[daggerok]

What changed

The shared updater blocks of .claude/tools/etfs/etf-std (byte-identical, applied with apply-block.ts, the current nport-reader template with the EDGAR state-suffix fix), wired like Schwab #75

• nport-reader (ETF-H2): N-PORT holdings come from readNportHoldings with one SEC state per run (resetSecState()), the series id from SEC's company_tickers_mf.json (parseSecFundTickers) and tickers from nportTickers with SEC's company names (parseSecCompanyNames). The sheet keeps the WisdomTree headers (nportCell and nportIdentifier)
  • the old copy read the .txt submission through the series Atom feed, took only the first feed entry, built the archive folder from the accession prefix (the filer agent), accepted a filing whose series id was missing, replaced the published sheet on >= its date, wrote 0 for a missing weight or value, published CUSIP 000000000 as the identifier and gave every row a name-matched ticker
  • now: newest report period first from the WisdomTree Trust's own folder (primary_doc.xml), only the fund's own series, strictly newer than the published N-PORT sheet, missing numbers empty, the position's own identifier (ISIN or other id instead of 000000000), a ticker only for a U.S. equity in USD and never on two securities
  • WT-2 ticker hygiene: bonds, MBS/ABS, FX forwards and other derivatives no longer take an issuer or counterparty ticker (6,365 published rows have one today, 6,233 of them debt), Tokyo or other non-USD listings no longer take an OTC ADR ticker (DXJ FANUF)
  • freshness choice: the published N-PORT sheet is what a filing is compared with. A published product-page top-N list (WSPC, BTCW) is no full schedule: the first N-PORT filing replaces it whatever its date (the issuer dates that list every day, so a strict date check would block N-PORT forever)
  • product-page top holdings use the same SEC name key (nportNameKey, EDGAR suffixes like /DE/ dropped); the old normalizer and company map are gone
• holdings-guard (Q24): holdingsWeightCount(publishedHeaders, publishedRows) as the 5th argument at both guard calls, so even the relaxed WSPC cannot replace numeric published weights with an all-blank sheet; HOLDINGS_GUARD_RELAX is exported for the test
• windowed-chart (Q25): a HISTORY_RANGE window is merged with the published history and older distributions (the published [Ex-Date, ..., Total Distribution] rows are mapped to [date, amount]); a window that does not join keeps the published history (the old copy shrank the history to the window)
• calendar-dates: toIsoDate is calendarIso(raw) ?? raw (the old copy went through Date.parse, a UTC suffix kept month names on the right day, but a timestamp with an offset such as 2026-10-24T22:00:00-07:00 became the 25th)
• young-fund rule: withoutYoungReturns with { yr1, cagr3y, cagr5y, cagr10y, siAnn } on the official month-end and quarter-end tables (main, NAV and index rows) and on the final returns. The old copy nulled only since inception, below 365 days. Now the calendar anniversary, inclusive, every horizon; the age counts from the earliest known start (catalog inception, Yahoo first trade, first history day) so GCC (joined WisdomTree in 2020, trades since 2008) keeps its official 10-year return. The "derived since inception only from a series that starts at inception" part of fix(metrics): since-inception never annualized below one year or from a history window #86 stays (annualizedSince)
• fund-selection (Q27, Q28): assertKnownTickers after the catalog load (live catalog plus published funds; the old check also accepted stale catalog-only rows), previous rows enter the catalog only with dataFile, the index map drops stale catalog-only rows (withoutStalePlaceholders). The pre-run NEW FUNDS line (every live ticker missing from the index, also failed reads) is replaced by newFundsLines(newFundsReport(previous, final)) after the index is written, in the log and as ### NEW FUNDS in the step summary. DROPPED FUNDS stays (now only funds with published data)
• README: N-PORT source and identity, freshness, identifiers and tickers, empty missing values, young-fund rule, calendar dates, NEW FUNDS semantics, TICKERS and HISTORY_RANGE rows

Values that change on the next runs

• N-PORT sheets: every published fund keeps its sheet until SEC has a strictly newer report period (today all 92 N-PORT funds are as of 2026-05-31 or 2026-06-30, the same as their newest filing). With the next filing per fund: empty instead of 0 (102 rows), own identifiers instead of 000000000 (17,576 rows), no ticker on 6,365 non-equity rows and on non-USD equities, weights at most 8 decimals, Shares Held without trailing zeros, Coupon/Maturity on every row of a debt sheet
• WSPC and BTCW keep the top-N list until a first N-PORT filing exists (WSPC: none yet, BTCW: not in SEC's fund table); WSPC gains LHX for L3Harris from the new name key
• Young-fund rule: WQTM loses its official 1-year (1.21) and GTR its official 5-year (4.22, 4.98 years old at the 2026-09-30 table); 8 funds younger than a year (GDT NTSD WAMA WDIG WDRN WIMA WSPC WTLS) lose since inception, which fix(metrics): since-inception never annualized below one year or from a history window #86 already does on their next run

How it was checked

• bun test: 57 pass / 400 expects on main -> 65 pass / 428 expects, exit 0; the same with the config controls exported, exit 0
• each new test fails with its wiring reverted (dates, both age-rule applications, earliest start, N-PORT names, top-N freshness, window merge, guard weights, Q27 assert and stale rows, Q28 lines, published date)
• bun build --target=bun scripts/update-data.ts --outfile=/dev/null, bun build --target=bun src/main.tsx --outfile=/dev/null, git diff --check
• check-readme.ts, check-scripts.ts, apply-block.ts --check holdings-guard,nport-reader,windowed-chart,fund-selection: ok, all five blocks identical
• tc/check.sh: 64 -> 52 error lines (all pre-existing [ide] kinds, no new kind)
• real runs (REQUEST_SLEEP=0, VERBOSE=true), then git checkout -- api/ && git clean -fd api/:
  • TICKERS="DGRW USFR": both unchanged, both log N-PORT ... is not newer than the published holdings
  • N-PORT path with api/wisdomtree/funds/{DXJ,SHAG} removed: DXJ rebuilt from accession 0000940400-26-036274 (2026-06-30, 439 rows, weight sum 104.56, 0 000000000), SHAG from 0000940400-26-028881 (2026-05-31, 1,190 rows, weight sum 99.52); rows with a ticker on non-equity or debt rows 687 -> 0, all equity tickers of DXJ (Tokyo listings) dropped; USFR in the same run not newer
  • N-PORT path with funds/DGRW removed: 198 rows, weight sum 99.97, 197 U.S. tickers (171 on the published first page before), 0 on non-equity rows
  • HISTORY_RANGE=1y TICKERS="DGRW USFR": zero diff, history rows 3,377 and 3,200 kept
  • TICKERS="GCC WQTM DGRW": GCC and DGRW unchanged, WQTM 1-year and since inception nulled