63 verified issues held: Do you want them posted now or later?

[synfonia-llc]

Hi @dgunning,

Since early October, AI-assisted reviews of EdgarTools have turned up more defects than the ones I've filed. I've held them back during the 6.0 window so you can stay focused on the release, and because several touch code that is still changing. My plan was to wait until all the breaking changes are in, re-run every one on 6.0.0, drop what no longer reproduces, and file the rest one issue per bug through the templates.

Where they stand:

• 63 findings, each reproduced on main in mid-October. Where a value is involved, it was checked against the filing itself. Each was searched against open and closed issues for duplicates.
• 33 are data quality, 1 is performance, and 29 are bugs, documentation errors and a few feature requests.
• They cluster in statement and XBRL parsing, the Financials getters, 8-K earnings releases, the AI skill docs, currency conversion, the MCP tools, funds and BDCs, and IFRS 20-F filings.
• All 33 data-quality findings also reproduce on 5.62.0.

Two questions:

1. Now or later? Would you like these issues filed now, or after 6.0.0 as planned?
2. If now, which branch for fixes? For a fix that applies to both 5.62.x and main, would you want two PRs, one against 5.62.x and one against main, or a single PR against main that you carry back to 5.62.x yourself?

Thanks for the quick turnaround on the last batch.

[synfonia-llc]

Hi @dgunning,

Following up on the question above: the held findings were re-verified on today's main (2323d732) and on the v5.62.0 tag, and filed one issue per defect through the templates, 131 in all. Each body shows the output of both builds, so a defect fixed in 6.0 shows that difference; nothing was filed that no longer reproduces on main, and findings that your recent fixes resolved (#1443, #1445, #1468, #1469, #1483, #1581, #1582, #1614) were closed on our side instead.

They group as follows (the grouping is by subsystem; the labels are the templates'):

Data quality (54)

• Currency
  • [DATA] CurrencyConverter ignores the currency and hedge axes: GSK's average rate comes from EUR hedge and JPY rows, not US$/£ #1650
  • [DATA] CurrencyConverter inverts a USD-per-GBP rate: a 1.31 average gives GBP 100 = USD 76.34, not USD 131 #1651
  • [DATA] CurrencyConverter returns no closing rates for Novo and GSK: a NaN period_end skips the period_instant fallback #1652
  • [DATA] CurrencyConverter returns no exchange rates for SAP: its divide-unit USD/EUR rate facts fail the unit-ID filter #1653
  • [DATA] Non-USD cash-flow statement opening and closing cash show $ despite correct amounts #1655
  • [DATA] Statement.to_dataframe() gives every row of a concept its first occurrence's display sign #1657
• Entity, filings and search
  • [DATA] EntityFacts.time_series("Assets") returns values of other concepts: a bare concept name is matched by substring #1658
• Financials getters
  • [DATA] Williams' Q2 2026 10-Q: income_statement() selects the Transco co-registrant role, so get_revenue() and get_net_income() return Transco's figures #1648
  • [DATA] Financials.get_stockholders_equity() returns the liabilities-and-equity total or None instead of stockholders' equity #1659
  • [DATA] Financials.get_capital_expenditures() returns accrued, unpaid capital expenditures as capex when the cash capex tag is not one of its two mapped concepts #1660
  • [DATA] Financials.get_current_assets() and get_current_liabilities() return a narrower subtotal, or None, instead of the filed current totals because they never look up the concept #1661
  • [DATA] Financials.get_financial_metrics() returns quarter income figures beside year-to-date cash flows, with no period on any value, on Q2 and Q3 10-Qs whose cash flow is year-to-date only #1662
  • [DATA] Financials.get_net_income() returns income from continuing operations as net income when its concept lookup misses #1663
  • [DATA] Financials.get_operating_income() returns pretax income from continuing operations as operating income when the filer reports no operating income #1664
  • [DATA] Financials.get_revenue() returns None for five filings whose income statement total is us-gaap:RegulatedAndUnregulatedOperatingRevenue (NextEra, Xcel, Spire, Duke Energy, American Water) #1665
  • [DATA] Financials.get_revenue() returns a bank's deposit service-charge line as total revenue when the statement has no total row #1666
  • [DATA] Financials.get_total_assets() returns the current assets subtotal when the total line is not labeled Total assets #1667
  • [DATA] get_shares_outstanding_basic() and get_shares_outstanding_diluted() return None when the income statement tags WeightedAverageNumberOfShareOutstandingBasicAndDiluted #1668
  • [DATA] Financials.get_net_income() returns ProfitLoss before noncontrolling interests when the statement has no NetIncomeLoss row (Caterpillar FY2025 10-K: 8,882M, filed 8,884M) #1688
• Funds and BDCs
  • [DATA] PortfolioInvestments reports 'SCSP', 'Middle' and a company name as industries for ARCC #1669
• IFRS and 20-F
  • [DATA] Standard views of four IFRS 20-F filings keep note, subsidiary and segment rows #1670
  • [DATA] Cash-flow getters return None for some IFRS 20-F filers (Shell, BP, TSMC, Toyota) (affects 6.x-rc) #1785
• MCP and AI skills
  • [DATA] MCP comparison and EntityFacts AI context derive YoY growth from unqualified fact rows #1671
• Other
  • [DATA] Basic-share getter can return period-end shares instead of the weighted-average denominator #1672
  • [DATA] Company cash-flow view lists the same dividends payment twice for JPM and KO #1674
  • [DATA] Derived Q4 changes with input order when a year-to-date period has been recast #1675
  • [DATA] extract_filing_fees_xbrl returns ixt:fixed-zero fees as an em dash and an ixt:booleantrue final-prospectus flag as False #1681
• Statements and rendering
  • [DATA] Comprehensive income rows with a negated label keep the filed sign in to_dataframe() and render() #1676
  • [DATA] Facts API takes a concept's preferred label from the first presentation role that has one, so an income-statement fact gets preferred_sign -1 from a disclosure role #1677
  • [DATA] Financials getters return None, and the income statement shows prior-year columns, for a 10-Q whose DocumentFiscalPeriodFocus is FY #1678
  • [DATA] SAP cash-flow dividends row shows the equity-statement total, not the filed financing line #1679
  • [DATA] Statement rows for us-gaap-supplement concepts are blank (GAIN, FSK, NMFC, HTGC 10-Ks filed 2023) #1680
  • [DATA] Statement rows merge distinct members sharing a label and attach the wrong amounts #1682
  • [DATA] Statements synthesize totals from non-additive per-share and percentage amounts #1683
  • [DATA] Viewer calculation validation mixes three-month net income into six-month cash flow #1684
  • [DATA] Synthesized statement totals add members in different units and cite one member's fact as their source #1689
• Stitching and multi-period
  • [DATA] Stitched balance sheet merges any two cash-bucket lines from different filings, so cash plus short-term investments can show as cash and cash equivalents #1685
  • [DATA] Stitched cash flow puts an activity's continuing-operations cash in the total row that includes discontinued operations #1690
• TTM, quarters, EPS and splits
  • [DATA] EPS share lookups use a misspelled combined-share element, so WeightedAverageNumberOfShareOutstandingBasicAndDiluted filers get no derived Q4 or TTM EPS #1686
  • [DATA] Split adjustment skips USD per-share amounts not named EarningsPerShare, so dividends per share and continuing-operations EPS mix pre- and post-split bases #1687
  • [DATA] TTM EPS averages the fiscal-year share count as if it were Q4's count, so the fiscal-year column differs from the filed EPS #1691
  • [DATA] TTM and derived Q4 EPS divide NetIncomeLoss instead of income available to common stockholders #1692
  • [DATA] TTM and derived quarters treat every monetary duration concept as additive, including concepts whose quarters do not sum to the year #1693
  • [DATA] TTM income and cash-flow statements sum four adjacent quarters without checking they are consecutive, so a missing quarter gives a 15-month total under a TTM column #1694
  • [DATA] TTM sums quarterly continuing-operations per-share amounts into a twelve-month value #1695
  • [DATA] detect_splits dates a split by the first-listed of its tagged dates, so pre-split facts filed after the announcement are never split-adjusted #1696
  • [DATA] discrete_quarters=True with use_optimal_periods=False leaves the cumulative date range on the derived quarter's header #1697
  • [DATA] get_ttm() sums four quarterly weighted-average share counts, so a TTM share count is about four times too large #1698
  • [DATA] get_ttm_revenue(as_of=...) labels quarters with the wrong fiscal year and builds them from filings made after the as-of date #1699
  • [DATA] TTM diluted EPS reuses quarterly option inclusion for a loss year #1782
• XBRL parsing (instance, linkbases, dimensions)
  • [DATA] PresentationParser drops the ending arc of negated period-boundary pairs, and Beginning Balance rows show period-end values #1700
  • [DATA] Statement member nesting uses domain-member arcs from every role, so Tesla's 'Services and other' nests under a note-only subtotal #1701
  • [DATA] Statement rows take their calculation weight and parent from an unrelated role when the statement's own role has no arc for them #1702
  • [DATA] Synthesized plain rows sum overlapping or out-of-scope dimensional members #1703

Bugs and documentation (64)

• Currency
  • Statements treat facts in different currencies as duplicates of one fact and keep only one of them #1704
  • to_dataframe(): a concept presented under two parents gets the first occurrence's parent on both rows #1705
• Docs
  • Company.md, XBRL.md and Statement.md (api-reference, .docs) teach calls that fail in 5.62.0 and on main #1706
• Entity, filings and search
  • BofA shelf_registration intermittently returns None because unordered file numbers select a truncated fast path #1707
  • EFTS search results carry the raw display name in EFTSResult.company ("NVIDIA CORP (NVDA) (CIK 0001045810)") #1708
  • EntityFacts.time_series omits duration_days on empty results and periods=0 #1709
  • XBRL.entity_info names a co-registrant as the filer when its cover facts use srt:ConsolidatedEntitiesAxis #1710
  • XBRL.to_pandas() omits available fact metadata because the catalog lookup uses colon-form IDs #1711
• Funds and BDCs
  • BDC industry cleanup discards all but the last member of an enumeration set #1712
  • extract_nonaccrual(BDCEntity) selects ARCC 10-K/A with no investment evidence instead of the original 10-K #1713
• MCP and AI skills
  • AI XBRL skill examples fail on Apple FY2025: missing attributes, unqualified exact=True, fact['unit'], instant period_end #1714
  • AI holdings skill: NEW HOLDINGS recipe returns 40 rows with Lennar CL B twice, not the 114 its comment states #1715
  • AI ownership skill: the 'Screen Large Transactions' pattern matches only net purchases, so large net sales never print (AAPL Form 4 0001140361-26-038674) #1716
  • DEF 14A example in the reports skill and ProxyStatement.sections docstring uses section keys unguarded #1717
  • Holdings AI skill and docs call Filings.filter(company=...), which raises TypeError #1718
  • MCP edgar_read advertises full_text but returns success with null on 40-F, prospectuses, registrations and 6-K #1719
  • MCP edgar_read earnings section calls FinancialTable properties as methods and returns only a placeholder #1720
  • edgar_company and edgar_trends report Python AttributeError text when company facts are missing #1721
  • install_skill() and package_skill() omit the funds skill #1722
  • test_install_skill_uses_symlinks fails on Windows without symlink rights, where install_skill() falls back to copying #1723
• Other
  • Auditor ICFR attestation parses the valid boolean value 1 as false #1724
  • Axis.default_member_id is None in axes_for_role() when the axis default is declared in another role #1725
  • ixt-sec:exchnameen returns NYSEARCA and NYSETXSE, which are not DEI exchange codes (NYSEArca, NYSETX) #1726
  • set_identity() logs the full SEC identity, including the email address, at INFO #1727
  • Balance metadata gives share counts 'credit', the operating cash total 'debit', and leaves 87 of 135 USD us-gaap concepts without a balance #1728
  • Define the total-liabilities numerator of Financials debt_to_assets #1729
• Standardization
  • Abstract status of standard concepts is guessed from case-sensitive name patterns, so Rollforward headings and domain members read as non-abstract #1730
• Statements and rendering
  • Calculation and definition role names ignore a definition already loaded from the schema #1731
  • Company.income_statement, balance_sheet and cash_flow_statement return None for invalid arguments that EntityFacts rejects #1732
  • ConceptGraph.validate() applies calculation weights to sign-flipped display values, so a cash flow that foots is reported as not footing #1733
  • Document.to_dataframe() crashes when generated duplicate-header suffixes collide with literal headers #1734
  • FactQuery.to_dataframe() returns stale rows after limit() or sort_by() changes #1735
  • Prohibited relationships remain active in calculation, presentation, definition and label parsing #1736
  • Rendered statements scale and format tax rates, share counts and per-share values as money unless the concept is in two built-in lists #1737
  • Statement.to_dataframe labels some quarters ending in a month's first week one fiscal quarter too high #1738
  • Statement.validate(level="detailed") promises line-item rollups but runs section checks #1739
  • Viewer validate() silently omits a total when one calculation child has no value #1740
• Stitching and multi-period
  • Stitched 10-Q DataFrame keys six- and nine-month columns by bare end date; multi-period guide calls it a quarterly statement #1741
• TTM, quarters, EPS and splits
  • Calculation parser ignores xlink:arcrole, keeps only the first of two arcs with different weights, and accepts a missing or zero weight #1742
  • Inline XBRL formats match by local name under any namespace, and many TR4/TR5 rules are missing (ixt:date-month-year keeps its display text) #1743
  • LabelsParser keeps only xml:lang="en-US" labels, so IFRS taxonomy documentation (xml:lang="en") is dropped #1744
  • PresentationParser keeps only the last non-empty extended link of a repeated role; IMAX FY2018 cash_flow_statement() returns None #1745
  • Statement unit display infers EPS units from the opaque reference ID #1746
  • XBRL unit with several sibling measures (a product such as feet x feet) keeps only its first measure #1747
  • edgar.documents HTMLParser inserts a space between an Inline XBRL fact and its ix:continuation, so a token split across them gains a space #1748
• XBRL parsing (instance, linkbases, dimensions)
  • 10-Q: 'Part I, Item 1' returns the contents page when the Donnelley contents parser drops the Part I rows #1749
  • A nil numeric fact outranks a valued duplicate in precision ranking because its missing decimals scores 0 #1750
  • A typed dimension member filed as xsi:nil is reported as the element name '{namespace}QName.domain' #1751
  • Cross-role dimensional domains lose their nested member hierarchy #1752
  • Inline XBRL parsing resolves contexts and units only within one HTML file, so EX-13 facts get context and unit None #1753
  • Inline duration transforms return fractional years and months such as P9.5Y, which are not valid xs:duration values #1754
  • Inline nonNumeric identifiers are classified as numeric facts #1755
  • Inline-parsed XBRLFact objects carry no fact id and lose footnotes linked by ix:relationship #1756
  • Inline-parsed Document.xbrl_facts has no nil state: a nil fact written with the xs: prefix reads as an empty string #1757
  • Presentation parser drops same-child relationships with distinct order or preferred labels #1758
  • STANDARD statement view drops nested dimension members: only direct children of a node named '*Domain*' count as valid #1759
  • Set-valued enumeration facts keep filed line breaks, so the XBRL and inline paths return different strings and duplicate removal misses them #1760
  • Typed dimension values are flattened to the first child's text or tag, and colons are rewritten #1761
  • XBRL context periods are kept as raw text: dateTime and whitespace-padded dates drop out of reporting_periods, and forever contexts never reach a statement #1762
  • XBRL contexts ignore dimensions in xbrli:scenario, so scenario-qualified facts read as undimensioned totals #1763
  • XBRL facts reported with @precision parse with decimals None and outrank a more precise duplicate #1765
  • XBRL fraction items are parsed as their numerator and denominator digits joined (1/3 reads as 13) #1766
  • XBRL.from_files() drops fact-to-fact footnote arcs and ignores the arcrole of fact-to-footnote arcs #1767
  • Escaped Inline XBRL TextBlock values keep nested ix: tags and ix:exclude content #1770

Performance (1)

• XBRL parsing (instance, linkbases, dimensions)
  • [PERFORMANCE] XBRL footnote attachment walks the fact list from the start for every relationship #1771

Feature requests (2)

• Financials getters
  • Accept include_quarterly on MultiFinancials.income_statement() #1783
• Funds and BDCs
  • Read filer industry subtotals on EquitySecuritiesByIndustryAxis and warn on low industry_coverage #1784

8-K earnings releases (10)

• 8-K
  • 8-K item detection accepts any number, so a line starting 'ITEM 2024' is reported as item 2024 and ends the item before it #1772
  • EntityFiling.parsed_items returns "" for 8-K items headed "Item - 5.02" with a dash (fixed only in CurrentReport by #1590) #1773
  • MCP edgar_read press_release returns an attachment list instead of the 8-K release text #1774
  • [DATA] 8-K caption column shifts the prior period under the current header #1775
  • [DATA] 8-K earnings lookup skips an EX-99 or EX-99.1 exhibit whose description and filename omit ex-99 #1776
  • [DATA] ASGN 8-K earnings table: 3 header labels for 5 value columns, so columns are mislabelled, nine-month columns dropped, get_key_metrics() all None #1777
  • [DATA] Alta Equipment 8-K get_key_metrics returns equipment sales of $265.6 million instead of total revenues of $481.2 million #1778
  • [DATA] Cardlytics 8-K thousands table scaled as millions, so get_key_metrics revenue is $63.249 billion instead of $63.2 million #1779
  • [DATA] EarningsRelease income table drops "% Change" headers, so a percent sits under the year-ago period #1780
  • [DATA] Mueller Water 8-K: EarningsRelease.income_statement is the FY2024 segment table, so quarter EPS 0.33 and FY2025 EPS 1.22 are missing #1781

On the second question (which branch for a fix that applies to both lines): we will open any fix against main only, unless you say you want a 5.62.x backport as well.

Thanks, and apologies for the volume of notifications in one day.