TenK: combined-item headings ("Items 1 and 2", "Items 7. and 7A.") lose items; #710 filings regressed in 5.29.0 / 5.34.0

[kelpplease]

Issue Type

• Data Quality Issue (incorrect financial values, missing data)
• XBRL Parsing Issue (statement rendering, concept mapping)
• Filing Access Issue (download failures, attachment problems)
• Performance Issue (slow operations, memory problems)
• Compatibility Issue (platform/dependency conflicts)
• Documentation Issue (unclear instructions, missing examples)

Environment

EdgarTools Version: 5.59.1 (bisected across 5.23.4 – 5.59.1)
Python Version: 3.12
Operating System: Linux Mint 22.1

Bug Description

What you expected to happen:
When a 10-K files two items under one heading, the section map holds one section covering both, and a lookup for either item returns it. #710 set this up in 5.23.4 for "Items 1 and 2. Business and Properties": tenk.business and tenk["Item 2"] both returned the combined section, verified on Viper Energy and Devon Energy.

What actually happened:
On 5.59.1 there are three failure shapes.

1. The #710 filings have regressed. Part I now starts at Item 1A. There is no Item 1, no Item 2 and no combined key, and the Business and Properties text (~360K characters) is in no section at all.

Filing	Last release that worked	First release that fails
Viper Energy 0002074176-26-000010	5.28.5	5.29.0
Devon Energy 0001193125-26-056485	5.33.0	5.34.0

"Worked" means sections contained part_i_items_1_and_2._business_and_properties and tenk.business returned ~359,700 characters. Both still fail on 5.59.1. I haven't traced the cause; for reference, 5.29.0 added agent-aware TOC parsing and 5.34.0 added the per-form section schema.

2. Headings with a period after each item number have never resolved. Freeport-McMoRan and Cheniere Energy write Items 1. and 2.:

• Freeport 0000831259-25-000006: Items 1. and 2. Business and Properties.
• Cheniere 0000003570-26-000005: TOC row Items 1. and 2. Business and Properties (non-breaking space after "Items"); body heading ITEMS 1. AND 2.    BUSINESS AND PROPERTIES

On 5.23.4 – 5.28.5, detection built keys that kept the punctuation: part_i_items_1._and_2._business_and_properties (Freeport) and part_i_items\xa01._and_2._business_and_properties (Cheniere). The #710 lookup regex part_[iv]+_items_(?:{inum}_and_\d+|\d+_and_{inum}) needs an unbroken _1_and_2, so the period after 1 defeats it: tenk["Item 2"] returned None on both, and tenk.business returned None on Freeport. On every release from 5.45.1 to 5.59.1 no combined key is built at all, and both filings look like shape 1.

3. The second item of a combined heading is unreachable. Freeport's filing also combines Items 7. and 7A. Management's Discussion and Analysis of Financial Condition and Results of Operations and Quantitative and Qualitative Disclosures About Market Risk. On 5.59.1 that content comes back only as part_ii_item_7 (which opens on the combined heading). There is no 7A key, and tenk["Item 7A"] returns None.

Error message (if any):
None. tenk["Item 2"] emits the FutureWarning "found no such item in this filing".

Reproduction

Company/Ticker: VNOM, DVN, FCX, LNG
Time Period: FY2024 (FCX) and FY2025 10-Ks
Relevant Forms: 10-K

Minimal code to reproduce:

import warnings
from edgar import find, set_identity

warnings.simplefilter("ignore", FutureWarning)
set_identity("Your Name your@email.com")

def size(text):
    return None if text is None else f"{len(text):,} chars"

FILINGS = {
    "Viper Energy":     "0002074176-26-000010",  # from #710
    "Devon Energy":     "0001193125-26-056485",  # from #710
    "Freeport-McMoRan": "0000831259-25-000006",
    "Cheniere Energy":  "0000003570-26-000005",
}
for name, acc in FILINGS.items():
    tenk = find(acc).obj()
    print(name, acc)
    print("  Part I keys:  ", [k for k in tenk.sections.keys() if k.startswith("part_i_")])
    print("  business:     ", size(tenk.business))
    print("  Item 2:       ", size(tenk["Item 2"]))
    print("  Item 7A:      ", size(tenk["Item 7A"]))

Output on 5.59.1:

Viper Energy 0002074176-26-000010
  Part I keys:   ['part_i_item_1a', 'part_i_item_1b', 'part_i_item_1c', 'part_i_item_3', 'part_i_item_4']
  business:      None
  Item 2:        None
  Item 7A:       5,815 chars
Devon Energy 0001193125-26-056485
  Part I keys:   ['part_i_item_1a', 'part_i_item_1b', 'part_i_item_1c', 'part_i_item_3', 'part_i_item_4']
  business:      None
  Item 2:        None
  Item 7A:       2,119 chars
Freeport-McMoRan 0000831259-25-000006
  Part I keys:   ['part_i_item_1a', 'part_i_item_1b', 'part_i_item_1c', 'part_i_item_3', 'part_i_item_4']
  business:      None
  Item 2:        None
  Item 7A:       None
Cheniere Energy 0000003570-26-000005
  Part I keys:   ['part_i_item_1a', 'part_i_item_1b', 'part_i_item_1c', 'part_i_item_3', 'part_i_item_4']
  business:      None
  Item 2:        None
  Item 7A:       1,344 chars

Additional Context

• The TenK.business returns None when filing uses combined "Items 1 and 2" heading #710 regression test injects already-normalised section keys, so it could not catch detection no longer producing them. A test that parses one of these filings end to end would.
• The punctuation problem isn't specific to Items 1 and 2 (shape 3 is 7/7A). A suggestion, not based on reading the current code: when building a combined key, drop punctuation and normalise whitespace (including \xa0) around the item numbers and the conjunction, so Items 1. and 2., Items\xa01. and 2. and Items 1 and 2. all yield items_1_and_2.
• It would help callers if a combined section said which items it covers, as the items_1_and_2 key used to. We consume tenk.sections directly and would otherwise have to parse heading text to know that Item 7 also holds 7A.
• Related: Bug: Not Capturing 10-K Item 1 text #45 (Permian Resources, combined Items 1 and 2 left unfixed in 2024), TenK.business returns None when filing uses combined "Items 1 and 2" heading #710.
• TenK.business returns None when filing uses combined "Items 1 and 2" heading #710 asked for more filers using this pattern: Freeport-McMoRan, Cheniere Energy and Talos Energy (0001193125-26-067807) all use it.

[dgunning]

Thanks @kelpplease, this is an excellent report. The bisect points and the observation that the #710 test never exercised detection are especially useful. I reproduced all three shapes on main. This is a regression and is being tracked as P1. The fix will normalise combined-item keys across punctuation and \xa0, and it will come with end-to-end tests on these filings.