render-finance-dashboard

You render the booky finance dashboard from the current state of data/finance.db. The visual design is fixed by the template; you only fill it with data. The mechanical part (SQL → HTML, live-price fetch, asset inlining) lives in scripts/render_dashboard.py — invoke it, don't re-derive it.

Where things live

• Canonical renderer: scripts/render_dashboard.py
• HTML template (placeholders in {{NAME}} form): templates/dashboard.html.tpl
• CSS: templates/styles.css — inlined into the HTML at render time
• Chart constructors: templates/charts.js — inlined into the HTML
• Vendored offline assets (gitignored, produced by scripts/bundle-assets.py):
• templates/vendor/chart.umd.min.js
• templates/vendor/chartjs-adapter-date-fns.bundle.min.js
• templates/vendor/fonts-inline.css (EB Garamond + Cormorant Garamond, base64 woff2)
• Telegram credentials: the [telegram] section of .secrets/findash (bot_token=… and chat_id=…)
• Local DB: data/finance.db
• Output: a single self-contained file at output/dashboard.html
• SQL schema and example queries: docs/sqlite-schema.md
• Visual rules: docs/design-system.md

Flow

1. Run the renderer. From the project root:

   python3 scripts/render_dashboard.py

The script handles everything mechanical:

• Live-fetches prices for currently-held securities and USD/ILS, GBP/ILS FX rates from Yahoo Finance (query1.finance.yahoo.com/v8/finance/chart/{symbol}). User-Agent header required.
• Yahoo symbol translation: RR.LSE → RR.L. Quote units: USD direct; LSE pence (÷100 → GBP); TASE agorot (÷100 → ILS).
• Persists fetched prices into prices and the new FX into fx_rates so a history accumulates day-by-day.
• Computes net worth in ILS, brokerage market value, unrealized P/L per position.
• Excludes accounts with closed_on IS NOT NULL from Cash and from the headline (they still show in SQLite-data).
• Sorts the positions table by ILS-equivalent market value, descending (so SPY ranks above RR.LSE despite fewer shares).
• Orders brokerage deposits newest first.
• Builds the SQLite-data debug tabs (first 50 rows + count per table).
• Inlines CSS, fonts, Chart.js, and the chart-app JS into a single HTML file.
• Writes output/dashboard.html and a small meta file at /tmp/dashboard_meta.json (used by the Telegram step).

If any vendor file is missing, the script stops with a clear message. Run python3 scripts/bundle-assets.py once to fix.

1. Send to Telegram. Delivery lives in one committed, allowlisted command so the

unattended daily claude -p run can run it without per-call approval (it can't build the read-secret + curl send inline — the permission analyzer won't auto-approve a command that sources .secrets and expands ${bot_token}, and the Write tool isn't granted unattended). The token is read at runtime and never printed.

   scripts/send_telegram.sh            # add: --note "<line>" to append a caption line

What it does (= the old steps 2+3, now deterministic):

• Sends output/dashboard.html as a document, caption Finance — <AS_OF_DATE> · <net worth>

(literal · U+00B7), read from /tmp/dashboard_meta.json.

• If data/last_sync_summary.md exists, prepends Sync — <AS_OF_DATE> and sends it as

sendMessage (plain text; split into ≤4000-char parts on \n boundaries), then deletes it on success. Absent → prints Sync summary: Skipped (no summary).

• Telegram not configured (no [telegram] in .secrets/findash) → prints Telegram delivery skipped: … and exits 0 (local

file still useful). Never deletes the local dashboard on a Telegram failure.

Pass --note "<line>" to append a caption line — the morning flow uses this to surface a best-effort fetch failure. The script prints Sent to chat <id> / Telegram delivery failed: <description> / the Sync summary: … line — relay them in step 3.

1. Report. Print:

• The absolute path of output/dashboard.html.
• The Telegram delivery status: Sent to chat <chat_id>, Telegram delivery skipped: …, or Telegram delivery failed: <description>.
• The sync summary delivery status: Sync summary: Sent (<N> bullets), Sync summary: Skipped (no summary), or Sync summary: Failed: <description>.
• Any tickers Yahoo couldn't price (the script logs them on FAILED: — flag them so the user knows which positions fell back to cost basis).

Where judgment still lives (not in the script)

• Deciding whether a fresh balance figure looks plausible or whether a sync mis-parsed something. If a number looks wrong, check the source doc and explain rather than rendering a broken figure (e.g. a closed Hapoalim sub-account showing an apparent negative balance from a parser artifact when the real closing balance was 0).
• Whether to mention a particular caveat in the Overview footnote (e.g. "no price for X" or "this account was just opened").
• The contents of the Telegram caption when the dashboard contains something unusual.

Principles

• Don't invent design. All visual choices are in styles.css and the template. Don't add inline styles, classes, or markup not already defined.
• Be honest with empty states. If a section has no data yet, show a brief italic muted note ("No payslips yet"), not a fake placeholder row.
• Number formatting: group thousands; 2 decimals for amounts <10,000, 0 decimals for ≥10,000. Right-align money columns.
• Hebrew text passes through unchanged. dir="auto" on table cells is already handled by the script.
• The "SQLite data" tab is for debugging. Show raw rows truthfully. Don't reformat or hide columns.