The entire bull case rests on one disputed number — is loan-loss coverage at 15.4% analytical or accounting?

• Coverage flat, book exploding. Expected-credit-loss coverage held at exactly 15.4% in both FY24 and FY25 while the gross portfolio grew 56% and the loan book grew 77%. Bulls call it pricing skill; bears call it a denominator artifact that lags the cycle.
• Q1 broke the spell. March-quarter expected-credit-loss expense jumped 31% QoQ to $1.72B; the 15-90 day NPL bucket spiked 89 bps to 5.0%; EPS missed at $0.18 vs $0.20. The stock has fallen 36% from its Jan 29 peak.
• One print decides it. Q2 results the week of Aug 13–14, 2026 are the falsifying event. Provisions ≤$1.4B with stable 90+ NPL leaves the model intact; ≥$1.6B with a forced coverage uplift would re-rate ROE expectations toward incumbent levels.

A 30% ROE digital bank growing 27% — earned the multiple, now defending it.

Mature cohorts earn $27 in monthly ARPAC on $0.80 of cost-to-serve — that spread is the entire moat, and it widens as the existing book ages. The pressure comes from two costs moving in the wrong direction together: cost of risk (18.6% of revenue in FY25) and cost of funding (81% of CDI on $41.9B in deposits). Every 100 bps of funding drift is ~$420M of pre-tax compression.

The market prices Nu's funding cost as a moat — peer disclosure says it's a 160 bps gap.

• Inter funds cheaper. Inter & Co. discloses funding at 65.3% of CDI in its FY25 20-F; Nu funds at 81%. That 160 bps gap on $41.9B in deposits is ~$670M of pre-tax that consensus has not priced in. The brand and scale moat is real; the funding moat is overstated.
• Mexico is a different game. Mexico is 8% of revenue, growing 52% YoY to 6.6M card customers — but ranked #3 behind BBVA (10.7M) and Banamex (9.2M), and Mercado Pago, not the incumbents, sets the competitive price. If Mexico settles at 18–22% ROE rather than 30%, group ROE drifts to 24–27% and the multiple compresses with it.
• The cost moat is the real one. 14,000 customers per employee vs Itaú 1,234 and Bradesco 850. Efficiency ratio 19.9% vs incumbents 39–49%. The cost-to-serve advantage is structural and observable in the filings; the funding advantage is not.

Three departures, one giveback, and a wholesale rebuild of credit oversight — all before the quarter that broke the chart.

• Berkshire walked. After trimming 19% in Q3 2024 and 54% in Q4 2024, Berkshire zeroed its position in Q1 2025 (disclosed on a 13F/A filed Aug 14, 2025). The most-cited validator is no longer a holder.
• Founder cashed $500M. David Vélez sold 33M Class A shares for ~$500M via Rua California on Aug 18, 2025 — nine months after voluntarily terminating his 2021 contingent share award (saving Nu $356M in non-cash comp). Two opposite signals; the later one is louder, and he still controls 74.4% of votes on 18.6% of economics.
• The credit chief is gone. President/COO Youssef Lahrech — ex-Capital One, six years running Nu's credit function — departed May 21, 2025. Vélez now personally owns credit oversight, at the exact moment cost of risk turned.

One print decides the next twelve months — everything else is positioning around it.

• Q2 2026 print — Aug 13–14. Consensus EPS $0.22 on revenue ~$5.42B. Provisions ≤$1.4B with stable 90+ NPL closes the cost-of-risk debate; ≥$1.6B with a forced coverage uplift sets up a re-rate toward incumbent P/B near 3×.
• Selic cuts begin. Brazil's first cut in five meetings landed March 18, 2026: 15.00% → 14.75%. Next Copom decisions June 17, Aug 5, Sep 16. Every 25 bps is funding relief for Nu but NIM compression for the sector.
• US bank charter clock. OCC granted Nubank, N.A. a conditional national-bank charter on Jan 29, 2026. Capitalization due by Jan 29, 2027 and operational launch by July 29, 2027 — the smallest near-term event with the largest optionality.