Franklin

One of the app's features lets you select Franklin ETFs in the Watchlist and aggregate their holdings to see how often each ticker appears across the selected funds. Repeated holdings make overlapping exposure visible: the more selected funds include a ticker, the greater its potential influence on the portfolio; gains in that holding may help, while declines may hurt, and actual impact also depends on each fund's position size. Another feature makes it faster and easier to find funds with stronger growth over different periods, higher dividend yields or distributions, greater Total Return (price performance plus dividends), and other key performance metrics. A single-file client-side tool that reads the generated ./api/franklin static feed (franklintempleton.com ETF listings, per-fund product pages, Yahoo Finance daily history, SEC EDGAR N-PORT-P as holdings fallback) into a searchable ETF/category catalog with per-fund tabs, watchlist aggregation, ticker copy and CSV/TXT export - the same look, feel, columns and business logic as the sibling applications.

bunx degit daggerok/Franklin#main ./12345 && cd $_
bun install
bun run serve
open http://localhost:1234

Production build (Parcel, Tailwind v4) into dist/ with api/ copied to dist/api:

bun run build
bun run build-github-pages

build-github-pages adds the /Franklin/ public URL and is what the Pages workflow runs.

The published application is available at https://daggerok.github.io/Franklin/

Every column of the ETF catalog and of the Watchlist, Holdings, History and Distributions tabs has a type: text (ABC), number (123), percentage (%), money ($), date (D), date and time (DT) or time of day (T). The type is detected from the texts the column shows (80% of the filled cells must agree, otherwise text) and is written in the badge next to the column title: click it to cycle the type, Shift+click to return to auto-detection. Dates are read as 2024-06-15, 6/15/2024, 15.06.2024, Jun 15, 2024 or 15-Jun-2024, date and time as 2024-06-15T09:30:00Z or 2024-06-15 09:30, time as 09:30, 16:00:00 or 9:30 PM

A row of filter inputs sits under the column headers (the Filters button hides it, Clear filters empties it). Filters of different columns are combined with AND, the search box applies on top, and Copy Tickers and the exports use the filtered rows. Filters and type overrides are remembered in the browser. Sticky # (next to Filters, off by default, remembered in the browser) numbers the rows by their rank in the table sorted by the current column before the column filters, so a filtered fund keeps its rank and the numbers keep gaps; the sort, the search and the category and blacklist choices rank again. The catalog starts sorted by Net Assets, largest first, unavailable values sort last in both directions, and every export starts with the # column. The red Clear button opens a dialog that lists what can be reset (the selection, searches, sort order, open tab, shown columns, column filters, remembered table views and the blacklist), all ticked the first time and afterwards as they were left at the last OK; Enter confirms, Esc or a click outside cancels, the theme is always kept, so the page looks like a first visit (also after a reload)

Inside one filter: a space means AND, a comma means OR, a leading ! means NOT, ? matches an empty or unavailable value and !? a value that is there; a value that is unavailable matches only ? and negated conditions. An unquoted space ends the value, so quote values that contain one (>="2024-06-15 09:30")

The Columns menu next to Filters lists every column of the ETF table from the first to the last, all of them shown by default, with a search box and the All, Clear, Toggle and Reset buttons. Use and Ticker are listed but locked. Hiding a column only removes it from the table: the filters, the sorting, the exports and Copy Tickers still use it. The choice is remembered in the browser (localStorage, never the data) and the menu is shown on the ETF catalog only

The asset classes are one Asset classes multi-select next to the All ETFs pill instead of one tab per class: every class is selected by default (= all ETFs), Only or unchecking narrows the table, and the All ETFs pill is lit only while nothing narrows it (all or none of the classes checked); clicking the pill clears the selection. The choice is remembered in the browser (localStorage, never the data)

Run the updater with Bun:

bun test
bun ./scripts/update-data.ts

Run bun ./scripts/update-data.ts --help to print every control with its default and examples.

Defaults live in scripts/update-data.config.json (every control as a string). Explicit environment variables override the file, and a blank environment variable overrides it with an empty value. The Update Franklin ETF data GitHub Actions workflow uses the same resolveControls resolver as the CLI: individual workflow_dispatch inputs are blank by default and inherit the file, the advanced input accepts a JSON object with any control (for example {"VERBOSE":"true","PERFORMANCE_10Y":"5:"}), and the precedence is file defaults < advanced JSON < nonblank individual inputs < protected Actions variable/env. GitHub allows at most 25 inputs, so HISTORY_RANGE, PERFORMANCE_10Y, TOTAL_RETURN_10Y, VERBOSE and SEC_UA are set through advanced. Scheduled runs (Sundays at 00:00 UTC) have no inputs and use the file defaults. The real SEC contact belongs in the protected repository Actions variable SEC_UA, which wins when nonblank; the config default is daggerok ETF feed daggerok@gmail.com. The workflow always writes to api/franklin and commits only that directory. All supplied filters use AND logic.

Each fund carries a derived metrics object that powers the catalog columns shared with the sibling sites:

• ytd / tr1y - official YTD and 1-year returns -> YTD Return, TR 1Y

• cagr3y / cagr5y / cagr10y - published annualized 3Y/5Y/10Y figures -> CAGR 3Y/5Y/10Y

• tr3y / tr5y / tr10y - cumulative 3Y/5Y/10Y figures in percent, ((1 + CAGR/100)^n - 1) * 100 -> TR 3Y/5Y/10Y

• siAnn - since-inception annualized -> SI Ann.

• dividendYield - the official figure when the product page prints one, otherwise an estimate; meta.json yields.dividendYieldKind says which. Order: (1) the official 12-month yield (official 12-month distribution yield from the product page), (2) the official Distribution Rate, labelled Distribution Rate at NAV published on the official Franklin fund page as of <date> with yields.dividendYieldAsOfDate, (3) only when the page prints neither, the indicated yield (latest Yahoo distribution x payments per year / price, labelled indicated: ...). The page defines the Distribution Rate (its footnote) as the most recent distribution amount paid, annualized and divided by the closing market price or NAV as of the printed date (the page shows it "At Net Asset Value (NAV)", updated daily); special distributions are excluded, it is not a trailing 12-month figure and not a quotation of performance. A rate of a monthly payer therefore reacts to the last payment, and funds that print no rate (for example semiannual payers) keep the indicated yield

• dividendYieldBasis - short code for the definition behind dividendYield, null exactly when dividendYield is null; it is retained together with the yield it describes. Franklin uses three of the shared codes:

  Code	Meaning for Franklin	yields.dividendYieldKind in meta.json
  official-trailing-12m	the 12-month distribution yield printed on the product page	official 12-month distribution yield from the product page
  official-distribution-rate	the page's Distribution Rate: latest distribution annualized over NAV or market price, with its own as-of date	Distribution Rate ... published on the official Franklin fund page ...
  indicated	updater estimate: latest Yahoo distribution x payments per year / price	indicated: ...

  official-other (a provider yield of unclear definition) is only the fallback for an unrecognised published label; computed-trailing-12m is not used by Franklin

• premiumDiscount (index row premiumDiscountValue, with premiumDiscountAsOfDate) - computed as (price / NAV - 1) * 100, rounded to two decimals, from the official NAV and a price of the same trading date: the closing Market Price of the page's own Share Prices block (one as-of date for both; the page defines it as the official closing price), else the Yahoo regular-session close of the NAV date. meta.json premiumDiscount.source is computed from market price / official NAV, same date and priceSource names the price used. When the dates differ or either figure is missing the value is null, never a mix of dates. The page's own "Avg. of Market Price vs. NAV at Close" matches this figure on the pages checked; the since-inception average is not used. Funds that hold foreign stocks can show more than 1% on days when their NAV was struck before the local close

• secYield - 30-day SEC yield when published; null otherwise

• returnsBasis - always a non-empty text describing how the returns are computed: official Franklin Templeton average annual total returns at market price (finder and product page), 3Y/5Y/10Y cumulative figures compounded from the official annualized values, not derived from Yahoo Finance

• performanceAsOf - ISO date (YYYY-MM-DD) the returns are as of: the "Average Annual Total Returns As of ..." date of the Franklin performance table (a month-end or quarter-end, not the NAV or holdings date), kept from the previous run when a page is unavailable; null when unknown or when the fund has no returns. The hub uses it to flag stale returns. YTD is refreshed daily by the issuer, so it can be newer than this date

Caveats:

• Official figures come from franklintempleton.com product pages; history, distributions and the derived yield come from Yahoo Finance market prices and are estimates, not official NAV data
• Unavailable values are published as missing, never as 0 (a missing weight, value or share count of an N-PORT holdings sheet is an empty cell); a bounded return, yield or AUM filter excludes funds that have no figure for it
• A fund younger than one year (calendar anniversary of its inception, inclusive, against performanceAsOf) publishes no 1-year or since-inception return, and no 3, 5 or 10-year one before those anniversaries
• Dates (U.S. 10/24/2026, month names such as Oct 24 2026 or 24 Oct 2026, ISO) are read as plain calendar text, never through the local time zone, so local runs match CI; a day that does not exist stays unread
• Each fund records its source and as-of metadata in meta.json
• Nothing is invented: a fund the sitemap lists but no page has been read for has null name, terValue and exchange, and an index row without meta.json has dataFile: null with a full all-null metrics object. terValue is the net expense ratio and meta.json expenseRatio.gross the gross one when published
• Funds are published whole: a fund is computed completely in memory, then written (page files, then meta.json, then stale page removal; every JSON file goes through a temp file and a rename). When the product page or the Yahoo chart worked for a fund before and fails now, the fund keeps its previous files and counts as failed, so a new return is never published next to stale prices. The workflow may therefore commit a partial run: every fund in it is either fully updated or fully kept. The updater exits 1 only when every selected fund failed
• A product page counts as fully loaded only when it carries all three anchors: identifiers/fees (Overview), Total Net Assets (Portfolio) and the Share Prices block (Pricing). The rendering proxy sometimes returns the page with the lazy sections collapsed (empty Performance, Portfolio, Pricing headings); the updater then asks the next proxy candidate once (never more than the existing three candidates, still one global proxy gate of at least 3.2 s) and otherwise uses the partial page. For a partial page every section that was published as official before and is missing from it (pricing: NAV and market price, from which the premium/discount is computed again; total net assets; SEC yield and the 12-month yield or Distribution Rate; month-end returns with performanceAsOf; official holdings unless a newer N-PORT filing exists) counts as a failed read: the previous block is kept as one unit with its as-of date, never replaced by Yahoo-derived values or null, and one [ kept ] line per fund names the sections kept. A page that loaded fully and lacks a field (no market price, no SEC yield) is an honest absence and publishes null (the market price then falls back to the Yahoo close, labelled as such). Returns are the exception: young funds have no returns table, so a loaded page cannot vouch for them and an absent returns section is always kept. A NAV that fails the plausibility check against the price is a failed read too and the published NAV stays
• Holdings carry their own asOfDate and source; when no fresh holdings are available the previous ones stay under their own date
• SEC N-PORT-P: the newest filing of the fund's own series is looked up on EDGAR every run (never a pinned accession; newest report period first, read as primary_doc.xml from the trust's own EDGAR folder), it is used only when it names that series id and it replaces the published holdings only when its report period is strictly newer than theirs (an unknown date on either side keeps them). The series comes from FRANKLIN_SERIES_MAP (series SEC's fund ticker table lacks), else from company_tickers_mf.json; a failed SEC ticker table request is asked once more later in the run. N-PORT identifiers are the position's own (CUSIP, else ISIN, else the filer's other code; a derivative never takes its reference security's), a ticker is set only for a U.S. equity traded in USD (the filer's own, else SEC's company name table) and never on two different securities
• Every request has a 45 s deadline covering headers and body (8 s for the direct franklintempleton.com attempt before the proxy fallback), retried per MAX_RETRIES. The SEC contact in SEC_UA is sent to sec.gov only and is redacted in the config printout
• Unbounded runs (MAX_FETCHES=0, and TICKERS runs) process the stalest fund first: funds without published data, then the oldest published as-of date (the latest of the price, NAV, holdings and performance dates in funds/<TICKER>/meta.json), ties alphabetical. The run stops taking new funds after 25 minutes (the first fund is always taken) and still writes the index; the skipped funds keep their published files and are first in line next time, so the tail of the list never starves. The log and the step summary report how many funds were refreshed and the oldest remaining as-of date. A bounded run (MAX_FETCHES > 0) keeps walking the alphabetical cursor
• NEW FUNDS: A, B is printed (and added to the GitHub step summary) when a fund is published for the first time, also when its first read failed and a later run succeeds; NEW FUNDS (not published yet): C names new catalog funds whose first read failed. A catalog-only row that the live catalog (the product sitemap and the seed list) no longer lists, for example an old mistyped TICKERS entry, is dropped
• franklintempleton.com is behind a WAF that may return 403 to bare fetch. The updater tries a direct fetch with a browser-like User-Agent first, then falls back to https://r.jina.ai/http://... (Jina AI rendering proxy) which returns Markdown. Both paths are parsed by the same parseFranklinCatalog / parseFranklinProductPage helpers. After ISSUER_DIRECT_DENIAL_LIMIT consecutive direct 403s, the updater uses the proxy only. Product pages skip the http:// proxy (it renders a redirect stub) and go through https://r.jina.ai/https://..., which is asked exactly once more after a 429 or a snapshot without the page content (it sometimes renders the page empty)
• HISTORY_RANGE limits the Yahoo history request window (max starts at period1=0, <N>y sends explicit period1/period2 for the last N years); a window keeps the older published history and distributions (adjusted closes rescaled by the overlapping days), and a window that does not join them keeps the fund as published
• Output layout:

api/franklin/
  index.json                 # catalog + month-end metrics
  update-state.json          # cursor for bounded runs
  funds/{TICKER}/
    meta.json                # per-fund identifiers, returns, metrics, manifest
    holdings/001.json        # paginated current holdings
    history/001.json         # paginated daily history

The UI loads index.json first, then lazily fetches meta.json and the paginated sheets for the selected watchlist.

The table matches scripts/update-data.config.json exactly.

TICKERS="FLIN FLGR FLEE" bun ./scripts/update-data.ts
MAX_FETCHES=10 REQUEST_SLEEP=0 bun ./scripts/update-data.ts
AUM=large TER=:0.40 bun ./scripts/update-data.ts

The browser app is built with Parcel and Tailwind v4: src/index.html carries the markup, src/index.css the styles and src/main.tsx is the TypeScript application. bun run build writes dist/, .github/workflows/github-pages.yml deploys it to GitHub Pages. No tsconfig.json needed, Bun runs TypeScript out of the box.

Verification before every publish:

bun install --frozen-lockfile
bun test
bun build --target=bun scripts/update-data.ts --outfile=/dev/null
git diff --check

bun test runs the single scripts/update-data.test.ts: parsers, the config resolver (precedence, validation, defaults), parity of the config file, --help and the controls table, the workflow shape, the README structure and the browser contract ids of src/index.html and src/main.tsx

MIT — same as all sibling ETF repositories.

Franklin Templeton® and Franklin® and the fund names/tickers referenced here are trademarks of Franklin Resources, Inc. This is an independent, unofficial tool; it is not affiliated with, endorsed by, or sponsored by Franklin Templeton or Franklin Resources, Inc. All data is reproduced from Franklin Templeton's own public fund pages, public SEC EDGAR filings and Yahoo Finance for research purposes. All other trademarks, including index names, are the property of their respective owners.