Independence
Our credibility rests on the filings, not on us.
Two questions decide whether a measurement like this is worth anything: who pays for it, and whether the matching behind it can be challenged. This page answers both, and gives you the accession numbers to check the second one yourself.
Who commissions the work
The independence question is not who we are. It is who pays us — and what they are allowed to change.
The quarterly corroboration file is commissioned by the auditor or the fund’s board — the parties testing the marks. Never by the fund being tested.
A manager who wants the same evidence for its own committee can take a seat on the platform and pull it directly. What we will not do is sell a manager a report about itself: the day such a report is unflattering, it does not go in the file — it gets buried, and the next one gets written to be buyable. That is the incentive that discredited issuer-paid ratings, and it is far more acute at five clients than at five thousand.
- Your fund is in the index either way
Being a client does not add you, remove you, or change how you are measured. Non-clients appear on exactly the same terms.
- No figure moves at a client’s request
An error is corrected for everyone the day it is shown to us, and the correction is dated. Nothing is delayed, softened or withheld.
- A report says what the filings say
If the marks sit above the co-lenders, the report says so. That is the only reason it is worth putting in a file.
- We are paid for the work, never for the conclusion
No engagement is priced on, contingent on, or renewed on what the numbers turn out to show.
A loan we refused to count
Anyone can write that their matching is careful. Here is a case where two positions looked identical and we rejected the pair — with the filings, so you can judge the call rather than trust the rule.
RYAN LLC
Same borrower. Same spread, to the basis point: +350 bps. Same seniority: first lien. Two unrelated managers. On the matching key alone, this is one loan held by two lenders — and their prices would have been compared.
- Maturity declared
- 2029-01-28
- Accession
- 0001193125-26-124286
The two maturities are 46 months apart. Two lenders holding the same instrument do not disagree by 4 years on when it comes due — so these are different tranches, and the pair was broken.
Maturity cannot be part of the matching key — only a minority of vehicles tag it, and the largest non-traded funds never do. It is used as a rejection test instead: where both sides declare it and they differ, the pair breaks.
This errs one way on purpose. A false negative costs one observation. A false positive publishes a fabricated gap against a named borrower.
- Comparable positions
- 314,960
- Dropped before publication
- 5,320
- Maturity conflicts between managers
- 31
RYAN LLC is one of them. The others are not hidden — they are simply absent from every figure we publish.
How often we are wrong
One case we refused is an anecdote, not an error rate. Two independent tests run on every published loan, and 12.11% of the loans they can examine get flagged — roughly one in eight. That is the weakest number in this dossier, and it is measured on the same run that produced every other figure here.
The all-in rate is not part of the matching key, so it is an outside witness. Two holders of one tranche report the same rate, give or take the fixing date. More than 50bps apart, PIK netted, and they are not holding the same instrument.
The rate test is structurally blind to two companies borrowing at the same spread — and the spread is in the key, so that is every group it forms. This test asks whether the shared word identifies anyone. "Avalara" does. "Foods", "packaging", "international" do not.
Flagged twice over. These are the groups where the match is least defensible, and they are excluded on either ground alone.
| Loans published this quarter | 1,036 | three managers or more |
| Flagged by at least one test | 115 | 11.1% of the panel — a floor, not a rate |
| Loans the rate test can examine | 925 | 89.29% — the rest declare too few all-in rates |
| Flagged among those | 12.11% | the comparable figure, quarter to quarter |
A flagged loan is dropped, not footnoted — it carries no price into any figure on this site. What the two denominators say is that the honest number is the second one: where neither test can see, we are not claiming a clean match, only an unexamined one.
the name test is calibrated on 8 cases of known truth — five real matches that must survive, three merged groups that must be caught. 8 is a small calibration set and we would rather say so than let the threshold read as settled
A second holder, from outside the BDCs
The two tests above run on our own filings — same regime, same chain. Business development companies are exempt from Form N-PORT; some registered funds that hold the same direct loans are not. Where one of them does, a loan we could only see through a single manager gets a second, independently filed price.
Every quarter, this many loans on the panel come from a single manager inside the BDC universe — no second filer to compare against.
A registered fund outside that universe holds the same loan and files N-PORT — 8.2% of single-holder loans gain one.
Carries a price and an accession number on this site. The rest failed the borrower-name link, not the mark itself.
On loans a BDC panel already prices by consensus, an N-PORT price lands within 2 points of it 95.9% of the time — against 55.7% for the same pairs with names shuffled. A matching rule guessing at random would not clear that gap.
On loans with no rate to cross-check, the false-match ceiling is 10.4–11.0% — higher than our internal panel, because a name is the only witness available. It is a range, not a point: the placebo draws a partner at random, and across 12 draws that choice alone moves the estimate by more than half a point. Restricted to the 270 loans backed by two independent N-PORT filings, it tightens to 5.5%.
One loan, one vote. The harshest of the three, and the one every other number on this page is quoted against.
Each loan carries its size across the $7.54bn unlocked. The unit that answers "how much of my book is mismatched" — still above our 4.5% internal bar.
Flattered, and shown only so the comparison is available: a loan filed by forty funds counts forty times, and the most-held loans never miss.
Of the 485 unlocked loans, the share where our price and the N-PORT price disagree by more than five points. A wrong match lands here — but so does a real pricing disagreement, which is why what follows is a ceiling and not a rate.
The same loans matched against unrelated N-PORT positions that pass the rate and maturity filters (n=4,602). This is what the tail looks like when the borrower name carries no information.
The same ceiling with the tail drawn at ten points instead of five. A wider bar admits fewer genuine disagreements into the tail, so the estimate falls — the ceiling is sensitive to where the line is drawn, and this says by how much.
The ceiling is those first two numbers divided: 3.5% ÷ 33.6%. If a share f of our matches were wrong, they would land in the tail at the placebo rate, so the observed tail bounds f from above. Nothing here is modelled — both tails are counted from filed prices, and the arithmetic between them is the whole estimate. That division lands on 10.4%, the bottom of the range above, because the placebo tail shown here is the one fixed draw the published figure uses; redrawing it moves the denominator, and that is what the other 0.6 of a point is.
read from 555 registered filers · 248,074 positions · $564.6bn tracked, as of 2026-03-31 · N-PORT is quarterly, not monthly — the gain in cadence comes from filers on an off-cycle fiscal year, not from a faster filing
Verify any figure yourself
No value here is modelled, smoothed or estimated, and none of it is licensed from a dealer. Every price is read from a Schedule of Investments and carries the accession number it came from. One field in the matching key is inferred rather than read: where a filer omits seniority, it is taken from the dominant seniority of the same borrower at the same spread — 17.8% of declarations this quarter.
Take the case on the home page. AmeriLife Holdings LLC — a $1.41bn loan held by 17 funds across 6 managers. The 1178 basis points between the highest and lowest value is not our estimate: it is the difference between two numbers those funds filed themselves. Here are the first of them.
17 filings in total · the rest are in the subscription
How we know the gap is real
The fair objection to everything above: maybe the disagreement we measure is our matching going wrong. It is testable — and the test was already in the panel.
The same lenders hold two kinds of loan. Some are widely syndicated and quoted by brokers, so every holder reads the same screen. Others are bilateral, with no secondary market, so each lender values them on its own assumptions. We match both the same way — same rule, same code, same filings. If our matching were the problem, both would look alike.
Priced off a screen
27bps
observable inputs — level 2 · 760 loans
Priced by model
104bps
unobservable inputs — level 3 · 1,189 loans
Where the price is observable, lenders land 27 basis points apart — rounding and a day’s timing. Where it comes from a model, they land 104 apart: 3.8× wider, on loans matched by the identical rule. The gap is a property of the credit, not of our method.
| Lenders on the loan | Priced off a screen | Priced by model | Ratio |
|---|---|---|---|
| 3 | 25 (345) | 84 (449) | 3.3× |
| 4 | 36 (173) | 108 (253) | 3× |
| 5 | 26 (99) | 106 (187) | 4.2× |
| 6 | 33 (69) | 149 (106) | 4.6× |
| 7 | 34 (16) | 114 (45) | 3.4× |
| 8 | 24 (28) | 130 (58) | 5.4× |
| 9 | 32 (17) | 150 (36) | 4.7× |
| 10 | 29 (8) | 291 (12) | 10× |
compared at equal lender counts, because a max-minus-min gap widens with the number of reporters · the level is declared by the filer in its own N-PORT, not assigned by us · 396 loans whose holders disagree on the level are excluded · a structural split points the same way — loans carrying a CUSIP sit 31 bps apart, those without 104, 3.3× — though it tracks the fair-value level rather than confirming it from outside
What we do not claim
For an entity with no track record, one caught overstatement costs more than a dozen honest limitations. So here are the limitations.
Our methodology has not been reviewed by an outside firm, and we will not imply otherwise. The day it is, we will say so and name them.
A fund’s schedule must tie to its balance sheet within 0.5% before its loan-level detail is published. That gate covers $350.3bn of the $531.9bn we track. The rest is measured but not published at loan level.
A wide gap does not indicate which lender is wrong. Marks can differ on stale dates, recovery assumptions, or genuine information. We publish the difference; the interpretation is yours.
Everything on this page is built from 145 filings as of 2026-03-31. If a number cannot be traced back to one of them, it is a bug and we will treat it as one — publicly, and dated.
Ask us anything about the method
Reply within one business day.
Kanonstone LLC · New Mexico