = plc.code "
+          "JOIN posts p ON p.id = pl.post_id "
+          "WHERE plc.ts >= NOW() - " + days_sql + " "
+            "AND plc.is_bot = FALSE "
+            "AND " + proj_clause + " "
+            "AND p.posted_at < NOW() - " + days_sql + plat_clause +
+        ") "
+        "SELECT n.clicks + o.clicks "
+        "FROM new_clicks n CROSS JOIN old_event_clicks o",
+        proj_params + proj_params,
+    )
+    row2 = cur2.fetchone() or (0,)
+    post_clicks_total = int(row2[0] or 0)
+
+    return {
+        "upvotes": upvotes_total,
+        "comments": comments_total,
+        "views": views_total,
+        "post_clicks": post_clicks_total,
+    }
+
+
+def _windowed_post_engagement(conn, name, days, platform=None):
+    """Sum engagement only for posts *created within the window*.
+
+    project_stats.get_post_stats aggregates engagement over ALL time for the
+    project, which is misleading when the window is a day or a week. Here we
+    filter by posted_at so upvotes/comments/views/post_clicks match the same
+    24h slice as the 'recent' post count.
+
+    When `platform` is set, also folds in the same platform/mention filter
+    that /api/style/stats uses so the Project Final Stats and Posts by
+    Engagement Style tables agree on the same denominator.
+
+    post_clicks: SUM of post_links.clicks attributable to short links minted
+    for posts in this project's window (post_id-keyed; reply-keyed clicks
+    excluded so we don't double-count engagement on replies hanging off
+    someone else's thread).
+    """
+    # upvotes is NET of the Reddit/Moltbook OP self-upvote (both platforms auto-
+    # apply a +1 to every post). Discounting per row before the SUM means the
+    # funnel reflects organic engagement, not (posts * 1) + organic. X /
+    # LinkedIn / GitHub have no equivalent auto-vote so they pass through.
+    # Matches top_performers.SCORE_SQL and bin/server.js upvotes_discounted.
+    plat_clause = _platform_sql_clause(platform, "p")
+    proj_clause, proj_params = _project_filter_sql(name, "p")
+    cur = conn.execute(
+        "SELECT COALESCE(SUM(CASE WHEN LOWER(p.platform) IN ('reddit', 'moltbook') "
+        "  THEN GREATEST(0, COALESCE(p.upvotes, 0) - 1) "
+        "  ELSE COALESCE(p.upvotes, 0) END), 0), "
+        "COALESCE(SUM(p.comments_count), 0), "
+        "COALESCE(SUM(p.views) FILTER (WHERE LOWER(p.platform) NOT IN ('moltbook', 'github', 'github_issues')), 0), "
+        "COUNT(*) FILTER (WHERE LOWER(p.platform) NOT IN ('moltbook', 'github', 'github_issues')), "
+        "COALESCE(SUM(pl.total_clicks), 0) "
+        "FROM posts p "
+        "LEFT JOIN ("
+        "  SELECT post_id, SUM(clicks)::int AS total_clicks "
+        "  FROM post_links WHERE post_id IS NOT NULL GROUP BY post_id"
+        ") pl ON pl.post_id = p.id "
+        "WHERE " + proj_clause + " AND p.posted_at >= NOW() - INTERVAL '" + str(days) + " days'"
+        + plat_clause,
+        proj_params,
+    )
+    row = cur.fetchone() or (0, 0, 0, 0, 0)
+    return {
+        "upvotes": int(row[0] or 0),
+        "comments": int(row[1] or 0),
+        "views": int(row[2] or 0),
+        "views_posts": int(row[3] or 0),
+        "post_clicks": int(row[4] or 0),
+    }
+
+
+def _seo_pages_count(conn, name, days):
+    """Count SEO pages published in window. seo_keywords.product matches project_name."""
+    cur = conn.execute(
+        "SELECT "
+        "(SELECT COUNT(*) FROM seo_keywords WHERE product = %s "
+        "   AND completed_at >= NOW() - INTERVAL '" + str(days) + " days' "
+        "   AND page_url IS NOT NULL) + "
+        "(SELECT COUNT(*) FROM gsc_queries WHERE product = %s "
+        "   AND completed_at >= NOW() - INTERVAL '" + str(days) + " days' "
+        "   AND page_url IS NOT NULL)",
+        (name, name),
+    )
+    row = cur.fetchone()
+    return int((row and row[0]) or 0)
+
+
+def _amplitude_signups_24h_from_cache(proj):
+    """For days==1, read the precomputed rolling-24h count from the cache
+    written by scripts/amplitude_24h_signups.py.
+
+    That script uses our own server-side PostHog `newsletter_subscribed`
+    event (real-time, partner_outcome IN ('partner_created','partner_reused'))
+    as the primary source, because Amplitude segmentation/export both lag
+    several hours behind real time and bucket by calendar day in the
+    project's display timezone.
+
+    Returns int (count) or None when:
+      - cache file missing / unreadable
+      - project not present in cache
+      - cache is older than 30 minutes (stale, fall back to live segmentation)
+    """
+    cache_path = os.path.expanduser(
+        "~/social-autoposter/skill/cache/amplitude_24h_signups.json"
+    )
+    if not os.path.exists(cache_path):
+        return None
+    try:
+        with open(cache_path) as f:
+            cur = json.load(f)
+        gen = cur.get("generated_at_utc")
+        if gen:
+            age_min = (datetime.now(timezone.utc) - datetime.fromisoformat(gen)).total_seconds() / 60
+            if age_min > 30:
+                return None
+        for p in cur.get("projects") or []:
+            if p.get("name") == proj.get("name"):
+                v = p.get("count_24h")
+                return int(v) if v is not None else None
+    except Exception:
+        return None
+    return None
+
+
+def _amplitude_signups(proj, days, env):
+    """Pull attributed end-product signup count from the client's Amplitude.
+
+    For projects with an `amplitude` config block (project_id, api_key_env,
+    secret_key_env, signup_event, attribution_filter). Returns total signups
+    matching the filter over the last `days`, or None if not configured /
+    creds missing / API errors. Errors are non-fatal — they collapse to None
+    so the dashboard falls back to the click-based metric.
+
+    Special case: days == 1 reads from the rolling-24h cache populated by
+    scripts/amplitude_24h_signups.py, which uses real-time PostHog data
+    instead of Amplitude segmentation (which lags hours and buckets by
+    calendar day in the project's display timezone). Falls through to the
+    segmentation path if the cache is missing or stale.
+    """
+    amp = proj.get("amplitude")
+    if not amp:
+        return None
+    if days == 1:
+        cached = _amplitude_signups_24h_from_cache(proj)
+        if cached is not None:
+            return cached
+    api_key = env.get(amp.get("api_key_env", ""))
+    secret_key = env.get(amp.get("secret_key_env", ""))
+    if not api_key or not secret_key:
+        return None
+    import base64
+    end_dt = datetime.now(timezone.utc)
+    start_dt = end_dt - timedelta(days=max(1, days) - 1)
+    e = json.dumps({
+        "event_type": amp.get("signup_event", "New User Sign Up"),
+        "filters": [
+            {
+                "subprop_type": "event",
+                "subprop_key": k,
+                "subprop_op": "is",
+                "subprop_value": v if isinstance(v, list) else [v],
+            }
+            for k, v in (amp.get("attribution_filter") or {}).items()
+        ],
+    })
+    qs = urllib.parse.urlencode({
+        "e": e,
+        "start": start_dt.strftime("%Y%m%d"),
+        "end": end_dt.strftime("%Y%m%d"),
+        "i": "1",
+        "m": "totals",
+    })
+    auth_b64 = base64.b64encode(f"{api_key}:{secret_key}".encode()).decode()
+    req = urllib.request.Request(
+        f"https://amplitude.com/api/2/events/segmentation?{qs}",
+        headers={"Authorization": f"Basic {auth_b64}"},
+    )
+    try:
+        with urllib.request.urlopen(req, timeout=20) as resp:
+            data = json.loads(resp.read())
+    except Exception as exc:
+        print(f"  amplitude signups fetch error ({proj.get('name')}): {exc}", file=sys.stderr)
+        return None
+    series = (data.get("data", {}).get("series") or [[]])[0]
+    return int(sum(int(x or 0) for x in series))
+
+
+def _post_stats_synthetic_null(conn, days):
+    """NULL-project sibling of ps.get_post_stats. Same shape, same upvote
+    discount logic; filters posts.project_name IS NULL instead of = name.
+
+    ps.get_post_stats lives in the chflags-locked project_stats.py, so the
+    synthetic '(no project)' bucket reuses this in build_project_entry
+    rather than passing a magic string into a function that would return
+    all-zeros for it.
+    """
+    cur = conn.execute(
+        "SELECT COUNT(*), "
+        "COUNT(*) FILTER (WHERE posted_at >= NOW() - INTERVAL '" + str(int(days)) + " days'), "
+        "COUNT(*) FILTER (WHERE status = 'active'), "
+        "COUNT(*) FILTER (WHERE status IN ('removed', 'deleted')), "
+        "COALESCE(SUM(CASE WHEN LOWER(platform) IN ('reddit', 'moltbook') "
+        "  THEN GREATEST(0, COALESCE(upvotes, 0) - 1) "
+        "  ELSE COALESCE(upvotes, 0) END), 0), "
+        "COALESCE(SUM(comments_count), 0), "
+        "COALESCE(SUM(views), 0) "
+        "FROM posts WHERE project_name IS NULL"
+    )
+    row = cur.fetchone()
+    if not row:
+        return {}
+    cols = ["total", "recent", "active", "removed", "total_upvotes", "total_comments", "total_views"]
+    return dict(zip(cols, row))
+
+
+def _platform_breakdown_synthetic_null(conn, days):
+    """NULL-project sibling of ps.get_platform_breakdown."""
+    cur = conn.execute(
+        "SELECT platform, COUNT(*) as cnt FROM posts "
+        "WHERE project_name IS NULL AND posted_at >= NOW() - INTERVAL '" + str(int(days)) + " days' "
+        "GROUP BY platform ORDER BY cnt DESC"
+    )
+    return {row[0]: row[1] for row in cur.fetchall()}
+
+
+def build_project_entry(proj, days, api_key, ph_pid, env, ph_results, platform=None):
+    name = proj["name"]
+    # All main-DB + bookings-DB per-project stats come from one consolidated
+    # HTTP endpoint (HTTP-only migration 2026-06-01). Booking scoping params
+    # (client_slug / table / require_utm) are computed locally from config and
+    # forwarded so the endpoint can read the separate bookings DB server-side.
+    # The endpoint folds the platform filter into post_stats.recent directly,
+    # so the legacy platform-override COUNT is no longer needed here.
+    client_slug = ps.get_client_slug(name)
+    booking_table = ps.get_booking_table(name)
+    require_utm = _bookings_require_utm(name)
+    from http_api import api_get
+    _detail = (api_get("/api/v1/stats/project-detail", query={
+        "project": name,
+        "days": int(days),
+        "platform": platform or "",
+        "client_slug": client_slug or "",
+        "booking_table": booking_table or "cal_bookings",
+        "require_utm": "1" if require_utm else "0",
+    }).get("data") or {})
+    post_stats = dict(_detail.get("post_stats") or {})
+    platforms = _detail.get("platforms") or {}
+    eng_recent = _detail.get("windowed") or {"upvotes": 0, "comments": 0, "views": 0, "views_posts": 0, "post_clicks": 0}
+    eng_period_total = _detail.get("period") or {"upvotes": 0, "comments": 0, "views": 0, "post_clicks": 0}
+    seo_pages_recent = int(_detail.get("seo_pages_recent") or 0)
+
+    domains = ps.get_project_domains(proj)
+    ph_override = proj.get("posthog", {}) or {}
+    ph_key = env.get(ph_override.get("api_key_env", ""), api_key)
+    ph_pid_proj = ph_override.get("project_id", ph_pid)
+    analytics_error = None
+    if domains:
+        per_domain = []
+        for d in domains:
+            stats = ph_results.get((ph_key, ph_pid_proj, d))
+            if stats is None:
+                stats = _empty_domain_stats(d)
+            if stats.get("error") and not analytics_error:
+                analytics_error = stats["error"]
+            per_domain.append(stats)
+        posthog = _ph_combine(per_domain)
+        if analytics_error:
+            posthog["error"] = analytics_error
+    else:
+        posthog = None
+
+    # Window-scoped: `created_paths` is now restricted to pages whose
+    # seo_keywords/gsc_queries `completed_at` falls inside `days`. Top tab →
+    # Pages sub-tab already filters rows on this set, so it becomes "pages
+    # created in the selected period" automatically.
+    # Window-scoped created paths come from the endpoint's db_created_pages
+    # ({host: [paths]}). With a window set, the filesystem scan is intentionally
+    # skipped (static page files carry no trustworthy creation timestamp), so
+    # the DB-derived set is the whole answer — matching _created_paths_for_project
+    # with days set.
+    created_by_domain = {
+        host: set(paths) for host, paths in (_detail.get("db_created_pages") or {}).items()
+    }
+    if posthog is not None:
+        for d, detail in (posthog.get("pageview_details") or {}).items():
+            paths = created_by_domain.get((d or "").lower(), set())
+            detail["created_paths"] = sorted(paths)
+
+    # Preserve the pre-rewrite, domain-wide totals for the analytics-broken
+    # canary below — it's meant to answer "is window.posthog wired up on this
+    # site at all?", which requires domain-level signal, not per-new-page.
+    domain_wide_pv = int(posthog["pageviews"]) if posthog else 0
+    domain_wide_signups = int(posthog["email_signups"]) if posthog else 0
+    domain_wide_sched = int(posthog["schedule_clicks"]) if posthog else 0
+    domain_wide_get_started = int(posthog["get_started_clicks"]) if posthog else 0
+
+    # Recompute funnel totals against the window-scoped created set so the
+    # Status tab → project funnel columns reflect "pageviews / signups /
+    # schedule clicks / download clicks ONLY on pages we generated in this
+    # window" instead of domain-wide traffic. cta_clicks and real_bookings
+    # are not tracked per-page so they stay domain/project-wide.
+    #
+    # Skip entirely when PostHog is errored: the top_pages maps are empty
+    # for errored domains, so scoping would silently collapse everything to
+    # zero. Keep the funnel values as None below so the dashboard renders
+    # 'err' instead of a misleading 0.
+    # Only pageviews get window-scoped to "traffic on pages we generated in
+    # this window". Conversion events (newsletter_subscribed, schedule_click,
+    # get_started_click) fire on dedicated landing pages (/, /use-case, /ig,
+    # etc.), almost never on the freshly-generated /blog/* and /t/* SEO pages
+    # we ship each cycle. Scoping those collapsed every project to 0 and made
+    # the dashboard's Email Signups / Schedule Clicks / Get Started columns
+    # useless. Domain-wide is the honest metric for those.
+    if posthog is not None and not analytics_error:
+        scoped_pv = 0
+        for d, detail in (posthog.get("pageview_details") or {}).items():
+            created = {_norm_path(p) for p in created_by_domain.get((d or "").lower(), set())}
+            if not created:
+                continue
+            for path, cnt in (detail.get("top_pages") or {}).items():
+                if _norm_path(path) in created:
+                    scoped_pv += int(cnt or 0)
+        posthog["pageviews"] = scoped_pv
+
+    bookings = _detail.get("bookings")
+
+    # When the PostHog batch failed, the aggregate numbers on `posthog` are
+    # all 0 but that doesn't mean there are no events, it means we couldn't
+    # read them. Surface null + an error string on the funnel so the
+    # dashboard renders 'err' instead of silently claiming "zero pageviews".
+    if analytics_error:
+        pvs = None
+        ctas = None
+        email_signups = None
+        schedule_clicks = None
+        get_started_clicks = None
+        cross_product_clicks = None
+        ctr = None
+        conv = None
+        dw_pv_out = None
+        dw_signups_out = None
+        dw_sched_out = None
+        dw_get_started_out = None
+        analytics_suspected_broken = False
+    else:
+        pvs = posthog["pageviews"] if posthog else 0
+        ctas = posthog["cta_clicks"] if posthog else 0
+        email_signups = (posthog["email_signups"] if posthog else 0)
+        schedule_clicks = (posthog["schedule_clicks"] if posthog else 0)
+        get_started_clicks = (posthog["get_started_clicks"] if posthog else 0)
+        # Cross-product stays domain-wide on purpose: it's a lightweight
+        # signal ("how many clicks went to a sibling product from this site")
+        # with no per-page top-pages breakdown, so there's nothing to scope.
+        cross_product_clicks = (posthog.get("cross_product_clicks", 0) if posthog else 0)
+        # Domain-wide counterparts for the "scoped (domain-wide)" dashboard
+        # rendering. domain_wide_* were captured before the window-scoping
+        # overwrote posthog["pageviews"] etc.
+        dw_pv_out = domain_wide_pv if posthog else 0
+        dw_signups_out = domain_wide_signups if posthog else 0
+        dw_sched_out = domain_wide_sched if posthog else 0
+        dw_get_started_out = domain_wide_get_started if posthog else 0
+        ctr = (ctas / pvs * 100) if pvs else None
+        conv = None  # computed below once `real` is in scope
+        # Canary: real traffic but zero tracked conversion events almost
+        # always means window.posthog was never wired up on the site (e.g.
+        # Fazm newsletter bug where signups worked but nothing fired to
+        # PostHog). Use domain-wide totals so the signal isn't diluted by
+        # the window-scoped funnel numbers above.
+        analytics_suspected_broken = (domain_wide_pv >= 500) and ((domain_wide_signups + domain_wide_sched + domain_wide_get_started) == 0)
+
+    real = bookings.get("real_bookings", 0) if bookings else 0
+    dm_clicks = int(_detail.get("dm_clicks") or 0)
+    dm_bookings = int(_detail.get("dm_bookings") or 0)
+    amplitude_signups = _amplitude_signups(proj, days, env)
+    if not analytics_error:
+        conv = (real / ctas * 100) if ctas else None
+
+    return {
+        "name": name,
+        "posts": {
+            "total": post_stats.get("total", 0),
+            "recent": post_stats.get("recent", 0),
+            "active": post_stats.get("active", 0),
+            "removed": post_stats.get("removed", 0),
+            # Lifetime engagement across ALL posts for this project (kept for context).
+            "upvotes": post_stats.get("total_upvotes", 0),
+            "comments": post_stats.get("total_comments", 0),
+            "views": post_stats.get("total_views", 0),
+            # Window-scoped engagement: only posts created in the last `days`.
+            "upvotes_recent": eng_recent["upvotes"],
+            "comments_recent": eng_recent["comments"],
+            "views_recent": eng_recent["views"] if eng_recent["views_posts"] > 0 else None,
+            # post_clicks_recent: SUM of post_links.clicks for short links
+            # minted for posts in this project's window. Pre-2026-05-07 rows
+            # may include bot prefetches; post-2026-05-07 rows are humans-only
+            # (Twitter card / LinkedIn unfurl / Slack preview filtered at the
+            # resolver via post_link_clicks.is_bot). See server.js /api/top.
+            "post_clicks_recent": eng_recent["post_clicks"],
+            # Period totals: engagement GAINED during the window across ALL
+            # posts (regardless of posted_at), mirroring the Trends-tab
+            # /api/{views,upvotes,comments,clicks}/per-day SUM. The dashboard
+            # renders each as "<scoped> (<period_total>)" in gray brackets.
+            # post_clicks_period_total counts post_link_clicks (is_bot=FALSE)
+            # in the window joined to this project's posts.
+            "upvotes_period_total": eng_period_total["upvotes"],
+            "comments_period_total": eng_period_total["comments"],
+            "views_period_total": eng_period_total["views"],
+            "post_clicks_period_total": eng_period_total["post_clicks"],
+        },
+        "seo": {"pages_recent": seo_pages_recent},
+        "platforms": platforms,
+        "posthog": posthog,
+        "bookings": bookings,
+        "funnel": {
+            "pageviews": pvs,
+            "cta_clicks": ctas,
+            "email_signups": email_signups,
+            "schedule_clicks": schedule_clicks,
+            "get_started_clicks": get_started_clicks,
+            "cross_product_clicks": cross_product_clicks,
+            "real_bookings": real,
+            "dm_clicks": dm_clicks,
+            "dm_bookings": dm_bookings,
+            # Attributed signups on the client's product (Amplitude), filtered
+            # by the UTM source we forward (config.json projects[].amplitude).
+            # null when the project has no `amplitude` block or the fetch
+            # fails — dashboard falls back to get_started_clicks.
+            "amplitude_signups": amplitude_signups,
+            # Filter shape (e.g. {"utm_source": "studyly.io"}) for tooltip;
+            # null when the project has no `amplitude` block.
+            "amplitude_filter": (proj.get("amplitude") or {}).get("attribution_filter") if proj.get("amplitude") else None,
+            "ctr_pct": ctr,
+            "conv_pct": conv,
+            # Domain-wide siblings: the dashboard shows each as "<scoped>
+            # (<domain>)" so "0 pv for mk0r" doesn't hide 62 real visits
+            # that happened to land on older pages.
+            "domain_pageviews": dw_pv_out,
+            "domain_email_signups": dw_signups_out,
+            "domain_schedule_clicks": dw_sched_out,
+            "domain_get_started_clicks": dw_get_started_out,
+        },
+        "analytics_error": analytics_error,
+        "analytics_suspected_broken": analytics_suspected_broken,
+    }
+
+
+def main():
+    import argparse
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--days", type=int, default=1)
+    parser.add_argument("--project", help="Filter to a single project name")
+    parser.add_argument(
+        "--platform",
+        default="",
+        help=(
+            "Filter to a single platform (twitter|reddit|linkedin|github|moltbook). "
+            "'x' is folded into 'twitter'. Empty / 'all' = no filter. "
+            "Matches the same normalization used by /api/style/stats."
+        ),
+    )
+    parser.add_argument(
+        "--posts-only",
+        action="store_true",
+        help=(
+            "Emit ONLY the per-project posts.* engagement counters; skip the "
+            "PostHog batch (pageviews/CTAs), the bookings DB, Amplitude, and "
+            "SEO page counts. Drops the python runtime from ~30s+ to ~1s. "
+            "Used by /api/funnel/stats as a fast overlay path when the "
+            "dashboard's platform pill changes — those slow sources are "
+            "platform-independent so the all-platform snapshot's values for "
+            "them stay correct, and only the engagement columns need to "
+            "react to the filter."
+        ),
+    )
+    args = parser.parse_args()
+
+    # Normalize platform early; pass empty string when no filter so build_project_entry
+    # can splat it unconditionally without spreading the alias logic everywhere.
+    platform = _normalize_platform(args.platform)
+    # Safety: enforce the same regex /api/funnel/stats accepts so a bad CLI
+    # value can't smuggle SQL through _platform_sql_clause.
+    if platform and not re.match(r"^[a-z0-9_]{1,32}$", platform):
+        print(json.dumps({"error": f"invalid platform: {args.platform!r}"}), file=sys.stdout)
+        sys.exit(1)
+
+    ps.load_env()
+    env = os.environ
+    config = ps.load_config()
+
+    api_key = env.get("POSTHOG_PERSONAL_API_KEY")
+    project_id = env.get("POSTHOG_PROJECT_ID", "330744")
+    bookings_db_url = env.get("BOOKINGS_DATABASE_URL")
+
+    _bridge_per_project_posthog_keys_from_keychain(config, env)
+
+    # Fast path: --posts-only skips the slow PostHog/Amplitude/bookings work
+    # and emits ONLY the per-project posts.* counters. Used as a low-latency
+    # overlay on top of the cached all-platform snapshot when the dashboard's
+    # platform pill changes (see /api/funnel/stats in bin/server.js). Runs
+    # in ~1s instead of ~30s because there are no external HTTP calls AND
+    # the per-project SQL is collapsed into 3 batched GROUP BY queries
+    # (the naive per-project loop pays N x ~180ms Postgres round-trip).
+    if args.posts_only:
+        # posts-only batch: all 8 batched GROUP BY / synthetic queries now run
+        # server-side at /api/v1/stats/posts-batch (HTTP-only). The dicts below
+        # are keyed by project_name with the synthetic NULL bucket under
+        # SYNTHETIC_NO_PROJECT_NAME, matching the prior local computation.
+        from http_api import api_get
+        _batch = (api_get("/api/v1/stats/posts-batch",
+                          query={"days": int(args.days), "platform": platform}).get("data") or {})
+        lifetime = _batch.get("lifetime") or {}
+        windowed = _batch.get("windowed") or {}
+        period = _batch.get("period") or {}
+        period_clicks = _batch.get("period_clicks") or {}
+
+        # Project list: real projects from config + the synthetic NULL bucket.
+        proj_list = list(config.get("projects", [])) + [{"name": SYNTHETIC_NO_PROJECT_NAME}]
+        out_projects = []
+        for proj in proj_list:
+            name = proj["name"]
+            if args.project and args.project.lower() != name.lower():
+                continue
+            life = lifetime.get(name) or {}
+            w = windowed.get(name) or {"upvotes": 0, "comments": 0, "views": 0, "views_posts": 0, "post_clicks": 0}
+            pe = period.get(name) or {"upvotes": 0, "comments": 0, "views": 0}
+            out_projects.append({
+                "name": name,
+                "posts": {
+                    "total": int(life.get("total", 0)),
+                    "recent": int(life.get("recent", 0)),
+                    "active": int(life.get("active", 0)),
+                    "removed": int(life.get("removed", 0)),
+                    "upvotes": int(life.get("total_upvotes", 0)),
+                    "comments": int(life.get("total_comments", 0)),
+                    "views": int(life.get("total_views", 0)),
+                    "upvotes_recent": w["upvotes"],
+                    "comments_recent": w["comments"],
+                    "views_recent": w["views"] if w["views_posts"] > 0 else None,
+                    "post_clicks_recent": w["post_clicks"],
+                    "upvotes_period_total": pe["upvotes"],
+                    "comments_period_total": pe["comments"],
+                    "views_period_total": pe["views"],
+                    "post_clicks_period_total": int(period_clicks.get(name, 0)),
+                },
+            })
+        print(json.dumps({
+            "generated_at": datetime.now(timezone.utc).isoformat(),
+            "days": args.days,
+            "platform": platform or "all",
+            "posts_only": True,
+            "projects": out_projects,
+        }))
+        return
+
+    if not api_key:
+        print(json.dumps({"error": "POSTHOG_PERSONAL_API_KEY not set"}), file=sys.stdout)
+        sys.exit(1)
+
+    # Per-project main-DB + bookings-DB stats now come from HTTP endpoints
+    # (build_project_entry calls /api/v1/stats/project-detail). No direct
+    # Postgres connection is opened here anymore (HTTP-only, 2026-06-01).
+    selected_projects = []
+    for proj in config.get("projects", []):
+        name = proj["name"]
+        if args.project and args.project.lower() != name.lower():
+            continue
+        selected_projects.append(proj)
+
+    # Synthetic '(no project)' bucket: surfaces posts.project_name IS NULL rows
+    # (e.g. IG drafts that landed without a project tag) so the funnel total
+    # lines up with /api/style/stats. No website/landing_pages/posthog block,
+    # so get_project_domains() returns [] -> PostHog/SEO/booking lookups all
+    # become no-ops; per-project SQL helpers route through _project_filter_sql
+    # to use `IS NULL` instead of `= name`.
+    if not args.project or args.project.lower() == SYNTHETIC_NO_PROJECT_NAME.lower():
+        selected_projects.append({"name": SYNTHETIC_NO_PROJECT_NAME})
+
+    # Group domains by (api_key, project_id) so we issue one batched set of
+    # HogQL calls per PostHog bucket instead of one-per-domain. Projects that
+    # share a bucket collapse into a single batched fetch; projects with
+    # dedicated credentials run in their own bucket concurrently.
+    after = (datetime.now(timezone.utc) - timedelta(days=args.days)).strftime("%Y-%m-%dT%H:%M:%S")
+    buckets = {}
+    for proj in selected_projects:
+        domains = ps.get_project_domains(proj)
+        if not domains:
+            continue
+        ph_over = proj.get("posthog", {}) or {}
+        ph_key = env.get(ph_over.get("api_key_env", ""), api_key)
+        ph_pid_proj = ph_over.get("project_id", project_id)
+        bucket_domains = buckets.setdefault((ph_key, ph_pid_proj), set())
+        for d in domains:
+            bucket_domains.add(d)
+
+    # One batched fetch per bucket. When a batch fails after retries, mark
+    # every domain in that bucket as errored rather than rendering zeros.
+    #
+    # Concurrency is capped low (2) on purpose: PostHog's query endpoint
+    # enforces a short-window burst limit (429 "throttled", recovery 1-12s),
+    # and the personal API key is shared across most buckets. Firing 8
+    # buckets at once (each ~10 sequential HogQL queries) created a
+    # thundering herd that all hit the limiter together, all backed off
+    # together, and re-collided on retry until the 4 attempts were
+    # exhausted, marking whole buckets errored ('err' on the dashboard for
+    # every project sharing them). Two-at-a-time keeps us under the burst
+    # ceiling while the Retry-After-honoring backoff in _hogql absorbs the
+    # occasional 429.
+    ph_results = {}
+    if buckets:
+        pool_size = max(1, min(2, len(buckets)))
+        with ThreadPoolExecutor(max_workers=pool_size) as ex:
+            futs = {
+                ex.submit(_ph_batch_counts, k, pid, sorted(ds), after): (k, pid, ds)
+                for (k, pid), ds in buckets.items()
+            }
+            for fut, (k, pid, ds) in futs.items():
+                try:
+                    per_domain = fut.result()
+                    for d, stats in per_domain.items():
+                        ph_results[(k, pid, d)] = stats
+                except HogqlError as e:
+                    msg = f"PostHog unavailable: {e}"
+                    print(f"  PostHog batch error (pid={pid}): {e}", file=sys.stderr)
+                    for d in ds:
+                        ph_results[(k, pid, d)] = _empty_domain_stats(d, error=msg)
+                except Exception as e:
+                    msg = f"PostHog batch error: {e}"
+                    print(f"  PostHog batch unexpected error (pid={pid}): {e}", file=sys.stderr)
+                    for d in ds:
+                        ph_results[(k, pid, d)] = _empty_domain_stats(d, error=msg)
+
+    out_projects = []
+    for proj in selected_projects:
+        name = proj["name"]
+        try:
+            out_projects.append(build_project_entry(
+                proj, args.days, api_key, project_id, env, ph_results,
+                platform=platform,
+            ))
+        except Exception as e:
+            out_projects.append({"name": name, "error": str(e)})
+
+    # `overall.recent` also respects the platform filter so the dashboard's
+    # "N project(s)" / total header stays self-consistent with the per-row data.
+    from http_api import api_get
+    _overall = (api_get("/api/v1/stats/posts-overall",
+                        query={"days": int(args.days), "platform": platform or ""}).get("data") or {})
+    total_all = int(_overall.get("total") or 0)
+    total_recent = int(_overall.get("recent") or 0)
+
+    print(json.dumps({
+        "generated_at": datetime.now(timezone.utc).isoformat(),
+        "days": args.days,
+        "platform": platform or "all",
+        "projects": out_projects,
+        "overall": {"total": total_all, "recent": total_recent},
+    }))
+
+
+if __name__ == "__main__":
+    main()