Data sources

All default paths are free and require no API key. Vite middleware proxies (/proxy/yahoo, /proxy/fred, /proxy/coingecko) bypass browser CORS and briefly cache upstream responses. Yahoo is fetched with curl server-side (Node fetch is often HTTP 429’d).

Indicator Endpoint Auth Notes
BTC spot https://api.coinbase.com/v2/prices/BTC-USD/spot None CORS open; polled ~7s
BTC 24h stats https://api.exchange.coinbase.com/products/BTC-USD/stats None open / high / low / last
Circulating supply https://blockchain.info/q/totalbc None Satoshis → BTC; remaining = 21_000_000 − circulating
Daily issuance Computed — 3.125 BTC × ~144 blocks/day ≈ 450 BTC/day (post-2024 halving)
200-week MA https://api.kraken.com/0/public/OHLC?pair=XBTUSD&interval=10080 None Weekly closes; see calculation below. Published on its own; a hung ETF fetch does not keep this card on Loading.
ETF holdings + flows GET /api/etf-flows?asset=btc&days=21 (same-origin). Upstream: https://xoomar.com/api/markets/etf-flows?asset=btc&days=21, then issuer files if that GET sends no body. None See ETF section below.
Corporate treasuries Holdings: /proxy/coingecko/api/v3/companies/public_treasury/bitcoin (fallback CEBE companies + XOOMAR SEC). 7-day net flow: https://cebetracker.io/api/v1/deals.json — sum signed btc_amount on non-superseded btc_purchase / btc_sale rows with event_date in the last 7 days. USD = Σ(amount × deal btc_price) when present. None Holdings list is CoinGecko-broad when available. 7d flow coverage is CEBE’s tracked fleet only (filing-verified deals; sparse when no 8-Ks). Footer states deal count / tickers / fleet.
Performance tracker /proxy/yahoo/v8/finance/chart/{symbol}?interval=1d&range=10y for Bitcoin (BTC-USD), Gold (GC=F), S&P 500 (^GSPC) as first-class rows; Mag7 (AAPL MSFT GOOGL AMZN NVDA META TSLA) as a separate equal-weight group + individuals None Timeframes: 1M, 3M, 6M, 12M, YTD, 2Y…10Y. Return = last price vs closest close on/near period start. UI: chip selector + benchmark cards.
DXY / USD Primary: /proxy/yahoo/v8/finance/chart/DX-Y.NYB. Fallback: FRED CSV DTWEXBGS (Trade Weighted U.S. Dollar Index: Broad) None Card footer names which series is shown
Gold Primary: /proxy/yahoo/v8/finance/chart/GC=F. Fallback: https://api.gold-api.com/price/XAU None Fallback is live spot (day change may show 0 without prior close)
Fed balance sheet /proxy/fred/graph/fredgraph.csv?id=WALCL None Millions USD → shown as $B; WoW change
U.S. Treasury yields /api/yields (server-side, edge-cached s-maxage=60, client polls every 60s). Per tenor: CNBC/Tradeweb quote cache (US2Y/US5Y/US10Y/US30Y, real-time) → Yahoo ^FVX / ^TNX / ^TYX (already in percent; no 2Y symbol) → Treasury.gov daily par yield curve CSV (official EOD). Day/week change vs official Treasury closes. Last resort: deploy-time FRED snapshot (public/data/fred/DGS*.csv), flagged stale. None FRED is blocked from Vercel egress, so it is no longer the primary source. Footer shows actual source + as-of date. Never invented: unavailable tenor = null.
Investment returns Same Yahoo 10y daily chart proxy as Performance for Mag7, BTC-USD, GC=F, ^GSPC, plus BTC treasury names (MSTR, MARA, RIOT, COIN) None Horizons 3M…10Y. Lump-sum = (invest ÷ past close) × latest. DCA = fixed $ every week/month from period start, shares summed, valued at latest. Amount/freq/selected assets in localStorage. Horizon chips sticky; results ranked best→worst by % return. Live marks ~7s (history ~10 min).
Digital credit Pref quotes: `/proxy/yahoo/v8/finance/chart/{STRC SATA}?interval=1m&range=1d(~7s). Issuer BTC: CEBEcompanies.json(MSTR, ASST). Credit-line BTC: CEBEdeals.jsonfiltered byinstrument_id (STRC, ASST_SATA`) — SATA ATM raises + funding-group–paired purchases when present; STRC currently repurchase-mode (0 STRC ATM→BTC in recent window). BTC spot: live exchange median. None
Balance sheet health /api/sec-financials?ticker= → server/secFinancials.mjs (SEC EDGAR companyconcept XBRL: freshest cash + same-period marketable securities, debt, preferred, pref dividends, interest, cover-page shares) merged with server/treasuryDisclosures.mjs (latest 8-K/press-release overrides: Strategy 8-K USD Reserve + USD Cash; Strive 8-K + strive.com/api/treasury for ASST/SATA; curated yen constants for Metaplanet (not an SEC filer, secondary-sourced cash/interest); XXI/RIOT from 10-Qs). Equity quotes: Yahoo chart. BTC holdings: CEBE/CoinGecko treasuries feed (null-symbol safe). BTC spot: Coinbase. None Mag 7 = cash+marketable securities vs debt, plus market cap. MSTR/ASST pinned + top treasuries by holdings: Market cap = live price × filing shares; mNAV / Amplification / Net Reserve follow Strategy's published definitions (see “Strategy standard & company reconciliation” below; the old EV ÷ BTC multiple is kept as “EV ÷ BTC”); BTC breakeven ARR = (annual pref dividends + interest) ÷ BTC assets; Duration = (BTC assets + USD reserve) ÷ annual (pref dividends + interest), "—" if no obligation. Every unpublished input → "—" / partial, never invented; as-of dates and sources shown per tile.

Spot ETF holdings and flows

The browser does not wait on a raw cross-origin XOOMAR call. fetchEtfFlows hits /api/etf-flows (Vercel edge in production, the Vite data proxy in npm run dev / preview) with a 14s AbortSignal deadline. If that route itself errors before the deadline, the client tries XOOMAR directly for whatever time is left (capped at 8s). A failure sets the card to Offline. It does not stay on Loading, and it does not block the 200-week MA or the other macro cards — each macro source publishes as soon as it settles.

/api/etf-flows does two things:

  1. XOOMAR (7s server timeout). When the JSON arrives, it is passed through unchanged. Rows include holdings (BTC) and aumUsd. The footer still says XOOMAR reporting.
  2. Issuer files run in parallel with that attempt. If XOOMAR already lists every target ticker, its body is returned as-is. If it lists only some, those rows are kept and the missing tickers are appended (source: "xoomar+issuer"). If it does not answer, the issuer payload is the whole response (source: "issuer-fallback"). IBIT’s multi-day download starts only when XOOMAR did not already return IBIT.
    • IBIT — iShares public get-product-data holdings for portfolio 333011, one request per recent weekday. USD flow = change in coins × that day’s implied price (market value ÷ coins). BTC flow = the same coin change.
    • BITB — latest coins and AUM from the Bitwise page (bitbetf.com __NEXT_DATA__).
    • ARKB — ARK holdings CSV (coins and market value).
    • HODL — VanEck fund-dataset JSON linked from the HODL product page (coins and market value).
    • EZBC — Franklin Templeton public PDS Portfolio.dailyholdings for the Bitcoin line.
    • FBTC, GBTC, BTC (Grayscale Mini), and a coin change for any one-day file above — btcetfdata.com/v1/current.json, which republishes the issuer’s latest coins and the change versus the prior file. A change is applied to an official one-day file only when the date and coin count match. Canadian BTCC is not included.
    • Nasdaq quote-summary AUM (value × 1,000; the field is labeled in thousands) is attached only when AUM ÷ coins is within 8% of the median issuer-file BTC price. A divergent print is dropped.

USD for a fund with no market value in its own file is the coin change × the nearest issuer-file BTC price (same day, else within four days). A target with no holdings source is omitted. The footer lists included tickers and no public holdings file: for the rest (today that is usually BTCO, BTCW, BRRR — Invesco, WisdomTree, and CoinShares do not publish a fetchable holdings file from this host). Each fund row shows that fund’s latest flow in USD and BTC; the card’s aggregate daily / 7-day nets are unchanged. Click a row for the daily history.

The fallback is shaped like a XOOMAR payload so totals and the per-fund rows stay on the same parser. Successful responses are cached about 5 minutes in memory and on the edge (s-maxage=300).

Every macro HTTP call (ETF, 200-week MA, treasuries, DXY, gold, Fed balance sheet, Treasury yields) uses a 14s abort. refreshMacros applies each result on its own instead of one Promise.all that cannot resolve while any fetch is still open.

200-week moving average

  1. Fetch Kraken public weekly OHLC for XBTUSD (interval=10080 minutes = 1 week). No API key; CORS enabled.
  2. Use candle close prices (field index 4). Kraken returns hundreds of weeks of history.
  3. Take the most recent 200 weekly closes.
  4. 200WMA = average(those 200 closes).
  5. Compare to the live Coinbase spot price:
    • deltaUsd = price − 200WMA
    • deltaPct = deltaUsd / 200WMA × 100
  6. A compact sparkline shows the rolling 200WMA over the last ~52 weeks.

Refreshed with the macro poll (~5 minutes); the displayed gap vs spot updates on every BTC price poll (~7s).

Optional keys (not required)

Env var Purpose
VITE_FRED_API_KEY Reserved if you later switch WALCL / DGS* to the FRED JSON API. CSV endpoints used today need no key.
VITE_SOSO_API_KEY Reserved for SoSoValue OpenAPI if you want full-universe ETF history beyond XOOMAR’s free feed.

If a series fails, its card shows an Offline badge (not silent zeros). Demo-data badges are only for a future keyed fallback — nothing ships that way by default.

Attribution

Full spot-ETF roster (added Sep 2026)

/api/etf-flows now also returns roster.funds (built in server/etfRoster.mjs, cached with the rest for 5 min) so the card lists every US spot Bitcoin ETF: IBIT, FBTC, GBTC, BTC (Grayscale Mini), BITB, ARKB, HODL, MSBT (Morgan Stanley, launched Apr 8 2026), BRRR, EZBC, BTCO, BTCW, plus DEFI (Hashdex, liquidated Aug 2026 — listed with status, holds 0 BTC, not counted).

Field Source
BTC held issuer-file pipeline / btcetfdata.com (freshest), else Bitbo per-fund pages
AUM SoSoValue net assets
Daily USD flow Farside ($m, one decimal) + SoSoValue latest session
Daily BTC flow SoSoValue latest session, btcetfdata 1-day change, Bitbo dated changes (only when consistent with that day's USD flow)
DEFI status SEC 8-K (Aug 24 2026), hashdex-etfs.com/defi

Anything no source publishes stays —; nothing is estimated. The client (src/api/etfRoster.ts) merges roster + XOOMAR/issuer rows and totals only funds with real holdings.

Live cadence & 1D / 1W (added 2026-09-30)

Spot ETF flows (rules, 2026-09-30)

Performance: bitcoin-treasury stocks and the Bitcoin vs MSTR card

BTC period bases (1D / 1W / 1M / YTD) — src/api/btcBases.ts

BTC's live price is Coinbase spot, so its bases are Coinbase Exchange candles too (public /products/BTC-USD/candles, finest granularity first: 1m → 5m → 1h; Yahoo BTC-USD 1m/1h only if Coinbase returns nothing; "—" if everything fails). "Price at T" = open of the first candle at or after T (if no trade yet, close of the last candle before T).

Treasury leaderboard and Purchase tracker (added 2026-10-02)

Served by GET /api/treasury-board (server/treasuryBoard.mjs, edge-cached s-maxage=300, 30 s when a source was throttled). Nothing is estimated: a field with no real source is null and shown as “—”.

Per-field sources

Field Source (priority order) Notes
BTC held 1) company 8-K weekly table (Strategy, Strive; parsed directly from SEC EDGAR) or RNS (The Smarter Web Company: Quarterly Investor Update / Bitcoin Purchase, via Investegate) → 2) CEBE Tracker companies.json (dated, filing-anchored) → 3) SEC 10-Q/10-K XBRL CryptoAssetNumberOfUnits (dated) → 4) CoinGecko /companies/public_treasury/bitcoin (undated) As-of date + source shown per row. A CEBE/XBRL value more than 5× away from CoinGecko is rejected and listed as “also reported”. XBRL unit series with >50× jumps (typos, e.g. CLSK) are cleaned.
BTC value holdings × live BTC price (exchange median, src/api/liveQuotes.ts) live, re-marked every ~6 s
Share price Yahoo Finance quote (liveQuotes.ts, shared 5 s cache) Metaplanet: MTPLF (USD OTC) for 3350.T; Smarter Web: TSWCF (USD OTCQB; LSE: SWC trades in GBX). No FX is applied anywhere in the leaderboard.
Shares, cash, debt, preferred SEC EDGAR XBRL via /api/sec-financials (existing); Metaplanet via the disclosure overrides already in that endpoint; Smarter Web (not an SEC filer): RNS figures via CEBE Tracker companies.json (GBP facility converted to USD by CEBE at the RNS-date rate) inputs older than 150 days (shares) / 200 days (cash, debt) are not used → mNAV “—”
Market cap live price × cover-page shares
mNAV (market cap + debt + preferred − cash) ÷ BTC value (balanceMetrics.mnav) “n/m” when BTC < 5% of enterprise value
Premium / discount to BTC NAV (mNAV − 1) × 100 descriptive
Avg cost / BTC, unrealized P/L Strategy 8-K aggregate purchase price and average price; SEC XBRL CryptoAssetCost ÷ CryptoAssetNumberOfUnits at the same balance-sheet date (covers all crypto assets); CoinGecko total_entry_value_usd (company disclosures aggregated by CoinGecko, undated); Smarter Web RNS “Net Average Purchase Price” in GBP as published P/L = (costBtc × spot − cost) ÷ cost, using the BTC the cost covers. Smarter Web: GBP-denominated, so no USD P/L is shown.
BTC per 1,000 shares BTC ÷ shares × 1000 (same share count as market cap) basic cover-page shares
BTC yield ((BTC ÷ shares)ₜ₂ ÷ (BTC ÷ shares)ₜ₁ − 1) × 100 only with two dated points; Strive: 8-K “Assumed Fully Diluted Shares”, weekly; others: SEC XBRL balance-sheet dates with units and basic shares; a pair whose later date is >215 days old is dropped
mNAV history Yahoo daily closes (equity and BTC-USD), 8-K weekly holdings (step function from each filing date) Point-in-time: equity close, BTC close, BTC held. Held constant at the latest filed values: shares, debt, preferred, cash. Built only for Strategy (since the first 8-K in the scan window) and Strive (since 2026-06-30, after the merger/convert retirement); other companies show “—”.

Purchase tracker

Attribution and etiquette

Performance tracker

Treasury leaderboard toggles, Digital credit dashboard, Purchase chart (added 2026-10-02)

Treasury leaderboard: definitions and bases

Digital credit dashboard (/api/credit-rates, /api/credit-volatility, /api/yields)

Purchase tracker chart

Bitcoin dashboard (#bitcoin, home “Bitcoin” section)

Removed per Luka’s request: mempool fees, difficulty / retarget, all-time low (and the difficulty mentions in the hashrate tile and halving estimate).

Cross-check rule (server/crossCheck.mjs, used on the server and in the browser): every metric collects one observation per independent source. ≥3 sources → the median is shown; 2 sources → the primary is shown and a gap beyond tolerance is flagged; 1 source → labelled “single source / not cross-checked”. A source farther from the median than the tolerance is marked an outlier and the tile carries “⚠ sources disagree”. Every tile has a <details> “sources” disclosure listing each source’s value and as-of time. Tolerances (% of median): price 0.25, supply 0.1, market cap 1, 24h volume 25, 30D volume 20, hashrate 5, chain height 0.0005, ATH 1, dominance 5, realized vol 5.

Edge proxy /api/btc-metrics?part=market|dominance|network|sentiment|history (server/btcMetrics.mjs, api/btc-metrics.ts); s-maxage market 20 s, dominance 45 s, network 45 s, sentiment 90 s, history 240 s (+ stale-while-revalidate).

Metric Sources (cross-check) Method
Price (live) Coinbase, Kraken, Bitstamp WebSockets (REST poll every 2 s per venue if a socket is silent); CoinGecko / CoinPaprika / CoinLore / Blockchair / blockchain.info as slower cross-checks median of venues that ticked in the last 20 s; UI throttled to 4 updates/s; Binance.com is geo-blocked (HTTP 451)
Market cap live price × median supply; CoinGecko, CoinPaprika, Blockchair, CoinLore caps as cross-checks tol 1 %
Supply blockchain.info, Blockchair, CoinGecko, CoinPaprika + the protocol emission schedule at the chain height (CoinLore excluded: its supply lags ~120k BTC) median
ATH / drawdown CoinGecko, CoinPaprika, highest daily high in Kraken (720 d) and Bitstamp (1000 d) history median; drawdown = price ÷ ATH − 1, recomputed per tick; a live price above the ATH becomes the new ATH
24h / 7d / 30d / 1y change CoinGecko % → reference price (= price at fetch ÷ (1 + chg)) live price ÷ reference − 1 on every tick
200-week MA, % vs MA Kraken weekly OHLC only (no other free source has 200 weekly bars) single source, labelled
Avg volume 30D CoinGecko market_chart (primary), CoinPaprika history; Kraken + Bitstamp + Coinbase BTC volume shown as a lower bound (not averaged) 30 complete UTC days; BTC = USD volume ÷ that day’s price; tol 20 % (they differ ~24 % → flagged)
24h volume CoinGecko (primary), CoinPaprika, CoinLore (single exchanges excluded: far smaller) tol 25 %
Realized vol 30D / 90D Kraken, Bitstamp, Coinbase daily closes per-venue st-dev of log returns × √365, median
Fear & Greed alternative.me (single source; CoinMarketCap’s index has no key-free API) UTC-day as-of
Dominance CoinGecko global (primary); CoinPaprika, CoinLore, Blockchair range; method variants from CoinGecko /coins/markets see below
Hashrate mempool.space currentHashrate (recent blocks), blockchain.info hash_rate (24 h), Blockchair hashrate_24h (24 h) median; windows differ so a 965 vs ~871 EH/s gap is flagged
Chain height / next halving mempool.space, blockchain.info, Blockchair median; halving = next multiple of 210,000; date = blocks × 10 min

Dominance methods. Standard (CoinGecko global, all coins incl. stablecoins) ≈ 58.6 %; providers 56.4–59.5 %. Excluding stablecoins from the top-125 coins gives ≈ 64.7 % with CoinGecko market caps (≈ 63.6 % with CoinPaprika’s, stablecoins matched by a symbol list — both are shown and cross-checked), the closest reproducible figure to the ~64.90 % Luka expects (TradingView’s BTC.D uses a different, non-public coin universe; TradingView and CoinMarketCap have no key-free public feed and are not scraped, so 64.90 % itself cannot be verified). The dashboard shows the CoinGecko primary + range and the ex-stablecoin variant with its method label.

Rate limits. CoinGecko’s free tier answers HTTP 429 to Vercel’s shared egress IPs now and then. The server keeps the last good heavy CoinGecko/CoinPaprika answers in memory for 15 min, and the browser carries a source’s last value forward with its original as-of (carryObs, shown in the sources list) for ≤15 min (ATH: indefinitely; history series: until replaced). A stale CoinGecko spot price (>2 min) is never used as a cross-check. If CoinGecko is down, % changes fall back to CoinPaprika (24h / 7d only) and ex-stablecoin dominance to the CoinPaprika-computed variant, each labelled.

Live behaviour (src/api/livePrice.ts, src/hooks/useBtcMetrics.ts, src/api/btcModel.ts): price via WebSocket; market cap, drawdown, % changes and % vs 200W MA are re-derived on every tick. Browser refresh: market 30 s, dominance 60 s, network 60 s, sentiment 120 s, history 300 s, 200W MA 30 min. LIVE badge → STALE when the last price tick is older than 15 s (or the tab is hidden); a slower feed is “delayed” after 3× its interval. Numbers flash green/red for ~0.7 s on change (no re-mount, so no flicker). Coinbase’s REST API is blocked from Vercel’s edge, so Coinbase is read directly in the browser.

Strategy metrics (src/api/strategyMetrics.ts, StrategyMetrics.tsx)

Definitions: Strategy “New Metrics” (July 2026) and Q2 2026 deck. Inputs: Strategy weekly 8-K (USD Reserve, USD Cash, ATM table, repurchase programs: parseStrategyUsdReserve, parseStrategyAtm), investor briefing 2026-09-13 (annual interest + dividends $1.65B), 10-Q debt, CEBE preferred notional.

BTC / Gold ratio (#btc-gold, /btc-gold, src/api/btcGold.ts) — added 2026-10-03

News channel (/api/news, server/news.mjs, home News rail / News page feed, /news) — added 2026-10-03

Hard assets vs dollar (#hard-assets, /hard-assets, src/api/hardAssets.ts, src/api/assetReturns.ts) — redesigned 2026-10-04; real-estate rows rebuilt 2026-10-05 (FHFA ZIP-5 Miami Beach, Zillow Palm Beach island, FHFA county Manhattan)

One headline chart, “What $100 became”: for a holding period of 1–10, 15, 20, 30 … 100 years, the total and annualised return of bitcoin, the S&P 500, gold, Miami Beach, Palm Beach island, Manhattan, US homes and the US dollar's purchasing power. Below it: a dollar strip (CPI and M2), a “what it takes to buy prime property today” block, and the documented sales and same-property resales behind a disclosure.

SEO, share, embeds, alerts — added 2026-10-03

Live ticker strip and one-number rule

The strip at the top of every page (Bitcoin, S&P 500, Gold, IBIT, MSTR, Magnificent 7) reads the same layers as everything else: BTC = live exchange median (24h % = price ÷ price 24h ago, src/api/btc24h.ts, also used by the hero, Explore row and Bitcoin sheet); gold = COMEX future via the BTC/gold layer, % vs the GC=F previous close; equities/index = Yahoo interval=1m&range=1d through liveQuotes.ts (10 s poll, 5 s cache), % vs previous close. Spot-ETF and treasury dollar values are holdings × the live BTC price. Missing values show "—".

Strategy standard & company reconciliation (added 2026-10-04; UI simplified 2026-10-04 PM)

What the UI shows (Balance sheet → “Key metrics” section, src/components/KeyMetricsList.tsx, data from src/api/keyMetrics.ts): a plain list per company — metric name, value, one short definition — with a single source line at the top (“Source: · as of … · definitions: strategy.com/notes”, or “Definitions: strategy.com/notes · inputs: latest company filings, live BTC and equity quotes” when a company publishes no metrics). Values are exactly what strategyStandard() computes (headline mNAV = close-paired when the stock is closed, live-BTC figure in its definition). Formulas sit in each row’s definition text/tooltip. No comparison table or “ours vs published” wording is shown in the UI. The reconciliation against company-published values (reconcile(), reconRows(), scripts/strategy-standard.test.ts) is kept as internal checks/tests only. The old Overview tiles for mNAV / Amplification / Net Reserve / EV ÷ BTC / Net leverage / BTC Rating and the MSTR “Strategy metrics” duplicates were removed (MSTR’s remaining extras are in the “Strategy detail” section).

Luka's request: follow each company's OWN published methodology; apply Strategy's definitions to every treasury; never invent numbers. Code: src/api/strategyStandard.ts (pure, tested in scripts/strategy-standard.test.ts), server/publishedMetrics.mjs (company-published values), src/components/CompanyRecon.tsx (the per-company “Reconciliation vs company-published” section).

Strategy definitions used (source: strategy.com/notes “New & Updated Metrics”, effective 2026-07-23)

Company-published values (“Reconciliation vs company-published”)

Company Where read How
Strategy https://api.strategy.com/btc/bitcoinKpis and /mstrKpiData (the JSON behind strategy.com; public, no auth, CORS *), definitions at strategy.com/notes fetched live in /api/sec-financials?ticker=MSTR (edge-cached ≤15 min); dated snapshot (2026-10-04) if unreachable
Metaplanet metaplanet.jp/en/analytics, whose data feed is https://data.strategytracker.com/latest.json → 3350_T.v<version>.json (processedMetrics) live; snapshot fallback. Provides cash, debt, MERCURY preferred, basic + diluted shares, EV mNAV, “Amplification Ratio”, sats/share
Strive strive.com/treasury, feed https://strive.com/treasury/api/dashboard/calculated live; snapshot fallback. Provides EV mNAV (“EV / Treasury Asset Value”), assumed diluted shares, sats/share
Twenty One 10-Q Q2 2026 BPS 12,547 sats snapshot; no mNAV / amplification published
MARA IR / 10-Q publishes holdings only; no mNAV, amplification or reserve metric

Findings recorded 2026-10-04

Extended-hours equity quotes (pre-market · after-hours · overnight) — added 2026-10-04

Sheet navigation (every full-screen sheet) — added 2026-10-04