Reference data — BSL universe and BDC holdings

The production reference corpus is built exclusively from primary regulatory filings. No commercial data-vendor licence sits in the chain. Illustrative demonstration data used on the public sample report is acknowledged separately below and is not held out as production reference.

1. Principle — primary regulatory sources only

The Verus production reference corpus draws exclusively from primary regulatory filings: SEC EDGAR, UK Companies House, and equivalent disclosure regimes. markst does not license, redistribute, or embed content from commercial market-data- vendor feeds (LSTA, LCD, Pitchbook, LSEG/Refinitiv, S&P Capital IQ, Morningstar, IHS Markit, Bloomberg) in the production corpus. Customers with active subscriptions to such feeds can overlay them on their own data; Verus ingests them read-only and does not propagate them to any other tenant.

2. BDC holdings dataset — production, operational

Source. SEC EDGAR — quarterly 10-Q filings lodged by US-registered Business Development Companies (BDCs). These filings are mandatory public disclosures.
Scope. Schedule-of-investments rows (obligor, facility type, par, fair value as marked by the BDC, stated coupon). Every row carries the SEC accession number and filing URL; the raw row is content-hashed before insertion for idempotence.
Refresh cadence. Quarterly, within ten business days of each BDC's 10-Q filing date. Ingestion is rate-limit-compliant with SEC fair-access policy and uses a descriptive User-Agent per the EDGAR guidelines.
Licensing basis. US federal government works are not copyrightable (17 U.S.C. §105) and 10-Q filings are lodged with the SEC under public-disclosure obligations. Where factual data points are extracted (obligor, fair value, coupon), the underlying principle is that facts are not subject to copyright — the extraction is of information, not creative expression. markst reproduces these facts rather than the text of the underlying filings.

3. BSL facility dataset — production, under construction

Status. The production BSL (broadly syndicated loan) corpus is under construction as a primary- source-first rebuild; as of 26 September 2026 it is not yet populated for production use. The ingestion pipeline exists — an SEC EDGAR 8-K credit-agreement scanner that identifies filings with credit-agreement exhibits and extracts facility metadata at origination. The production corpus is populated from that pipeline only, with accession-number provenance on every row.
Scope, once populated. Facility-level metadata: obligor, tranche, tenor, origination margin / spread over reference rate, covenant metadata where publicly disclosed in the filed agreement, sector. Where a facility's economics are not disclosed in a regulatory filing, the facility is not in the corpus — private placements with no regulator trail are acknowledged as outside coverage.
UK and EU counterparts. UK Companies House filings (for UK-domiciled borrowers with filed security agreements) and equivalent disclosures in supported EU jurisdictions will be added as discrete ingest modules with the same primary-source discipline. Until those modules are operational, UK/EU facilities are not in the production corpus.
Licensing basis. Identical to §2 — extraction of facts from primary regulatory filings, not redistribution of any third-party compilation. No commercial-vendor citation sits in the production corpus.

4. Demonstration data — NOT the production corpus

The public sample report and guided product walkthroughs render against an illustrative sample dataset that pre-dates the production primary-source discipline described in §§1–3. It is scaffolding for the public demonstration surface, not a compliance-grade reference corpus.

File. data/comparables_seed.csv, loaded by a demonstration-flagged path (app/engine/comparables.py::load_seed_comparables) that logs a warning at startup identifying itself as sample data. The boundary is documented in data/README.md.
Source mix. The seed mix is retired from the live platform as the production primary-source corpus fills out. A per-citation inventory is available to procurement and audit reviewers under NDA.
Use. Public sample report, screenshots, and development / test fixtures only. Never held out to a customer or procurement reviewer as the production reference corpus. Customer-visible reference queries on a live book draw from the production corpus described in §§2–3.
Forward discipline. No new commercial-vendor citations are added to this file. It is frozen except for backfills toward primary sources. As the production BSL corpus fills out under §3, the demonstration seed will be retired from the live platform and retained only as a test fixture.

5. Benchmark market data (rates & spreads)

OIS curves and other public-domain series (US Treasury yields, SOFR, ECB and BoE reference rates) are retrieved from the FRED economic-data service operated by the Federal Reserve Bank of St. Louis. markst calls the FRED API directly at request time, observes FRED's 2 req/sec rate limit, and cites FRED on any output that consumes the data. As US Treasury and Federal Reserve outputs are not subject to copyright, these series may be cached locally for performance and reproducibility.
CDS index prints (CDX and iTraxx family daily medians) are computed by markst from the public price-dissemination repositories mandated under CFTC Part 43 (DTCC Swap Data Repository). This is regulatory dissemination data — free to persist, aggregate, and display; block-masked notionals are excluded from the medians.
ICE BofA OAS series (BAML*) are not used.The platform's spread context is the Verus spread composite — markst's own calculation from SEC EDGAR BDC filings and the DTCC CDS prints above. No ICE BofA index data is fetched, persisted, displayed, or reproduced in any markst surface or client deliverable, so no third-party index licence is implicated. (A storage-layer guard still refuses BAML* persistence as a permanent Terms-of-Use tripwire.)

6. Client sublicensing posture

Because the production corpus contains no commercial-vendor content, no sublicence pass-through is required for clients to use Verus. This stands independent of the demonstration data in §4, which is not exposed to customer tenants.
If a client chooses to overlay their own licensed market- data feed (LSTA, Pitchbook, LSEG, S&P MI, etc.), the client remains responsible for that feed's terms. Verus ingests them read-only into the client tenant only and does not propagate them to any other tenant or to the production reference corpus.
Custom reference datasets built for a specific client under an engagement (for example an internal managed-account peer group) are scoped to that client's tenant and not redistributed.

7. What the production corpus is used for

Spread calibration and peer comparables on live Verus valuations, where the client enables reference-overlay.
IFRS 13 Level-2 supporting evidence (observable comparable pricing) — when a BDC-disclosed or EDGAR-8-K-sourced BSL mark sits inside the configured tolerance of the DCF output, the run is classified Level 2.
Outcomes-analysis and challenger-model inputs for model validation (see /model-risk §5–§6).

8. Limitations

Primary-source coverage is uneven. It is dense for US BDC holdings (directly filed with the SEC) and for US BSL facilities with public credit-agreement exhibits; sparser for UK / EU private placements and for bespoke club deals where no regulator-filed document exists. Absence of a reference comparable does not invalidate a Verus mark — it widens the defensibility surface and pushes the valuation toward Level 3 under the IFRS 13 resolution.

9. Contact

Questions about sourcing, licensing, or adding a primary source to the production corpus: support@markst.com.

Reference data provenance