networthy 0.1.0__tar.gz

networthy-0.1.0/PKG-INFO

@@ -0,0 +1,235 @@
+Metadata-Version: 2.4
+Name: networthy
+Version: 0.1.0
+Summary: Track your complete net worth on your own machine — NSDL/CAMS statement parsing, live prices, retirement projections. India-first, private by design.
+Author: Manoj Awasthi
+License: MIT
+Project-URL: Homepage, https://networthyhq.com
+Project-URL: Source, https://github.com/awmanoj/networthy
+Keywords: net-worth,personal-finance,india,nsdl,cas,mutual-funds,fire
+Classifier: Development Status :: 4 - Beta
+Classifier: Environment :: Web Environment
+Classifier: Framework :: FastAPI
+Classifier: Intended Audience :: End Users/Desktop
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Programming Language :: Python :: 3
+Classifier: Topic :: Office/Business :: Financial
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+Requires-Dist: fastapi==0.115.6
+Requires-Dist: uvicorn[standard]==0.34.0
+Requires-Dist: jinja2==3.1.5
+Requires-Dist: python-multipart==0.0.20
+Requires-Dist: pdfplumber==0.11.5
+Requires-Dist: pikepdf==9.4.2
+Requires-Dist: httpx==0.28.1
+
+# Networthy HQ
+
+**A private, self-hosted net-worth tracker for Indian investors.** Upload your
+**NSDL CAS** and **CAMS/KFintech** statements, add everything else by hand — property,
+gold, foreign equity, crypto, bank balances, loans — and see your complete net worth
+in one place, priced live where it can be.
+
+> **Privacy is the whole point.** Statements are parsed **on your own server** and the
+> parsed database lives under `data/` (gitignored). Nothing about your holdings ever
+> leaves the machine — the *only* things that egress are public price lookups by symbol
+> (a ticker, a currency pair, a coin), never a value, quantity, PAN, or identity.
+
+**Try it without signing up:** the landing page has an **Explore the live demo** button
+(`GET /demo`) that drops you into a fully-loaded demo account — no email required.
+
+![Dashboard](docs/screenshots/dashboard.png)
+
+---
+
+## What it does
+
+- **Parses password-protected CAS PDFs** — NSDL e-CAS (demat holdings across NSDL +
+  CDSL, plus mutual-fund folios) and CAMS/KFintech CAS (all-AMC mutual funds). Decrypt
+  → extract → classify by asset class, all locally.
+- **A complete Assets & Liabilities tree** — every asset class as a navigable leaf:
+  mutual funds, direct & foreign equity, crypto, fixed income (PPF/EPF/FDs/bonds/NPS),
+  gold & silver, physical gold, real estate, alternate investments, private business,
+  bank & cash, forex — netted against every kind of loan.
+- **Live valuation** — equities from Yahoo Finance (by ticker), mutual-fund NAVs from
+  AMFI's public bulk feed (looked up locally), USD/FX and crypto and gold priced live,
+  all cached and fail-soft so a slow endpoint never breaks a page.
+- **Expenses planner** — recurring spend normalised to a monthly/annual burn, with the
+  net-worth connection: runway and a FIRE target (25×, the 4% rule).
+- **Goals** — target-by-date planning that computes the **monthly SIP** needed to get
+  there, plus a read-only Retirement (FIRE) goal mirrored from your expenses.
+- **"Where do you stand?"** — rank your net worth among adults in India, the US, and the
+  world (computed client-side; nothing you type is sent anywhere).
+- **Multi-user** — sign in with email + a one-time code. Each account's data is isolated.
+- **Email digests** — an optional daily net-worth pulse and a weekly breakdown.
+
+## Screens
+
+| | |
+|---|---|
+| ![Net worth tree](docs/screenshots/networth.png) | ![Net worth over time](docs/screenshots/nsdl-cas.png) |
+| **Net worth** — the full Assets/Liabilities tree, values rolled up and priced live. | **NSDL CAS** — net worth over time from each uploaded statement. |
+| ![Mutual funds leaf](docs/screenshots/mutual-funds.png) | ![Goals](docs/screenshots/goals.png) |
+| **A data-backed leaf** — holdings from your CAMS import, live NAVs and values. | **Goals** — target, date, and the monthly SIP to reach it. |
+| ![Expenses](docs/screenshots/expenses.png) | |
+| **Expenses** — monthly/annual burn, category breakdown, runway & FIRE. | |
+
+## How it works
+
+```
+upload CAS PDF(s)  →  parse (decrypt + extract + classify)  →  SQLite (under data/)
+                                                                     │
+   manual entries (property, gold, loans, …) ──────────────────────┤
+                                                                     ▼
+                    live pricing (Yahoo / AMFI, by symbol only)  →  Dashboard · Net worth · Goals
+```
+
+1. Download your CAS — NSDL e-CAS from [nsdl.co.in](https://nsdl.co.in), or a CAMS CAS
+   from [camsonline.com](https://www.camsonline.com/Investors/Statements/Consolidated-Account-Statement).
+   It arrives as a **password-protected PDF** (password is usually your PAN in CAPITALS).
+2. Upload it (the app remembers your PAN as the password so you don't retype it).
+3. Add anything a statement doesn't cover — property, physical gold, bank balances,
+   foreign equity, crypto, loans — by hand.
+4. The Dashboard, Net-worth tree, and Goals reflect it all, valued live where possible.
+
+## Tech stack
+
+Server-rendered **FastAPI + Jinja2**, **SQLite** (stdlib `sqlite3`), no frontend
+framework and a single hand-written CSS design system ("Ink Navy & Copper", light + dark).
+PDF parsing via **pikepdf** (decrypt) + **pdfplumber** (text). Python 3.11+.
+
+## Run it on your own machine
+
+One command. No account, no sign-in, no server — your statements are parsed on your
+laptop and the database never leaves it.
+
+```bash
+uvx networthy
+```
+
+That's it: it starts on `http://127.0.0.1:8321`, opens your browser, and signs you in
+automatically (there's nobody else to authenticate against on your own machine).
+
+<details>
+<summary>Don't have <code>uv</code>?</summary>
+
+```bash
+curl -LsSf https://astral.sh/uv/install.sh | sh   # macOS / Linux
+# Windows: powershell -c "irm https://astral.sh/uv/install.ps1 | iex"
+```
+
+Or use pipx: `pipx run networthy`. Or plain pip: `pip install networthy && networthy`.
+</details>
+
+**Options**
+
+```bash
+networthy --port 9000              # pick a port (default 8321, or any free one)
+networthy --data-dir ~/my-finances # where the database lives
+networthy --no-browser             # don't open a tab
+```
+
+Your data is stored at:
+
+| macOS   | `~/Library/Application Support/Networthy/` |
+|---------|--------------------------------------------|
+| Linux   | `~/.local/share/networthy/`                |
+| Windows | `%APPDATA%\Networthy\`                     |
+
+It's a single SQLite file — back it up by copying it, move machines by moving it.
+
+### With Docker instead
+
+```bash
+docker run -p 8321:8321 -v networthy:/app/data \
+  -e NETWORTHY_LOCAL=1 -e APP_PORT=8321 awmanoj/networthy
+```
+
+### Developing on it
+
+```bash
+python3 -m venv .venv
+source .venv/bin/activate
+pip install -r requirements.txt
+
+NETWORTHY_LOCAL=1 uvicorn app.main:app --reload    # http://127.0.0.1:8000
+```
+
+Without `NETWORTHY_LOCAL=1` you get the hosted behaviour — sign in with your email, and
+the one-time code is printed to the server log (no email provider needed).
+
+## Testing
+
+Tests target the fragile logic — parsers, classification, pricing, the net-worth math,
+and the web routes — without needing a real password-protected PDF.
+
+```bash
+python -m pytest                                   # all
+python -m pytest tests/test_parser.py              # one file
+python -m pytest tests/test_goals.py::test_plan_active_required_monthly   # one test
+```
+
+> Use `python -m pytest` (not bare `pytest`) so the repo root is on `sys.path`.
+
+## Deployment
+
+Containerised, designed to sit behind a reverse proxy (e.g. Caddy for auto-TLS). The
+container port is set by `APP_PORT` (default 8000); the SQLite DB persists in a Docker
+volume mounted at `/app/data`.
+
+```bash
+docker build -t networthy .
+DOCKERHUB_USER=<name> ./deploy.sh [tag]   # build + push to Docker Hub
+DOCKERHUB_USER=<name> ./run.sh   [tag]    # run on the server (published on :8321)
+```
+
+**Email digests** (optional) recompute every user's net worth live, record a daily
+history point, and email a change summary (no-ops to a log without `RESEND_API_KEY`):
+
+```cron
+30 12 * * 1-6 docker exec networthy python -m app.digest daily    # 6 PM IST, Mon–Sat
+30 12 * * 0   docker exec networthy python -m app.digest weekly   # Sunday
+```
+
+**Backups** — `backup.sh` takes a consistent SQLite online backup, gzips it, and prunes
+old copies. Cron it every few hours and copy the archives off-box.
+
+## Privacy invariant
+
+This is load-bearing, not a footnote:
+
+- Statement contents and parsed financial data are **never written anywhere outside
+  `data/`**, and `data/`, `*.pdf`, `*.db` are gitignored.
+- The **one sanctioned egress** is `app/prices.py`, kept deliberately narrow: it sends
+  only a **public symbol** to a price API — an equity ticker to Yahoo, a currency pair,
+  a coin, a gold symbol — and looks mutual-fund NAVs up **locally** from AMFI's bulk
+  file. Never units, values, holdings, PAN, or identity. Every lookup fails soft, so the
+  view always falls back to the statement value.
+
+## Project structure
+
+```
+app/
+  main.py            FastAPI routes (dashboard, net-worth tree, goals, expenses, CAS)
+  models.py          Dataclasses shared across parser / storage / web
+  storage.py         SQLite persistence, per-user isolation
+  classify.py        Asset-class rule engine (section > ISIN > keywords)
+  prices.py          The ONLY module that egresses (live prices, by symbol)
+  networth.py        The declarative Assets/Liabilities tree + roll-up
+  expenses.py        Recurring-spend model (burn, FIRE)
+  goals.py           Target-by-date planning (required monthly SIP)
+  wealth.py          Net-worth percentile ranking
+  digest.py          Daily/weekly email digests
+  parser/
+    nsdl_cas.py      NSDL CAS parsing (the fragile core)
+    cams_cas.py      CAMS/KFintech CAS parsing
+    _common.py       Shared decrypt / text / float helpers
+  templates/         Jinja2 templates
+  static/            CSS + a tiny bit of JS (theme toggle, chart)
+data/                SQLite DB + uploads (gitignored)
+tests/               Parser, classify, pricing, net-worth, and route tests
+```
+
+For a deeper architectural tour — the parsing pipeline, the classification traps, the
+net-worth roll-up, and the design-system conventions — see [`CLAUDE.md`](CLAUDE.md).

networthy-0.1.0/README.md

@@ -0,0 +1,209 @@
+# Networthy HQ
+
+**A private, self-hosted net-worth tracker for Indian investors.** Upload your
+**NSDL CAS** and **CAMS/KFintech** statements, add everything else by hand — property,
+gold, foreign equity, crypto, bank balances, loans — and see your complete net worth
+in one place, priced live where it can be.
+
+> **Privacy is the whole point.** Statements are parsed **on your own server** and the
+> parsed database lives under `data/` (gitignored). Nothing about your holdings ever
+> leaves the machine — the *only* things that egress are public price lookups by symbol
+> (a ticker, a currency pair, a coin), never a value, quantity, PAN, or identity.
+
+**Try it without signing up:** the landing page has an **Explore the live demo** button
+(`GET /demo`) that drops you into a fully-loaded demo account — no email required.
+
+![Dashboard](docs/screenshots/dashboard.png)
+
+---
+
+## What it does
+
+- **Parses password-protected CAS PDFs** — NSDL e-CAS (demat holdings across NSDL +
+  CDSL, plus mutual-fund folios) and CAMS/KFintech CAS (all-AMC mutual funds). Decrypt
+  → extract → classify by asset class, all locally.
+- **A complete Assets & Liabilities tree** — every asset class as a navigable leaf:
+  mutual funds, direct & foreign equity, crypto, fixed income (PPF/EPF/FDs/bonds/NPS),
+  gold & silver, physical gold, real estate, alternate investments, private business,
+  bank & cash, forex — netted against every kind of loan.
+- **Live valuation** — equities from Yahoo Finance (by ticker), mutual-fund NAVs from
+  AMFI's public bulk feed (looked up locally), USD/FX and crypto and gold priced live,
+  all cached and fail-soft so a slow endpoint never breaks a page.
+- **Expenses planner** — recurring spend normalised to a monthly/annual burn, with the
+  net-worth connection: runway and a FIRE target (25×, the 4% rule).
+- **Goals** — target-by-date planning that computes the **monthly SIP** needed to get
+  there, plus a read-only Retirement (FIRE) goal mirrored from your expenses.
+- **"Where do you stand?"** — rank your net worth among adults in India, the US, and the
+  world (computed client-side; nothing you type is sent anywhere).
+- **Multi-user** — sign in with email + a one-time code. Each account's data is isolated.
+- **Email digests** — an optional daily net-worth pulse and a weekly breakdown.
+
+## Screens
+
+| | |
+|---|---|
+| ![Net worth tree](docs/screenshots/networth.png) | ![Net worth over time](docs/screenshots/nsdl-cas.png) |
+| **Net worth** — the full Assets/Liabilities tree, values rolled up and priced live. | **NSDL CAS** — net worth over time from each uploaded statement. |
+| ![Mutual funds leaf](docs/screenshots/mutual-funds.png) | ![Goals](docs/screenshots/goals.png) |
+| **A data-backed leaf** — holdings from your CAMS import, live NAVs and values. | **Goals** — target, date, and the monthly SIP to reach it. |
+| ![Expenses](docs/screenshots/expenses.png) | |
+| **Expenses** — monthly/annual burn, category breakdown, runway & FIRE. | |
+
+## How it works
+
+```
+upload CAS PDF(s)  →  parse (decrypt + extract + classify)  →  SQLite (under data/)
+                                                                     │
+   manual entries (property, gold, loans, …) ──────────────────────┤
+                                                                     ▼
+                    live pricing (Yahoo / AMFI, by symbol only)  →  Dashboard · Net worth · Goals
+```
+
+1. Download your CAS — NSDL e-CAS from [nsdl.co.in](https://nsdl.co.in), or a CAMS CAS
+   from [camsonline.com](https://www.camsonline.com/Investors/Statements/Consolidated-Account-Statement).
+   It arrives as a **password-protected PDF** (password is usually your PAN in CAPITALS).
+2. Upload it (the app remembers your PAN as the password so you don't retype it).
+3. Add anything a statement doesn't cover — property, physical gold, bank balances,
+   foreign equity, crypto, loans — by hand.
+4. The Dashboard, Net-worth tree, and Goals reflect it all, valued live where possible.
+
+## Tech stack
+
+Server-rendered **FastAPI + Jinja2**, **SQLite** (stdlib `sqlite3`), no frontend
+framework and a single hand-written CSS design system ("Ink Navy & Copper", light + dark).
+PDF parsing via **pikepdf** (decrypt) + **pdfplumber** (text). Python 3.11+.
+
+## Run it on your own machine
+
+One command. No account, no sign-in, no server — your statements are parsed on your
+laptop and the database never leaves it.
+
+```bash
+uvx networthy
+```
+
+That's it: it starts on `http://127.0.0.1:8321`, opens your browser, and signs you in
+automatically (there's nobody else to authenticate against on your own machine).
+
+<details>
+<summary>Don't have <code>uv</code>?</summary>
+
+```bash
+curl -LsSf https://astral.sh/uv/install.sh | sh   # macOS / Linux
+# Windows: powershell -c "irm https://astral.sh/uv/install.ps1 | iex"
+```
+
+Or use pipx: `pipx run networthy`. Or plain pip: `pip install networthy && networthy`.
+</details>
+
+**Options**
+
+```bash
+networthy --port 9000              # pick a port (default 8321, or any free one)
+networthy --data-dir ~/my-finances # where the database lives
+networthy --no-browser             # don't open a tab
+```
+
+Your data is stored at:
+
+| macOS   | `~/Library/Application Support/Networthy/` |
+|---------|--------------------------------------------|
+| Linux   | `~/.local/share/networthy/`                |
+| Windows | `%APPDATA%\Networthy\`                     |
+
+It's a single SQLite file — back it up by copying it, move machines by moving it.
+
+### With Docker instead
+
+```bash
+docker run -p 8321:8321 -v networthy:/app/data \
+  -e NETWORTHY_LOCAL=1 -e APP_PORT=8321 awmanoj/networthy
+```
+
+### Developing on it
+
+```bash
+python3 -m venv .venv
+source .venv/bin/activate
+pip install -r requirements.txt
+
+NETWORTHY_LOCAL=1 uvicorn app.main:app --reload    # http://127.0.0.1:8000
+```
+
+Without `NETWORTHY_LOCAL=1` you get the hosted behaviour — sign in with your email, and
+the one-time code is printed to the server log (no email provider needed).
+
+## Testing
+
+Tests target the fragile logic — parsers, classification, pricing, the net-worth math,
+and the web routes — without needing a real password-protected PDF.
+
+```bash
+python -m pytest                                   # all
+python -m pytest tests/test_parser.py              # one file
+python -m pytest tests/test_goals.py::test_plan_active_required_monthly   # one test
+```
+
+> Use `python -m pytest` (not bare `pytest`) so the repo root is on `sys.path`.
+
+## Deployment
+
+Containerised, designed to sit behind a reverse proxy (e.g. Caddy for auto-TLS). The
+container port is set by `APP_PORT` (default 8000); the SQLite DB persists in a Docker
+volume mounted at `/app/data`.
+
+```bash
+docker build -t networthy .
+DOCKERHUB_USER=<name> ./deploy.sh [tag]   # build + push to Docker Hub
+DOCKERHUB_USER=<name> ./run.sh   [tag]    # run on the server (published on :8321)
+```
+
+**Email digests** (optional) recompute every user's net worth live, record a daily
+history point, and email a change summary (no-ops to a log without `RESEND_API_KEY`):
+
+```cron
+30 12 * * 1-6 docker exec networthy python -m app.digest daily    # 6 PM IST, Mon–Sat
+30 12 * * 0   docker exec networthy python -m app.digest weekly   # Sunday
+```
+
+**Backups** — `backup.sh` takes a consistent SQLite online backup, gzips it, and prunes
+old copies. Cron it every few hours and copy the archives off-box.
+
+## Privacy invariant
+
+This is load-bearing, not a footnote:
+
+- Statement contents and parsed financial data are **never written anywhere outside
+  `data/`**, and `data/`, `*.pdf`, `*.db` are gitignored.
+- The **one sanctioned egress** is `app/prices.py`, kept deliberately narrow: it sends
+  only a **public symbol** to a price API — an equity ticker to Yahoo, a currency pair,
+  a coin, a gold symbol — and looks mutual-fund NAVs up **locally** from AMFI's bulk
+  file. Never units, values, holdings, PAN, or identity. Every lookup fails soft, so the
+  view always falls back to the statement value.
+
+## Project structure
+
+```
+app/
+  main.py            FastAPI routes (dashboard, net-worth tree, goals, expenses, CAS)
+  models.py          Dataclasses shared across parser / storage / web
+  storage.py         SQLite persistence, per-user isolation
+  classify.py        Asset-class rule engine (section > ISIN > keywords)
+  prices.py          The ONLY module that egresses (live prices, by symbol)
+  networth.py        The declarative Assets/Liabilities tree + roll-up
+  expenses.py        Recurring-spend model (burn, FIRE)
+  goals.py           Target-by-date planning (required monthly SIP)
+  wealth.py          Net-worth percentile ranking
+  digest.py          Daily/weekly email digests
+  parser/
+    nsdl_cas.py      NSDL CAS parsing (the fragile core)
+    cams_cas.py      CAMS/KFintech CAS parsing
+    _common.py       Shared decrypt / text / float helpers
+  templates/         Jinja2 templates
+  static/            CSS + a tiny bit of JS (theme toggle, chart)
+data/                SQLite DB + uploads (gitignored)
+tests/               Parser, classify, pricing, net-worth, and route tests
+```
+
+For a deeper architectural tour — the parsing pipeline, the classification traps, the
+net-worth roll-up, and the design-system conventions — see [`CLAUDE.md`](CLAUDE.md).

networthy-0.1.0/app/__init__.py

@@ -0,0 +1,3 @@
+"""Networthy — net worth tracking from NSDL CAS statements."""
+
+__version__ = "0.1.0"

networthy-0.1.0/app/analytics.py

@@ -0,0 +1,139 @@
+"""Business analytics — adoption metrics for the owner, computed live from the app's
+own tables. First-party by design: nothing here egresses, and it deliberately reads
+only account/usage **metadata** (signups, logins, which features were touched), never
+any financial values — so this surface holds no holdings data even if it leaked.
+
+The shared demo account is excluded from every number.
+"""
+
+from __future__ import annotations
+
+from datetime import date, datetime, timedelta
+
+from . import storage
+from .demo import DEMO_EMAIL
+
+# (label, table, extra WHERE condition or None) — a "feature" is used once a user has
+# at least one row here. Counts distinct users, demo excluded.
+_FEATURES: list[tuple[str, str, str | None]] = [
+    ("Uploaded NSDL CAS", "snapshots", None),
+    ("Imported CAMS", "networth_holdings", "source = 'cams'"),
+    ("Set a goal", "goals", None),
+    ("Tracked expenses", "expenses", None),
+    ("Real estate", "property_holdings", None),
+    ("Crypto", "crypto_holdings", None),
+    ("US equity", "foreign_holdings", None),
+    ("Fixed income (manual)", "manual_holdings", None),
+    ("Bank & cash", "bank_cash", None),
+    ("Physical gold", "gold_items", None),
+    ("Liabilities", "liabilities", None),
+    ("Alternate investments", "alt_investments", None),
+]
+
+# Tables that mean "this user put real data in" — for the activation funnel.
+_ASSET_TABLES = [
+    "snapshots", "networth_holdings", "property_holdings", "bank_cash",
+    "manual_holdings", "gold_items", "alt_investments", "crypto_holdings",
+    "foreign_holdings", "forex_holdings", "business_holdings", "liabilities",
+    "goals", "expenses",
+]
+
+
+def _iso(days_ago: int) -> str:
+    return (datetime.utcnow() - timedelta(days=days_ago)).strftime("%Y-%m-%d %H:%M:%S")
+
+
+def overview() -> dict:
+    with storage._connect() as conn:
+        demo = conn.execute("SELECT id FROM users WHERE email = ?", (DEMO_EMAIL,)).fetchone()
+        demo_id = demo["id"] if demo else -1
+
+        def scalar(sql: str, params: tuple = ()) -> int:
+            return conn.execute(sql, params).fetchone()[0]
+
+        total = scalar("SELECT COUNT(*) FROM users WHERE id != ?", (demo_id,))
+        new_today = scalar(
+            "SELECT COUNT(*) FROM users WHERE id != ? AND created_at >= ?",
+            (demo_id, _iso(1)))
+        new_7d = scalar(
+            "SELECT COUNT(*) FROM users WHERE id != ? AND created_at >= ?",
+            (demo_id, _iso(7)))
+        new_30d = scalar(
+            "SELECT COUNT(*) FROM users WHERE id != ? AND created_at >= ?",
+            (demo_id, _iso(30)))
+
+        # Sign-ins (from the durable login_events log).
+        logins_total = scalar("SELECT COUNT(*) FROM login_events WHERE user_id != ?", (demo_id,))
+        logins_24h = scalar(
+            "SELECT COUNT(DISTINCT user_id) FROM login_events WHERE user_id != ? AND created_at >= ?",
+            (demo_id, _iso(1)))
+        logins_7d = scalar(
+            "SELECT COUNT(DISTINCT user_id) FROM login_events WHERE user_id != ? AND created_at >= ?",
+            (demo_id, _iso(7)))
+        logins_30d = scalar(
+            "SELECT COUNT(DISTINCT user_id) FROM login_events WHERE user_id != ? AND created_at >= ?",
+            (demo_id, _iso(30)))
+        # Returning = signed in on ≥2 distinct days.
+        returning = scalar(
+            """
+            SELECT COUNT(*) FROM (
+                SELECT user_id FROM login_events WHERE user_id != ?
+                GROUP BY user_id HAVING COUNT(DISTINCT date(created_at)) >= 2
+            )
+            """, (demo_id,))
+
+        # Cumulative signups over time (for the growth chart).
+        rows = conn.execute(
+            "SELECT date(created_at) d, COUNT(*) n FROM users WHERE id != ? GROUP BY d ORDER BY d",
+            (demo_id,)).fetchall()
+        cum, chart = 0, []
+        for r in rows:
+            cum += r["n"]
+            chart.append({"date": r["d"], "value": cum})
+
+        # Feature adoption.
+        features = []
+        for label, table, cond in _FEATURES:
+            where = f"user_id != ?" + (f" AND {cond}" if cond else "")
+            n = scalar(f"SELECT COUNT(DISTINCT user_id) FROM {table} WHERE {where}", (demo_id,))
+            features.append({"label": label, "users": n,
+                             "pct": (n / total * 100.0) if total else 0.0})
+        features.sort(key=lambda f: f["users"], reverse=True)
+
+        # Activation funnel: signed up → put in real data → came back.
+        union = " UNION ".join(
+            f"SELECT user_id FROM {t} WHERE user_id != {demo_id}" for t in _ASSET_TABLES)
+        activated = scalar(f"SELECT COUNT(*) FROM (SELECT DISTINCT user_id FROM ({union}))")
+        funnel = [
+            {"label": "Signed up", "users": total, "pct": 100.0},
+            {"label": "Added real data", "users": activated,
+             "pct": (activated / total * 100.0) if total else 0.0},
+            {"label": "Came back (≥2 days)", "users": returning,
+             "pct": (returning / total * 100.0) if total else 0.0},
+        ]
+
+        # Recent signups with a last-seen — the email listing.
+        recent = conn.execute(
+            """
+            SELECT u.email, u.created_at,
+                   (SELECT MAX(created_at) FROM login_events e WHERE e.user_id = u.id) last_seen
+            FROM users u WHERE u.id != ?
+            ORDER BY u.created_at DESC LIMIT 200
+            """, (demo_id,)).fetchall()
+
+    def fmt(s: str | None) -> str | None:
+        if not s:
+            return None
+        try:
+            return datetime.strptime(s[:19], "%Y-%m-%d %H:%M:%S").strftime("%d %b %Y")
+        except ValueError:
+            return s
+
+    return {
+        "total_users": total, "new_today": new_today, "new_7d": new_7d, "new_30d": new_30d,
+        "returning": returning, "logins_total": logins_total,
+        "logins_24h": logins_24h, "logins_7d": logins_7d, "logins_30d": logins_30d,
+        "signup_chart": chart, "features": features, "funnel": funnel,
+        "recent": [{"email": r["email"], "joined": fmt(r["created_at"]),
+                    "last_seen": fmt(r["last_seen"])} for r in recent],
+    }