Any vendor can hand you a number for every borrower in a portfolio. We read 14 BDC filers' own Schedules of Investments — ARCC, BBDC, BXSL, CGBD, CSWC, FSK, GBDC, HTGC, MAIN, NMFC, OBDC, PSEC, SLRC, TSLX — score 1,985 of 3,065 borrowers with a score rebuilt on every BDC lender's own marks, and refuse 1,080 by name: holdings a lending BDC reports as its own 1940 Act control or affiliate investment, borrowers already on non-accrual, and borrowers no BDC lender's debt position resolves to in the SEC BDC Data Sets. Where a filing says nothing, we print an em-dash instead of an estimate.
A Schedule of Investments does not always disclose what a score needs. When a borrower's filing discloses neither a spread nor an exposure amount, the score prints an em-dash and the borrower is refused by name instead of estimated. That refusal is the first thing on this page because it is the rule everything else follows.
| Borrower · filer | Missing field | Score | State |
|---|---|---|---|
| Ivy Hill Asset Management, L.P. · ARCC | none - the lender's own 1940 Act affiliate | — | REFUSED |
| Credit Opportunities Partners JV, LLC · FSK | none - the lender's own 1940 Act affiliate | — | REFUSED |
| Senior Direct Lending Program, LLC · ARCC | none - the lender's own 1940 Act affiliate | — | REFUSED |
| National Property REIT Corp. · PSEC | none - the lender's own 1940 Act affiliate | — | REFUSED |
| First Tower Finance Company LLC · PSEC | none - the lender's own 1940 Act affiliate | — | REFUSED |
Four verification levels. Confidence scores in the served envelope are capped by validation grade. The badge on every figure tells you which grade applies.
Sample, synthetic or scaffold data with no fitted model behind it. Shown only to describe a structure, never as a basis for a decision.
Real underlying data and a fitted model, but no validation gate has been passed yet.
Validation evidence exists, in sample or out of time on a single window, but no point-in-time out-of-sample test has been passed and no track record is recorded.
Passed a held-out, point-in-time out-of-sample test on its recorded date. The level describes that record.
The disclosed-non-accrual grade is the better-founded of the two, because no duplicated position can inflate it — so it leads.
The score is concurrent classification: it reads impairment already disclosed in the same filing that supplies its features. There is no forward observation window, so it cannot tell you what happens next quarter, and we do not attach a horizon to it anywhere in the product.
What it is good for is triage across a book you have not read yet: it ranks which Schedules of Investments deserve your afternoon.
| Signal | What it measures | Grade | Level |
|---|---|---|---|
| BDC panel score | Disclosed impairment, concurrent — no horizon | LOBO AUC 0.5892 [95% CI 0.53, 0.65] | L2 |
| Fed statement tone | The tone of each FOMC statement, scored clause by clause from the Fed's own text, with its date and source. Pre-registered test: after a hold, does the tone call the next decision? It does not beat the rule "the next decision repeats this one": an honest negative, no accuracy claim. | 68.8% right, below the last-move rule | L1 |
| COT commodities positioning | WTI crude (CL), CI90 0.5341–0.6933, n=140, as of 2026-10-22 on the validation record. Of the 22 markets tested, 5 passed the point-in-time gauntlet (run 2026-07-23), but none beats a coin once all 22 are counted (re-graded 2026-10-22): an honest negative, no accuracy claim. | CL AUC 0.6137, not significant | L1 |
| NSS 5Y repricing | A repricing measurement, not a forecast | MAE 8bps | L2 |
| Treasury 10Y VaR coverage | One-day 99% VaR of a par 10Y Treasury, tested one day ahead, 2000–2026 (n=6,699). Coverage is on target (30Y too), but it does not beat a simple rolling normal VaR: no accuracy claim. | 1.09% of days vs 1% target | L1 |
| Industrial nowcast | MF7. The next unpublished month of US industrial production. Its edge over simple baselines is not significant: an honest negative, no accuracy claim | +3.6% vs the mean, not significant | L1 |
We publish the accuracy figures we withdrew and what killed each one. The recession probit's 0.862 AUC came down on 2026-07-20 after label leakage was found at the validation gate, and it has never been restated.
Five stages from first look to quarterly monitoring, with a gate at the end of each that shows its arithmetic. Your model stays your model: figures leave through a formula. It augments the workbook you already have; it never replaces it.
Our price is on the page, so you can budget before the first conversation.
All 14 built verticals · 2,500 API credits / month · Excel add-in (pilot)
Request accessModel documentation for SR 26-2 §VII vendor-model review · scoped onboarding
Request accessRole-based API access across four roles, hardened sessions, TLS in transit and default encryption at rest, per-tier rate limiting, and model cards written for SR 26-2 §VII vendor-model review.
We are not SOC 2 certified, and we won't claim otherwise before an auditor says so. SOC 2 is a readiness programme for us, not a certificate. For your vendor-risk review we will answer your questionnaire directly and share the model documentation; where an answer is ‘not yet’, we will say that.
Read the validation record, the verification levels and the retired-claims log before you speak to anyone here. If the evidence does not hold up, we would rather you found out on this page than in your first quarter.