On January 28, 2026, Gotham City Research published a short report on Carvana (CVNA) containing FOIA-obtained financial statements from DriveTime and GoFi LLC — two private entities controlled by the Garcia family that we hadn't previously analyzed in depth.
We took this new evidence back to our LLM committee (Opus 4.5, Gemini 3, ChatGPT-5.2, Perplexity) for a targeted Q&A session. The results were humbling: our January 7 analysis was directionally correct but significantly underweighted the scale and interconnectedness of the risk.
This post documents what we got right, what we missed, and — most importantly — how we're improving our methodology based on this case study.
Classification Upgraded
Investor Posture: Changed from HIGHER_SCRUTINY to AVOID. New investors should wait for: (1) DriveTime deleveraging, (2) auditor change with clean transition, or (3) transparent disclosure of servicing fee economics.
Key Claims from Gotham
Gotham obtained DriveTime's 2024 Annual Report and GoFi's 2024 financials via FOIA request. The key allegations include:
DriveTime Cash Burn: $1B+
DriveTime burned $324M (2023) and $708M (2024) in operating cash flow while CVNA booked ~$755M in gain-on-sale. DriveTime funded this with $1.4B in debt.
DriveTime Leverage: 20-40x
Net debt / Adj EBITDA reached 40.2x (2023) and 18.6x (2024) vs historical cap of 10.3x. Interest coverage collapsed to 0.5-1.0x from historical 2.3x+.
Same Auditor Across All Entities
Grant Thornton audits CVNA, DriveTime/Bridgecrest, AND GoFi. GT also gave clean opinions to Tricolor Holdings during its multi-year fraud (executives now indicted).
Servicing Fee Arbitrage (0.117% vs 2-3%)
"Third party" loan buyers pay 0.117% servicing fees vs 2-3% market rate. Gotham theorizes this subsidizes CVNA's gain-on-sale — an order-of-magnitude discount that's economically irrational unless it's a related-party subsidy.
Garcia Extraction: $352M
Garcia II pulled $352M in distributions from DriveTime while it burned $1B+ in cash and raised $1.4B in debt — insider enrichment while the related party deteriorated.
What We Got Right
Our January 7 analysis correctly identified several key risks that the Gotham report validates:
Gain-on-Sale Timing Optionality
We flagged that management has discretion over when to recognize loan sale revenue. Gotham's evidence confirms this is even more structural than we realized.
Governance Misalignment
We identified 84% voting control, non-arm's-length DriveTime transactions, and the TRA directing 85% of tax benefits to insiders.
Level 3 Fair Value Concerns
We noted the beneficial interest valuations were unobservable inputs. DriveTime's $900M loan markdown concurrent with CVNA's gain-on-sale validates this concern.
Securitization Dependency
We identified ABS market access as critical. The ecosystem's hidden leverage makes this dependency even more acute.
'Aggressive but Legal' Framing
We appropriately avoided crying 'fraud' while still flagging elevated concerns. This measured approach remains correct pending regulatory action.
What We Missed
The Gotham report revealed significant gaps in our analysis:
DriveTime as Funding Source, Not Just Governance
We treated DriveTime as a governance/related-party concern. It's actually a funding mechanism — DriveTime's debt is hidden leverage supporting CVNA's earnings.
Cross-Entity Auditor Concentration
We didn't flag that Grant Thornton audits all three Garcia entities. When the same auditor reviews both sides of related-party transactions, independent challenge is reduced.
Auditor Track Record (Tricolor)
We noted GT as 'mid-tier' but didn't investigate their track record. GT's recent miss at Tricolor Holdings (same industry, similar structures) is highly relevant.
Servicing Fee Mechanism
We didn't analyze the servicing fee economics. The 0.117% vs 2-3% discrepancy is an order of magnitude that demands explanation.
GoFi as Intermediary
We were unaware of GoFi's role. Gotham suggests it's another node in a circular money flow — 'what DriveTime is to CVNA, GoFi is to DriveTime.'
Scale of DriveTime Leverage
We knew DriveTime existed but didn't have their financials. 20-40x leverage vs historical 10x cap is a structural break, not normal variation.
Methodology Improvements
Based on committee consensus, we're proposing these additions to the Fugazi Filter:
Related-Party Funding Dependency Signal
New signal measuring earnings dependence on unconsolidated related parties. Triggers when >25% of EBITDA flows from related party transactions. If the related party's financials are opaque, assume Critical Fragility.
Auditor Concentration Risk Check
Flag when same auditor serves multiple interconnected entities under common control. Aggravated by: mid-tier auditor, recent fraud miss at comparable company, complex related-party flows.
Servicing Fee Benchmarking
In Stage 1/2, require benchmark comparison of implied servicing rates vs market. Flag when rates deviate >50% from market — this would have caught the 0.117% anomaly.
Cross-Entity Cash Flow Divergence
When subject company CFO improves while related party CFO deteriorates, flag as potential 'hidden leverage' pattern. DriveTime's $1B burn concurrent with CVNA's improvement is exactly this pattern.
New Monitoring Triggers
Based on the Gotham allegations, we've added these immediate monitoring triggers for CVNA:
| Event | Action |
|---|---|
| Grant Thornton resignation announced | Immediate re-analysis |
| 2025 10-K filing delayed | Immediate re-analysis |
| Any restatement of gain-on-sale or related party | Immediate re-analysis |
| DriveTime leverage >25x | Elevate fragility |
| Servicing fee disclosure in any filing | Verify market alignment |
What This Teaches Us
The Gotham report is a valuable stress test of our methodology. We got the direction right — ELEVATED concerns across both axes — but underestimated the magnitude and interconnection. The key lesson: related parties can function as funding mechanisms, not just governance concerns, and single-auditor oversight of an ecosystem reduces the independent challenge that would normally catch these patterns.
We've updated our CVNA analysis page with the new classification and findings. The methodology improvements will be incorporated into the Fugazi Filter v1.2 and applied to future analyses.
This is the value of transparent, multi-model analysis: when new evidence emerges, we can systematically update our views and share exactly what changed. No one gets every call right. The goal is rigorous process, honest assessment, and continuous improvement.