READ IN APP
My day job is increasingly BDC analysis for clients. Clients don’t love seeing their bids move away from their ask, which means publishing this probably costs me revenue. I am doing it anyway.
I have no access to loan docs, balance sheets, or cash flow statements. The methodology I run on those filings is, I’d argue, closer to economic reality than the marks the funds publish. Even if you don’t buy that — take this one: I run the same methodology on every BDC. My skin does not crawl equally.
I hate what these guys are doing like a fat kid hates celery. I cannot tell you how aggravating it is to hear the bullshit they spew. Alas — let’s begin.
❧ ❧ ❧
Ares Capital has two loans to Pluralsight marked 94 cents apart.
Same borrower. Same date. Same maturity. Same label. One trades at par. One trades at 6 cents. Both are on page 13 of the Q1 2026 10-Q, filed April 28, 2026:
Pluralsight, LLC and Pluralsight Holdings, LLC and Paradigmatic Holdco LLC. First lien senior secured loan, 6.67% SOFR+3.00, acquired 08/2024, matures 08/2029. Principal $21.8M. Cost $21.7M. Fair Value $21.8M.
Pluralsight, LLC and Pluralsight Holdings, LLC and Paradigmatic Holdco LLC. First lien senior secured loan, no coupon shown, acquired 08/2024, matures 08/2029. Principal $23.1M. Cost $21.9M. Fair Value $1.3M.
You do not see junior debt at pennies and senior debt at par in public markets. You can’t. When a public company gets in trouble, the whole capital structure moves together. Carvana senior at 70 cents — Carvana unsecured at 30. AMC senior at 60 — AMC sub at 5. They all bleed in unison because they’re all looking at the same default. The market prices default, not the label.
ARCC’s first 1L at Pluralsight is at par. ARCC’s other 1L at Pluralsight is at 6 cents. Same borrower. Same day. Same words on the page.
That is impossible in a market. It is routine in a model.
There is a private contract between the lenders that says one loan eats first and the other eats whatever’s left. The contract is called an AAL — agreement among lenders. ARCC doesn’t disclose it. There’s not even a footnote. They label both lines the same way and let the reader assume if they know how to know better.
Why is the gap there at all? Because Pluralsight is dead. It is a shittier version of Chegg. Chegg sold homework help to college students and got mauled by ChatGPT. Pluralsight sells pre-recorded video courses to developers — an audience even less sticky than students, even more AI-native, with zero loyalty to a streaming library when GitHub Copilot, Cursor, and Claude Code will answer the same question in the editor for free. Vista paid $3.5B for it in 2021. They marked the equity to zero in May 2024. Lenders took the company in August 2024. Here we are: page 13 of the Q1 2026 10-Q, two same-day senior loans, 94 cents apart, both still labeled “first lien senior secured loan.”
What follows is a line-by-line rebuttal of what Ares said on the April 28 earnings call. Management told you the software book is fine. The marks were “spread widening.” EBITDA grew 9%. The 67-quarter dividend streak is intact. Three names — out of about 600 — are “high risk.”
The same 10-Q the press release came out of tells a different story on every one of those claims. Below: their quote, the filing, what it means. At the end is the list of names management did not call high risk. There are at least 30.
- “The software book is fine. 1% high AI risk. 3 names.”
That’s the load-bearing claim from the April 28 earnings call. An “independent consultant” reviewed the software book. 85% low risk. 14% medium. 1% high — three names, sponsor-backed, 0.3% of the total portfolio. Can you feel the McKinsey?
Software hides everywhere it’s not labeled. AthenaHealth — a healthcare-revenue-cycle SaaS — sits in Health Care Equipment & Services, page 23 of the SOI. Symplr, also healthcare compliance SaaS, page 23. Himalaya / BCPE Hyperlink, payor SaaS, page 23. GHX, healthcare supply-chain SaaS, page 23. Vista Higher Learning — digital language learning K-12, the same business model that wiped Pluralsight — sits in Consumer Distribution & Retail, page 43. Imaging Business Machines, document scanning being eaten by AI vision, sits in Capital Goods, page 49. Add it up: $1.16B of “healthcare,” another $400M+ scattered across business services, insurance, capital goods, consumer.
Reclassified by what the businesses actually do, about $4.5B of ARCC’s portfolio sits at medium-or-higher AI risk. Equity NAV is roughly $14B. That’s 32% of NAV.
15× the headline. 5× the full at-risk number. 3 names became 50.
The consultant’s job was to give management a number low enough to anchor on. The job got done.
- “NAV moved on spread widening, not credit.”
This was the framing for the −$416M net unrealized loss on the Q1 press release. Spreads moved. The marks moved. Not a credit problem.
Read the SOI quarter-over-quarter.
I matched 768 debt positions across the Q1 2026 SOI and the prior-period schedule (also in the same filing). 144 of them were marked down by 1 cent or more in one quarter. Strip the FX-flagged Euribor and SONIA positions, strip the principal paydowns that drag fair value without changing marks, and you’re left with about $255M of clean credit-driven markdown.
122 of those 144 are NOT on non-accrual. They’re “proactive” markdowns — the manager cut the price but did not classify the loan as impaired. Cornerstone OnDemand’s full stack — 1L revolver, 1L term, 2L — got cut about 17 cents across the board, page 7 of the SOI, with no non-accrual flag. ATI Restoration’s four debt tranches all moved 100c → 87c, also without a non-accrual flag. Symplr’s six tranches each moved between −14 and −17 cents.
Spread widening doesn’t write Cornerstone down 17 cents and leave Anaplan alone. Spread widening doesn’t take Symplr down across six tranches at the same time. That’s specific borrower deterioration, lots of it, surfaced through marks before it surfaces through non-accrual.
The proactive bucket is 88% larger than the deepening-non-accrual bucket. That’s the part the non-accrual line hasn’t caught up to yet.
Now do the math on what the non-accrual line should actually be.
ARCC reports non-accruals at 2.1% at cost — about $622M. The proactive bucket is the bridge to where the number is going. Cornerstone should already be on non-accrual; AlixPartners is engaged. Symplr should already be there; every tranche is down 14 to 17 cents. ATI Restoration’s four tranches all moved 100c → 87c. AllClear’s legacy paper sits inside an executed recap. Add those and the smaller proactive cuts and you reach roughly $1.4-1.5B of credit-impaired exposure that belongs on the non-accrual line. Call it 5% at cost, not 2.1%.
Push harder. About $1B of forced-equity preferred is accreting at 10-15% PIK, marked at par. Another ~$1B of “performing” first liens carry meaningful PIK components — the full list is in section 4 below. That’s roughly $2B of phantom income — accreted to principal, booked as earned, never paid in cash. None of it is reaching the cash flow statement. If you treat it as economically non-accrual — which is what the cash flow statement already does — the rate lands closer to 10-12% at cost. Roughly 5x to 6x what management reports.
ARCC reports 2.1%. That number is below the BDC peer median of about 4%, and management cites it as proof of credit quality. Oh? It is also a rate that excludes 122 names quietly cut on the marks line, $1B of forced equity at par, and another ~$1B of PIK accreting on “performing” first liens. The peer median is a low bar. Everyone in this business marks from spreadsheets.
- “Portfolio company EBITDA grew 9%.”
A small detail. On the Q4 2025 call, management said “9% organic EBITDA growth.” On the Q1 2026 call, the word “organic” disappeared. Now it’s “9% on a comparable basis.”
That word did the work. “Organic” means same-store, ex-acquisitions. “Comparable” lets in roll-up math — buy a company in March, count its full-year EBITDA in April. PE owners love bolt-ons. Bolt-ons are how a flat operating company turns into a “growing platform” in a deck. The word that disappeared was the word that mattered.
The 9% is also a weighted average. No dispersion. No deciles. No count of how many borrowers grew, how many shrank. In credit, the average is meaningless. You collect cents per year of carry on every loan; you lose 100 cents of principal on any one. ARCC’s book needs a hit rate north of 90% to make money. The question isn’t what the average grew. The question is what fraction of the book shrank, and by how much. The 10-Q does not show that.
For comparison: the Lincoln Private Market Index, 1,600 borrowers across 175+ funds, had Q4 2025 EBITDA growth at 4.7% — decelerating. ARCC’s 9% is roughly twice that. Lincoln includes bolt-ons too, so the comparison isn’t pristine — but Lincoln publishes a methodology and a sample. ARCC publishes a number. ARCC’s own non-accruals also climbed 1.5% → 2.1% in twelve months. If borrowers were genuinely growing 9%, non-accruals would be falling. I don’t need to tell you what non-accruals are not being counted as they’ve been counted. You are looking at apples-to-apples deterioration.
- “67 consecutive quarters of stable or increasing dividend.”
Maintained at $0.48 per share for Q1 2026, per the press release. Core EPS was $0.47.
The dividend now exceeds core earnings.
What’s covering it? PIK accretion — payment-in-kind, which means the borrower can’t pay cash interest, so the unpaid amount is added to principal. Non-cash income on the income statement. The 10-Q books it as earned. The cash never arrives.
Read down the SOI and tally the PIK preferred carried at par:
AthenaHealth, $309M Series A pref at 10.75% PIK, page 23 Digicert, $208M Series A pref at 10.5% PIK, page 8 Cornerstone OnDemand, $179M Series A pref at 10.5% PIK, page 7 Centralsquare Technologies, $111M Series A pref at 15% PIK, page 7 Banyan Software, $95M Series A pref at 14% PIK, page 6 Cardinal Parent, three series totaling ~$80M at 11% PIK, page 6 Kaseya, $34M preferred at 14.33% PIK, page 11 ZenDesk, $10M Series A pref at 13.17% PIK, page 17 That is roughly $1B of forced equity wrapped as preferred, all marked at par. PIK accretes the principal balance every quarter. ARCC books the accretion as income.
Then the PIK components of “performing” first liens — another ~$1B of phantom income hiding inside loans that look like they’re paying:
Adonis Bidco — 2.88% PIK on $237M, page 5 Severin Acquisition — 2.25% PIK on $145M, software section ID.me — 5.25% PIK on $130M, software section Himalaya / BCPE Hyperlink — 2.25% PIK on $254M, page 25 GHX Ultimate Parent — 2.5% PIK on $109M, page 25 HS Purchaser / HelpSystems — 3.25% PIK on $63M, software section These are the loans where ARCC collects most of the coupon on paper and a fraction of it in cash. The PIK accretion is income on the income statement and zero on the cash flow statement.
Now the legal trap. ARCC is a Regulated Investment Company. By law, RICs must distribute at least 90% of taxable income each year to keep pass-through tax status. PIK accretion is taxable income — even though no cash arrived. ARCC isn’t just paying the dividend out of phantom income. They’re legally required to.
What that means in cash terms: PIK accretes to principal → 90% of it must go out the door as a cash distribution → cash dividend exceeds cash interest received → the gap fills from new equity issuance or new debt.
Speaking of debt. ARCC reports debt-to-equity at 1.10x. That means borrowings are 110% of NAV. Roughly $15B of debt against $14B of equity. Plus indirect leverage stacking through IHAM (a $12.5B PIV running its own facilities) and SDLP (a JV with its own debt). When the underlying marks move, the equity absorbs it first. The 67-quarter streak depends on the marks not moving.
Do I need to belabor points to answer why they are selling $1.5 billion at the market? 5. “Pluralsight was an isolated event.”
Pluralsight is the visible one. The 10-Q has eight more.
I scanned every borrower for the same fingerprint — two same-borrower, same-maturity loans both labeled “First lien” with a mark gap above 50 cents:
Pluralsight, page 7. 94-cent gap. Vista Equity wipeout closed August 2024. Kellermeyer Bergensons, 95-cent gap. Janitorial roll-up where Cerberus equity is no longer named in the post-2024 cap structure (KKR / Ares / BlackRock recap announced March 25, 2024). ARCC put on non-accrual August 2025.
Eagle Football Holdings, sports section. Two pairs (83-cent and 67-cent gaps). Cork Gully appointed administrator March 27, 2026 under Ares’ qualifying floating charge.
VPROP / V SandCo. Two pairs (83-cent gaps). Already emerged from Vista Proppants Chapter 11 case 20-42002, N.D. Tex., June 9, 2020, plan effective November 6, 2020. The 2024 “rescue” 1L is at SOFR+950 PIK.
AllClear Aerospace & Defense. Two pairs (60-cent and 58.8-cent gaps). June 9, 2025 “comprehensive recapitalization” — senior lenders took majority ownership; CEO demoted; prior sponsor’s residual equity extinguished.
Florida Food Products. 50-cent gap. October 2025 amend-and-extend — $92M new 1L plus extension of >$750M, 100% existing-lender support, Paul Hastings on the lender side, Weil on the company side.
Plus Visual Edge Technology, page 38, where there’s a four-deep ladder: 8.85% cash → 10.85% partial PIK → 15.85% full PIK → 15.85% PIK, all maturing 01/2029, all labeled “First lien.” Four tiers of internal subordination, one disclosure label.
Plus the Bain pattern — token cash 1L plus giant PIK preferred at par — across AthenaHealth, Himalaya / BCPE Hyperlink, Highstreet Insurance Partners (the cleanest of the four; insurance brokerage is sticky), and Cornerstone OnDemand.
Plus Apex Service Partners — the laundered counterfactual. The 14.25% PIK rescue tranche that ARCC carried in 2022-2023 disappeared from the SOI after Alpine Investors closed a $3.4B single-asset continuation fund in October 2023 (PEI Secondaries Deal of the Year, Americas). New term loan, lower coupon, no PIK in sight. Same operating company.
Pattern, not exception. We met Apex in my first dive into Cliffwater marks, and at this point the modus operandi is getting tired.
What’s already exposed
Three names was the floor. Three was the bare minimum management could publish without being caught lying. The rest of the list — the names where distress is already on the public record, in the 10-Q itself, in court filings, in administrator appointments — got bucketed below “high risk.”
Sponsor already wiped or company already taken by lenders:
Pluralsight — Vista wipeout, August 2024 Kellermeyer Bergensons — Cerberus exit March 2024; ARCC non-accrual August 2025 AllClear Aerospace & Defense — senior lenders took majority, June 9, 2025 Eagle Football — Cork Gully appointed administrator March 27, 2026; Botafogo Recuperação Judicial April 22, 2026; Crystal Palace 43% sold July 24, 2025 VPROP / V SandCo — emerged from Vista Proppants Ch11 in 2020; the 2024 “rescue” 1L at SOFR+950 PIK is the workout, not new lending Florida Food Products — October 2025 amend-and-extend, 100% existing-lender support Cornerstone OnDemand — AlixPartners retained for debt restructure, restructuring within 6 months Already on non-accrual in the same 10-Q ARCC just filed:
New in Q1 2026: Avalign Holdings, Team Acquisition Corp, Convey Health Solutions, Bamboo Purchaser.
Continuing from prior periods: Pluralsight, Kellermeyer Bergensons, Daylight Beta (marked 8 cents), Absolute Dental (three tranches), Pathway Vet Alliance, BAART Programs, KBHS / Alita Care, PS Operating, Continental Acquisition, Florida Food Products (legacy tranche), Teasdale Foods, VPROP / V SandCo.
Already written down to a fraction:
PCG-Ares Sidecar — 11.1 cents on cost. The only PIV ARCC will admit is impaired. Visual Edge Technology — sub note write-off Q3 2023 ($48M loss already booked); the four-tranche AAL ladder of “First lien” loans is what’s left. Quietly marked down 10+ cents in Q1, but not on non-accrual (the proactive bucket from section 2):
Cornerstone OnDemand full stack — 1L revolver, 1L term, 2L all between −16 and −19 cents Symplr Software — every one of six tranches, −14 to −17 cents ATI Restoration — all four 1L tranches, 100c → 87c CoreLogic 2L — −7.6 cents, page 6 Cloud Software Group 2L — −8.1 cents, page 6 EP Purchaser (TPG entertainment) — 73c → 63c That’s at least 30 names with documented distress, written-down marks, executed restructurings, or non-accrual flags on the public record before the call took place. Management called three of them “high risk.” The other 27+ got bucketed below.
That’s not an oversight. That’s the bare minimum.
- The “First lien senior secured loan” label.
Seven words. Every credit investor reads them as a single thing: senior, secured, first claim on the collateral.
The 10-Q proves the label can mean two different things at the same borrower.
ARCC knows how to disclose AAL structure. They do it for some borrowers. CEB Acquisition gets explicit “First Out” and “Last Out” suffixes, page 27. Denali Holdings, Sundance, BAART use the word “Unitranche” in the description. The technique exists in the same filing.
For Pluralsight, Kellermeyer, Eagle Football, VPROP, AllClear, Florida Food, Visual Edge — they don’t.
I’ve argued for months that the “unitranche” label oversells safety. That argument stands. This is bigger. This is one loan literally subordinated to another loan in the cash-flow waterfall, both filed under the same words.
2L is being filed as 1L. The mark dispersion is the strongest public signal that the priority isn’t equal — the AAL itself is private, but a 94-cent same-day same-maturity gap on identically-labeled paper is hard to produce any other way.
That isn’t unitranche optics. That’s the disclosure label and the cash-flow priority describing different things.
This reeks of fraud.
The Fraud Detective The pricing model, in plain sight
The 94-cent gap on same-borrower paper does more than fail at disclosure. It tells you exactly how these positions get priced.
A bond in the public market trades. Every day a buyer and a seller agree on a number, and the number reflects three things: probability of default per year, recovery in default, and carry — the cash interest collected between today and the end. Those three things produce a price, and the price moves when any of them moves.
That is not how ARCC marks Pluralsight.
ARCC takes a spreadsheet enterprise value for the borrower — what someone in a dark room thinks the company is worth — and pours it down the cap stack in waterfall order. First-out gets paid first to the spreadsheet limit, then last-out, then equity, until the spreadsheet runs out. Whatever is left over is the mark. I walked through this with a Lincoln professional earlier this year. He explained the mechanics in detail. He seemed pleased with how the math worked. It’s an open secret in the industry. It is still shocking to see the same plumbing carrying $300M of paper to a real operating company in the largest BDC in the world.
It’s bullshit logic built on bullshit logic.
It is trust on trust. I don’t trust. Fuck out of here with your trust.
The spreadsheet EV is a guess. Private companies don’t trade. There is no last print. Someone picks an EBITDA multiple and types in a number.
The waterfall is also a guess. The AAL is a private contract. The model assumes it works the way the lender says it works, until a workout proves otherwise.
There is no probability of default. No recovery distribution. No carry adjustment for PIK — and PIK is the thing that breaks carry. If the borrower can’t pay you cash today, you do not have carry. You have an IOU stacked on top of another IOU.
A real credit is priced PD × loss-given-default plus carry to maturity. ARCC’s are priced spreadsheet ÷ waterfall. That’s not a market. That’s a model. The model and the market agreed on Pluralsight at 84 cents in Q4. They agreed on 6 cents by Q1. The market got there first. The model is just catching up.
Markets don’t work like that. Models do. And every same-borrower mark dispersion in the SOI is the model’s fingerprint.
One more thing. Third-party valuation firms sign off on these marks. They use ARCC’s own inputs — the spreadsheet EVs, the AAL waterfalls, the borrower-reported EBITDA. Garbage in. Audited garbage out.
- “Markdowns this quarter were preemptive.”
The “What’s already exposed” list above is the answer. KBS — Cerberus exited March 2024. AllClear — lenders took majority June 2025. Pluralsight — Vista wipeout May 2024. Eagle Football — administrator March 2026. Florida Food — October 2025 amend-and-extend with 100% existing-lender support. VPROP — emerged from Chapter 11 in 2020 and got refinanced in 2024 at SOFR+950 PIK, which isn’t new lending — it’s a workout dressed up as a deal.
The marks aren’t preemptive. They’re catching up.
What ARCC Doesn’t Show You at All
Two vehicles inside the 10-Q absorb capital and emit no position-level disclosure.
Ivy Hill Asset Management, wholly owned by ARCC, $12.5B of AUM. ARCC carries it at $2.66B fair value — $1.9B Member interest equity plus a $762M subordinated revolving loan. The sub revolver took a $232M new par draw in Q1. Underlying portfolio composition is not disclosed. No condensed schedule. No sector buckets. If IHAM’s mix mirrors ARCC’s direct book, implied software exposure inside it is about $3.6B — first-loss equity over a $12.5B portfolio. Unverifiable.
SDLP — the Senior Direct Lending Program JV with Varagon. $4.4B of 1L loans across 45 borrowers. The condensed schedule names the borrowers above 5% of capital and buckets the rest as “Other.” One of those 45 loans went non-accrual in Q1. The certificates moved 4 basis points because the bad apple disappears into the JV waterfall. ARCC won’t name the loan.
PCG-Ares Sidecar is already marked at 11.1c of cost, page 28. ARCC has shown they apply real markdowns to PIVs they admit are impaired. They don’t apply them to IHAM.
Closing
I look at the filings because they tell the story. It’s supposed to be the point of truth. In a world of lazy nepohires and copy-paste bankers, this is my edge.
The book is still marked at fictions. But since the 10-Q has to print Q4 next to Q1, the walkdown shows the direction because the math ties. That’s the one thing they can’t omit. Everything else they choose — what to write, what to bucket, which consultant to hire, what to call “spread widening,” whether the word “organic” survives between calls. They cannot hide where this is going.
The next four to six quarters will print more walkdowns. Cornerstone has AlixPartners in the room. Vista Higher Learning sits at 98 cents and looks a lot like Pluralsight did a year ago. Ivy Hill keeps drinking capital. SDLP keeps hiding non-accruals in the certificate math. The PIK engine keeps booking income nobody pays.
Each one will be called isolated when it prints. Each will get a “preemptive markdown” headline. Each will be defended through the next tranche of PIK. They have the story straight. The book isn’t.
Fluent credit on the call. Consultants by the hour. The 67-quarter streak recited like a prayer. They could not tell you what one of these positions is actually worth without the spreadsheet — and the spreadsheet’s inputs are guesses.
What kind of public bond trades at 6 cents while its same-borrower, same-lien, same-maturity twin, trades at par?
None.
The 10-Q says so on page 13.
#analyst/nick-nemeth #thesis/fraud #sector/private-credit #short