Method · updated August 5, 2026
How each figure is derived, field by field
Enough detail to reproduce any row from the filing it names.
Extraction
- Form N-PORT. The filing's
primary_doc.xmllists each holding withvalUSD,balance, and the issuer's name and LEI where present. Price per share isvalUSD / balance. Both inputs are printed on the row. - S-1 charter exhibits. The exhibit is parsed for the defined term "Original Issue Price" and its per-series values. Four distinct drafting conventions are handled, including lists that place the series name before the price. A par-value figure is rejected as a round price by an explicit control.
- Rule 3-05 statements. Target revenue is taken from the audited income statement and consideration from the purchase-price disclosure, then graded by two independent extractors. Rows where the extractors disagree are adjudicated against the filing text, and a verdict whose supporting quote does not appear in the filing is voided rather than accepted.
Naming and identity
Issuer names as filed carry class tokens, legal suffixes, and occasional overlapping text runs from the filer's own XML. Names are normalised, and where a company appears under several spellings its LEI is used to collapse them. Where no LEI exists on any row and the spellings differ, the records are left separate rather than merged, because merging two genuinely distinct companies is the worse error.
What is checked before a row is published
- Every row must carry an accession number and a resolvable source URL. 7,412 of 7,412 rows do.
- Extraction tools each ship a self-test containing both a control that must pass and a decoy that must fail, so a passing run means something.
- Rows produced by disagreeing extractors are held for review rather than published as agreed.
Known open items
Stated because a dataset that lists none is not being honest about itself.
- One charter drafting convention remains unparsed on a specific filing, and that filing's rounds are absent rather than approximated.
- Business development company schedules of investments are a fourth rail, identified and not yet built.
- Two extractors can share a mistake. One measured case: both read an acquirer's own revenue as the target's and agreed with each other. Human review of the transaction rail is therefore a gate, not an option.