Finance lens

Most M&A teardowns are written at announce, the analyst restates management's deck and adds a one-line view. By the time enough information is on the table to test the case, the news cycle has moved on. The COF/DFS combination is unusual: it closed in May 2025, twelve months ago, so a full year of post-close disclosure now sits next to the announce-day framing. The S-4 says one thing. The Q2 2025 10-Q says another. The first four post-close earnings transcripts say a third. Putting all three on the same page is the work.

Engineering lens

The artifact stack mirrors the editorial pattern of the earnings-digest project: a Next.js + Tailwind single-page site for the narrative, a two-page PDF rendered with WeasyPrint from styled HTML for the recruiter-shareable note, and an openpyxl-generated eight-sheet Excel workbook with color-coded inputs (blue), formulas (black), and cross-sheet links (green), banker conventions that make the model auditable on first open.

Finance lens

Management telegraphed $2.7B in run-rate synergies by 2027 ($1.5B expense + $1.2B network), >15% Y2 adjusted EPS accretion, 16% ROIC, IRR >20%, and a $2.8B integration cost budget. The 26.6% premium against DFS's $110.49 unaffected price (Feb 16, 2024) sits above the post-pandemic U.S. bank M&A average of ~17–20% over 2022–23 (Mercer Capital). All numbers lifted from the joint investor presentation and the S-4, nothing in this section is interpretation.

Engineering lens

Source discipline as a principle: every load-bearing figure on the public site is hyperlinked or footnoted to a filing or a named secondary source (Mercer Capital for the premium benchmark, Banking Dive for the integration-cost trajectory). A "Verification status" section in the README enumerates which claims tie to which filings and where my earlier draft had to be corrected (the "in line with bank M&A average" framing was originally wrong, and the Discover-from-Sears history was off by seven years).

Finance lens

Management's >15% accretion guidance was almost certainly built on a more conservative COF standalone case than consensus would imply. COF reported $4.8B of net income in FY2024, depressed by a $10.9B provision build. Normalizing for credit roll-off implies a 2027 baseline closer to $8.1B. Once the standalone denominator is set there, even the announce-day synergy stack (full $2.7B at 100% phase) only delivers ~13% accretion in the model, not >15%. The framing matters: the contrarian view holds on management's own assumptions, before any Year-1 actuals come in.

Engineering lens

The model is built around an explicit "case toggle" cell on the cover sheet (1 = Mgmt, 2 = Base, 3 = Haircut) that drives synergy run-rate, phase-in, and integration cost across the rest of the workbook. Standalone projections sit as direct inputs on the combined sheet with stated source attribution, so every column on the A/D output sheet has a visible chain back to either a filing or a normalization assumption flagged in the methodology sheet.

Finance lens

The conventional read on a $35B all-stock deal is that pro forma CET1 takes a beating from goodwill and intangibles. The Q2 2025 10-Q tells a different story: $13.2B of goodwill and $13.4B of newly-recognized intangibles ($10.3B in purchased credit-card relationships, $3.1B in network intangibles) were created. DFS's loan book was marked down ~$9B for credit and rate at Day 1 (the same mark that feeds the $7.9B Discover-driven allowance build COF disclosed that quarter). And yet pro forma CET1 came in at 14.0%, up from COF's standalone 13.6% the quarter before. The equity issued to DFS holders added more to the capital base than the deductions consumed. Unusual for a deal this size, and underappreciated in the prints.

Engineering lens

The PPA build in the sources & uses sheet ties to the reported $13.2B goodwill number to within rounding ($13.0B in the model), with the loan-FV markdown line as the reconciling item. Two prior versions of this section had the wrong figures (a $4.5B step-up estimate and an ~11% pro forma CET1 projection); the post-verification audit replaced both with the actual reported numbers and reframed the CET1 line as a positive surprise, which is what it is once you read the 10-Q.

Finance lens

Three cases sit side by side on the workbook and the site. Management's announce-day case ($2.7B run-rate, 100% phased by 2027) yields +13.0% Y2 EPS accretion against the normalized standalone. The Base case ($2.5B "net synergies" by mid-2027, the framing Fairbank used on the Q2 2025 transcript, with a one-year slip) yields +8.6%. My Haircut case ($1.9B, roughly 30% off the announce headline, with another year of slip to reflect OCC remediation drag) yields +3.3%. The accretion compresses as the synergy stack shrinks and the timeline slips. The bridge on the site shows the move from the $2.7B announce headline to the $2.5B Q2 framing to my haircut as a visual; the table on the workbook shows the underlying math.

Engineering lens

The synergy phase-in schedule lives in its own block on the cover sheet so it can be tested independently of the run-rate. Each case has its own phase-in curve (Mgmt 10/50/100/100, Base 5/35/85/100, Haircut 5/25/65/90 across 2025–2028) that flows through the combined sheet and into the A/D output. A 4×4 sensitivity grid on its own sheet maps 2027 accretion across realization × phase-in aggressiveness so the headline numbers can be stress-tested.

Finance lens

The retrospective is built on six pieces of post-close disclosure: a $4.3B Q2 2025 GAAP loss (the first consolidation quarter), $9.4B in total Discover-related charges that quarter, $409M of H1 2025 integration spend against the $2.8B budget, Fairbank's admission that integration cost will run "somewhat higher than $2.8B," a $1.225B merchant restitution settlement that landed post-close on the historical card-misclassification matter, and an open-ended OCC remediation requirement tied to standing Discover Bank enforcement actions. Each one is sized and dated on the site. The contrarian framing is anchored to all six, not to any single headline number.

Engineering lens

Every actual cited on the site or the PDF is sourced to a specific filing or transcript and listed in the "primary sources" block on the resources section. The Verification round of the build replaced an earlier estimate-heavy draft with the verified numbers; the README captures which claims were revised and why, so a reader can audit the work end to end.