Docket Article
SEC BDC Investment Data: Borrowers, Instruments, Holdings
A 2026 protocol for SEC BDC investment data: distinguish borrowers, instruments and holder positions, reconcile repeated observations, and preserve filing evidence with a proposed ledger and fictional calculations.

Summary
- 01SEC BDC investment data starts with a reported holding. Preserve the reporting holder, disclosed issuer, instrument and economic date before linking financing evidence to a company record.
- 02Release refreshes can change extracted historical records without representing new financing. Preserve the archive, digest, literal headers and reporting contexts before comparing releases.
- 03Principal, cost and fair value describe a holder's position. They cannot establish borrower revenue, enterprise value or complete indebtedness; commitments also remain separate from funded amounts.
- 04Legal-entity matching and acquisition-group mapping require separate, dated evidence. Accept a supported financing observation while holding an unresolved connection to an operating brand or parent.
- 05Suppress repeated analytical observations while retaining source lineage. Different instruments, holders and economic dates remain distinct; the proposed ledger still requires authentic archive and header verification before ingestion.
Inside this article
Executive Summary
SEC BDC investment data is useful for recording dated financing observations about privately held businesses. Its starting unit is a reported holding, rather than an acquisition target or a complete borrower balance sheet. The Securities and Exchange Commission (SEC) describes its Schedule of Investments (SOI) extract as holding-level information. A researcher should preserve the reporting business development company (BDC), disclosed issuer, instrument and observation date before deciding which company record receives the evidence. [1]
Verification date for the SEC release-page and Data.gov catalog comparisons throughout this article: November 15, 2026. The SEC download page lists releases through October 2026. Its notices describe a September addition of qtrs to soi.tsv, a September parsing correction and an October refresh adding previously excluded records. These changes make release selection part of the analysis: a changed extracted row count need not represent new financing. The catalog at Data.gov still displays a September update and resources through August, so its distribution listing should not determine which SEC release is current. [2] [3]
The recommended workflow produces keep, deduplicate, hold or escalate decisions. Separate publication dates from economic observation dates; compare instruments within their reporting contexts; preserve source labels; and resolve legal entities using dated corroboration. Extensible Business Reporting Language (XBRL) distinguishes instant and duration facts and permits repeated presentations of the same fact. Legal Entity Identifier (LEI) data can supply legal names and accounting-parent relationships, but missing relationship information does not prove independence. [4] [5] [6] [7]
The financial boundary is equally important. A holder's principal, cost and fair value cannot establish borrower revenue, enterprise value or complete indebtedness. Blue Owl's filing distinguishes enterprise value from its allocation to individual securities. [8] Main Street's schedule distinguishes cost from principal, while Ares separately reports commitments and amounts funded. Federal Reserve staff estimate approximately $570 billion of BDC assets at the end of 2025, with about 80 percent in private credit loans; these are aggregate vehicle statistics, not measures of acquisition-target coverage. This report supplies a proposed ledger and fictional arithmetic, rather than a production parser: the permitted retrieval tools could not preserve the current ZIP archive for authentic header inspection. Any implementation must complete that verification before ingestion. [9] [10] [11]
Introduction and Background
For private equity (PE) origination, the useful question is whether a disclosure adds an attributable financing fact to a company record. Finding a borrower-like name is an intermediate result. The resulting record must distinguish the holder that filed, the legal issuer disclosed, the specific security and the business group that an acquisition team is researching.
The SEC publishes flat-file BDC data extracted from tagged submissions and describes monthly updates. Its warning that the extract does not replace the original filings defines the appropriate evidence hierarchy: use the extract for discovery, then return to the filing for acceptance. [2] The separate SEC Business Development Company Report concerns registrant identification. Its BDC identifiers should not be repurposed as identifiers for portfolio borrowers. [12]
Different instruments and reporting structures complicate company counts. Sixth Street Specialty Lending's investor material describes loans, bonds and equity; its portfolio-company count also includes structured-credit investments, counting each collateralized loan obligation series separately. A collateralized loan obligation (CLO) is therefore a relevant vehicle distinction in this source, rather than a reason to assume every reported name denotes an operating company. [13]
The protocol below makes separate decisions about extraction, identity and research use. It proposes an observation ledger with reversible links to company records. It does not assess investment attractiveness or offer a loan-underwriting model. Docket's perspective here is PE origination research: financing evidence is one input to a screening decision, with unresolved identity and scope questions carried forward.
The technical distinction matters before any matching algorithm runs. XBRL facts have entity and period context; an entity identifier is qualified by its identifier scheme. An identifier's meaning therefore comes from the reporting context and documentation, rather than from its appearance as a number beside a company name. [5] Use this report as a source-reconciliation specification, then test it against the actual release and original disclosures.
Key Changes
Release refreshes can change historical observations
The SEC's current page extends beyond the September snapshot described in earlier research notes. It lists October releases and states that the October refresh included additional SOI records from previously excluded filings. The September notice separately concerns duration discrimination and parsing. These are changes to the published extraction, with implications for comparisons between downloaded vintages. [2]
A proposed release policy should retain the entire archive and an integrity digest before transformation. The National Institute of Standards and Technology (NIST) explains that message digests can detect subsequent changes to a message. A digest can establish that two retained byte sequences differ; it does not establish that either financial observation is correct. [14]
Apply these controls:
- Release identity: Record the SEC download label, actual retrieval timestamp and source address.
- Byte identity: Calculate and retain a digest of the downloaded archive.
- Schema identity: Save the literal headers before selecting or renaming fields.
- Coverage identity: Record the accessions present and the range of filing dates.
- Comparison identity: Compare releases using matched submissions and contexts before describing portfolio change.
The distinction between a source object and the activity that transforms it follows the World Wide Web Consortium's (W3C) provenance model. A release manifest should explain both what was received and how the analytical output was produced. [15]
Documentation and actual headers need separate checks
The readme describes a dynamically created SOI report whose columns can vary. It excludes custom tags and custom axes while including custom members. The displayed SOI schema in the fetched readme lacks the newer qtrs field, while the release page announces that field's addition. This documentation mismatch makes a fixed column list an unsafe implementation assumption. [1] [2]
There is also a taxonomy-version question. The Financial Accounting Standards Board (FASB) says its implementation guides are not authoritative and may concern taxonomies pending SEC acceptance. An implementation guide can help interpret tagging without becoming evidence that a proposed requirement is already operative. [16]
FASB's conceptual Measurement chapter also distinguishes an estimated exit price from an observed transaction. That is an accounting-concept reference, rather than a substitute for the filing's policy or an adopted taxonomy requirement. [17]
No authentic TSV header was inspected for this report. The built-in fetch could not read the current ZIP, and the required Firecrawl fallback returned lossy binary-as-text. Consequently, this report supplies no executable column mapping, inferred qtrs convention or archive row count. A research engineer must download and inspect the original bytes before implementing these controls.
A practical stopping rule is to accept only the narrowest supported fact.
Implementation Considerations and Process Changes
Preserve dates and contexts before filtering
Maintain separate fields for filing date, reporting balance-sheet period, fact date and retrieval time. The readme distinguishes filed, period and ddate, with the latter dates rounded to month end in the extract. The original context remains necessary where exact boundaries matter. [1]
XBRL distinguishes an instant from a duration. A balance measured at an instant and income measured over a period are different observations even when their end dates coincide. Preserve the raw qtrs value and raw duration context until their current-release semantics have been verified. Do not infer a holding period, loan age or repayment event from an unvalidated duration code. [4]
SEC staff also distinguish numerical SOI tagging from the usefulness of detailed nonnumerical tags, including issuer and maturity information. A missing extracted label may therefore require checking the filing rather than an automatic rejection of the financing evidence. [18]
The proposed acceptance sequence is:
- Locate: Preserve accession, reporting BDC name and Central Index Key (CIK).
- Open: Follow the original inline filing link and identify the relevant SOI.
- Date: Select the intended economic observation and retain comparative periods separately.
- Read: Capture issuer labels, instrument descriptions, footnotes, units and currency.
- Reconcile: Check extracted amounts against the disclosed position.
- Classify: Assign an observation decision and a separate entity-match decision.
- Review: Record reviewer, supporting passage and unresolved issue.
W3C's provenance framework distinguishes data objects, activities and responsible agents. The ledger adapts that distinction to source retrieval, transformation and reviewer decisions; it is a proposed research design, rather than an SEC-mandated schema. [15]
Blank observation ledger
Table 1 groups the fields for a blank ledger. Each blank should remain explicitly unanswered until the source or reviewer supplies a value.
| Field group | Blank fields to complete | Acceptance purpose |
|---|---|---|
| Release | release label: ___; archive digest: ___; retrieved UTC: ___; header/version: ___ | Identifies the actual input bytes and transformation version. [14] |
| Filing | accession: ___; BDC CIK: ___; BDC name: ___; filed date: ___; original inline filing URL: ___ | Separates the reporting registrant from the disclosed investment issuer. [12] |
| Time context | balance-sheet period: ___; fact date: ___; qtrs raw: ___; original instant/duration context: ___ | Prevents comparison across incompatible periods. [4] |
| Disclosed position | raw issuer: ___; raw instrument label/type: ___; currency/unit/scale: ___; principal: ___; cost: ___; fair value: ___ | Retains disclosed amounts and labels before analytical normalization. [9] [19] |
| Entity evidence | proposed legal entity: ___; jurisdiction/registry ID: ___; dated match source: ___; parent relationship and scope: ___ | Keeps name matching distinct from accounting-parent evidence. [6] |
| Decision history | observation/version key: ___; duplicate-of: ___; reviewer: ___; decision: ___; unresolved issue: ___ | Records who accepted a transformation and why. [15] |
The ledger has separate provenance and interpretation fields because a valid extraction can still have an unresolved issuer match. A company master record should point to accepted observations instead of absorbing every source row as a new company. Keep the raw value beside any normalized amount, especially where a filing's scale excludes share counts. Barings, for example, labels financial amounts in thousands while excluding unit/share amounts from that scale. [19]
For an implementable export, use these groups as column families rather than storing the entire observation in one free-text cell. Require explicit missing-value states such as not disclosed, not extracted, not reviewed and conflicting evidence. XBRL guidance distinguishes a nil fact from both zero and omission, so a common empty-cell default would erase useful distinctions. [5]
- 01Locate and open the filing
Preserve the accession, reporting BDC name and CIK; follow the original inline filing link to the relevant SOI.
- 02Select the economic date
Select the intended economic observation and retain comparative periods separately.
- 03Read labels and context
Capture issuer labels, instrument descriptions, footnotes, units and currency.
- 04Reconcile the position
Check extracted amounts against the disclosed position.
- 05Classify separate decisions
Assign an observation decision and a separate entity-match decision.
- 06Record the review
Record the reviewer, supporting passage and unresolved issue.
Borrower and Acquisition-Group Reconciliation
Match the issuer before attaching the business group
The disclosed issuer may be a financing holding company, a vehicle or an operating entity. Treat a familiar trading name as a search candidate until corroborating evidence connects it to the legal borrower. GLEIF's basic reference information includes legal names and registered addresses, while its relationship information concerns direct and ultimate accounting-consolidating parents. These serve different matching tasks. [6]
A proposed matching procedure should retain:
- Source name: The exact issuer label as disclosed, including punctuation and suffixes.
- Candidate identity: The jurisdiction-qualified registry identifier or other supported legal identifier.
- Corroboration: Dated address, filing, corporate document or issuer statement linking the labels.
- Relationship meaning: Borrower, operating subsidiary, accounting parent or other specifically evidenced role.
- Time validity: The date at which the relationship is supported.
- Disposition: Accepted identity, ambiguous candidate or unresolved lead.
A Delaware search can corroborate an entity name, file number and formation information. Its own notice warns that a name-search result does not indicate current status, and the application reflects the database as of the search date. A present search result should therefore not be represented as proof of a historical structure. [20]
Keep the legal entity and acquisition-group identity in separate tables or field groups. A consolidated accounting relationship can be relevant evidence for group mapping without establishing the exact perimeter a buyer might acquire. Likewise, GLEIF permits exceptions to relationship reporting; the absence of public parent information cannot settle whether a sponsor or parent exists. [7]
Preserve uncertainty at the field level
The proposed confidence assessment is specific to each relationship. An accepted legal-name match need not imply an accepted parent link. A correct parent link need not imply that the disclosed debt belongs to every subsidiary. GLEIF's distinction between basic identity information and accounting relationships supports this separation. [6]
When names conflict, preserve both source statements and ask what differs: the entity, date, relationship definition or reporting boundary. W3C provenance supports retaining the activities and agents behind a derived record. Applied here, a merge should retain the reviewer and supporting evidence rather than silently replacing the original issuer string. [15]
Escalate corporate succession, guarantees, borrower-group boundaries and contractual obligor questions to counsel when they matter to the screen. Escalate valuation classification to an accountant familiar with investment-company reporting. The unresolved field should name the question and required expertise, rather than implying that professional review has occurred.
A practical stopping rule is to accept only the narrowest supported fact. For example, retain a filing-backed position under its disclosed issuer while holding its connection to an operating brand. This keeps useful financing evidence available without turning an unverified name resemblance into a company-level conclusion. Registry information and parent information each have their own scope limitations. [20] [7]
Position Accounting and Duplicate Review
Amounts are evidence about a position
Principal, cost and fair value are separate measures. Main Street's schedule says principal is net of repayments and cost is net of repayments and accumulated unearned income. That disclosure is a concrete reason to preserve the measure's original label instead of replacing multiple fields with a generic funding amount. [9]
Commitment and funding also require separate treatment. Ares reported approximately $2.1 billion of new commitments in the third quarter of 2026, with approximately $1.6 billion funded. Those figures describe different measures for the same reporting discussion. They should not be used interchangeably in a borrower screen. [10]
Table 2 is a claim matrix for interpreting the disclosed amounts. The boundaries are analytical acceptance rules for this protocol.
| Disclosed measure | What it can support after checking the filing | What it cannot establish by itself |
|---|---|---|
| Principal | A disclosed principal measure for the holder's instrument at its stated date and unit. Main Street explicitly reports principal net of repayments. [9] | Borrower-wide debt, revenue, enterprise value or the total size of a financing round. |
| Cost | The holder's reported investment cost under its disclosed accounting treatment. [9] | Cash received by the borrower during the current period, or current enterprise value. |
| Fair value | The disclosed value of the position in the security owned. [1] | The value of the entire issuer, repayment certainty or default probability. |
| Unfunded commitment | A separately identified future funding obligation or commitment under the disclosure. BCP describes commitments additional to scheduled amounts. [21] | Cash already advanced or an amount that can automatically be added to outstanding principal. |
| Foreign-currency amount | A position measure interpreted with the disclosed currency and translation context. Barings separately discusses foreign-currency investments. [19] | A comparable domestic-currency amount before conversion and date conventions are checked. |
The important distinction is between a measure and the claim attached to it. An amount can be correctly extracted but incorrectly labelled in a screening database. Borrowers may combine private credit term loans with bank revolving facilities, according to Federal Reserve staff, which reinforces why one holder's position is insufficient for a complete debt total. [11]
Blue Owl Capital Corporation's September 2026 filing describes fair value as the price for an investment in a current sale under an orderly market-participant transaction at the measurement date. It separately describes allocating enterprise value among securities according to their seniority, preferences and other features. That distinction supplies direct filing evidence for keeping security value separate from company value. [8]
Ares's disclosures also describe an investment in subordinated certificates of a co-investment program. Preserve that vehicle relationship rather than automatically assigning its underlying financing to a directly held borrower position. [10] Read footnotes for participation, unfunded, currency, nonaccrual or restructuring context when those features affect classification; do not infer that any one applies from an amount alone. Barings and BCP illustrate why unfunded information can sit in footnotes. [19] [21]
Design keys for review, rather than assume uniqueness
A proposed observation key, to test against authentic archives, combines accession, reporting BDC, raw issuer/member identity, instrument/member identity, fact date, period context, currency/unit and any other distinguishing dimensions. An economic-position candidate key may link observations across filings after instrument and issuer review. Neither is an official SEC key; the readme defines no SOI key. [1]
Use a separate version key for release digest, extraction version and source-row reference. Never sum observations across reporting dates to estimate a borrower balance. Do not put fair value into the stable identity of a position: a changed amount should prompt comparison, rather than automatically manufacture a new instrument. NIST's digest guidance supports integrity checking of inputs, not the economic equivalence of rows. [14]
Before suppressing a row, distinguish these situations:
- Repeated presentation: The same fact appears in multiple parts of an Inline XBRL report. [5]
- Comparative period: A later filing repeats an earlier dated observation; preserve its source lineage.
- New observation: The same instrument is reported at a later economic date.
- Separate instrument: Debt and equity under one issuer remain distinct positions.
- Separate holder: Different BDCs report their own interests, even when borrower names match.
- Amended version: Review what changed before selecting a preferred presentation.
Repeated Inline XBRL presentation is a documented technical phenomenon. It should not be confused with multiple economically distinct securities, nor should every repeated company label be deleted as a duplicate. [5]
- The same instrument at a later economic date is a new observation.
- Debt and equity under the same issuer remain separate positions.
- Different BDCs retain their own holder interests even when borrower names match.
- Check whether the same fact appears in multiple parts of an Inline XBRL report.
- An earlier observation repeated in a later filing retains its source lineage.
- Review an amended version's changes before selecting a preferred presentation.
An amount can be correctly extracted but incorrectly labelled in a screening database.
Data Analysis and Evidence
Scope and denominator discipline
Federal Reserve staff's aggregate estimates place BDC assets at approximately $570 billion at the end of 2025, with private credit loans comprising about 80 percent. These estimates contextualize BDC financing but do not establish how many privately held operating companies the SEC SOI extract can identify. No verified coverage percentage for the acquisition-target universe was established in this research. [11]
Data sources also differ in the direction of their exposure. The Office of Financial Research (OFR) distinguishes data covering BDCs borrowing from banks from reporting on private funds. BDCs as borrowers in a bank-exposure study are not the portfolio borrowers that this protocol seeks to identify. Mixing those denominators would create a misleading company count. [22]
Distribution metadata require the same caution. The opened Data.gov record displays a dataset update of September 24, 2026, while its latest listed resource is August. This proves an official distribution route, rather than independent editorial endorsement or completeness of the latest release. Use the SEC release page to select the current source vintage. [3]
Fictional repeated-quarter and debt-plus-equity records (Hypothetical Example)
All names, dates, amounts and counts in the remainder of this example are fictional assumptions. They describe no real company, financing or client result.
Suppose invented Harbor Manufacturing Holdings has a first-lien instrument reported by invented BDC Alpha at quarter-end. A later filing repeats that earlier observation and adds a new quarter. The repeated earlier fact is a duplicate observation for a position/date view, while the new quarter is a distinct observation. If a separate BDC Beta reports a holding in the same instrument, retain a separate holder position.
Table 3 applies the proposed outcomes to the fictional records.
| Fictional record | What the reviewer establishes | Outcome |
|---|---|---|
| Alpha reports Harbor debt at quarter-end; later comparative repeats the same date, instrument and amounts | One economic observation, two source presentations | Deduplicate the analytical observation; retain both source links. |
| Alpha reports Harbor debt at the next quarter-end | Same instrument, different economic date | Keep the later observation and prior history. |
| Beta reports its own Harbor debt position | A different reporting holder | Keep separately; do not call it Alpha's duplicate. |
| Alpha reports Harbor debt and Harbor preferred equity | Two instruments under one issuer | Keep both instruments; link to one issuer only after identity review. |
| A source label says Harbor Manufacturing, with no evidence connecting it to the holding company | Name similarity only | Hold the company match; retain the source observation. |
| An amended filing changes an amount, but the correction's scope is unclear | Competing versions need interpretation | Escalate to the research engineer and accounting reviewer. |
The table separates duplicate suppression from source deletion. XBRL's documented repeated presentations justify a duplicate-review stage, while its context dimensions explain why dates and units must participate in that review. The table's particular dispositions are proposed controls, not SEC classification rules. [5] [4]
For a second fictional calculation, assume the same position/date has cost $10 million and fair value $9 million. Fair value divided by cost equals 90 percent. This is a ratio of two position measures under the example's assumptions. It is neither enterprise value nor a default probability. This is a methodological limit: Blue Owl's disclosed valuation allocation is a valuation process, not a calibrated default model. [8] Confirm that both inputs refer to the same security, holder, currency and date before calculating.
Define the duplicate-observation rate as suppressed duplicate source rows divided by all source rows examined. With 24 duplicates among 120 source rows, the fictional rate is 20 percent. Define the filing-backed match rate as reviewed issuer records with accepted filing-backed entity matches divided by all reviewed issuer records. With 32 accepted matches among 40 reviewed records, the fictional rate is 80 percent. Preserve unresolved records in that denominator; disclose any pre-review exclusions separately.
Implications and Future Directions
The first operational implication is to separate extraction success from analytical acceptance. A filing's separation of enterprise value and security-level valuation is a reminder that a correctly extracted number still requires the correct analytical label. [8] A parser can produce rows without producing a reliable issuer match. SEC staff's discussion of nonnumerical tagging shows why issuer detail and other descriptive content deserve their own coverage checks. An extraction pipeline should report accepted observations, unresolved names and unreviewed records separately. [18]
The second implication is that evidence has both an economic date and a publication history. The Data.gov catalog and SEC release page currently display different coverage endpoints. A reproducible research team should retain both retrieval context and the archive actually used, rather than allowing a catalog timestamp to stand for the latest source version. [3] [2]
The third implication concerns changes in the current view. A disappearing position is an observation about a later disclosure, with an explanation still to establish. Sixth Street's release reports exits and repayments together, illustrating why even an aggregate description cannot settle what happened to an individual missing row. The screen should record “not observed in the selected later source” until a filing or issuer statement supports a more specific event. [23]
Recommended ownership of open questions is straightforward:
- Research operations: Define acceptance states, review denominators and handoff requirements.
- Research engineering: Verify archive bytes, headers, contexts, dimension preservation and reproducible transforms.
- Accounting reviewer: Interpret cost, fair value, units and footnote-driven accounting questions.
- Counsel: Review material uncertainty about obligors, succession, guarantees and acquisition perimeter.
- Origination analyst: Decide whether the narrow supported fact changes the research priority.
Further automation should begin with retained provenance and an exception queue. W3C's model supplies a useful foundation for linking source objects to transformations and responsible reviewers. NIST's digest guidance supplies a distinct integrity control. Neither replaces the substantive task of confirming what the filing discloses. [15] [14]
Conclusion
SEC BDC investment data can add dated financing evidence to a PE research record when the observation remains traceable to the underlying filing. The central discipline is to preserve the distinction between reporting holder, disclosed issuer, instrument, position and acquisition group. A company-level answer should be derived from accepted observations, with the unresolved links still visible.
The proposed protocol starts with release and header verification, preserves raw labels and context, then separates position reconciliation from legal-entity matching. It suppresses repeated analytical observations while keeping their source lineage. It retains different instruments, holders and economic dates rather than collapsing them merely because issuer names resemble one another.
Principal, cost and fair value should retain their disclosed meanings. Commitment is separate from funding; a partial holder position cannot become borrower-wide debt; and a fair-value-to-cost ratio cannot become an enterprise valuation or a probability forecast. Fictional worked records demonstrate those boundaries without identifying real businesses as acquisition targets.
The immediate implementation requirement remains authentic inspection of the current archive, because that step could not be completed with the permitted retrieval tools. Until it is complete, the ledger is a research specification, rather than a tested parser. The resulting handoff should state which financing observations were accepted, which presentations were duplicates, which issuer matches remain on hold and which questions require specialist review.
External Sources (23)
Published by
Docket Research
Docket provides deal-origination research software and managed research for private equity teams, screening companies against a mandate and showing the evidence behind each finding.
Editorial note
Researched and written by Docket Research with AI-assisted research tools; sources are cited inline and checked before publication. Report an error: https://docket.capital/evidence-standards#corrections.