Caris Life Sciences, Inc. [CAI] — Financial Model Notes

Phase Space Research · investment-memo v1.4.2 · Task 2 · 2026-07-28 Workbook: CAI_Model.xlsx — 6 tabs, live formulas throughout

Tabs

THE TIE-OUT (v1.4.2 rule — a model that balances but does not tie is confidently wrong)

The lesson from the inherited TWST model was that a zero balance check verifies internal consistency, not input accuracy. So every actual-year line here is reconciled to the filed statement, on the Income Statement tab, in a visible block — not asserted in prose.

SUM OF ABSOLUTE TIE-OUT ERRORS = 0 (live formula, cell Income Statement!D43).

Cost of services is derived, not assumed: CostsAndExpenses − R&D − S&M − G&A. FY2025 = 766,920 − 494,055 = 272,865, giving gross profit 539,168 and a 66.4% gross margin — which reproduces the 8-K's stated $539.2m and 66% independently. Same construction reproduces FY2024's $178.8m and 43%.

Verification — how the model was actually checked

The skill requires reading back computed cells rather than trusting that Python wrote a formula string without erroring. Excel-via-AppleScript hung on a dialog, so verification used formulas 1.3.4, which parses the workbook and independently recalculates the entire formula dependency graph.

That verification found two real bugs, both of which would have shipped silently:

1. Unit mismatch in the revenue build. Assumptions row 30 (molecular profiling revenue) is already in $000s — thousands of cases × dollars per case. The Income Statement multiplied it by 1,000 again, producing FY2026E revenue of $932,960 million. The balance check was unaffected. Caught only by reading the value.
2. Three off-by-one row references on the Balance Sheet and Cash Flow, caused by blank spacer rows: Total current liabilities summed rows 17:18 instead of 16:17; TOTAL LIABILITIES started one row late; Closing cash summed 17+18 instead of 16+17. The balance check was −$30.5m in FY2026E and −$1,013.6m in FY2031E before the fix.

Post-fix read-back:

The read-back is not ceremonial. It is the only control that caught either bug.

Key modelling choices

Revenue is built bottom-up from volume × ASP, not from a growth rate. Molecular profiling revenue = clinical case volume × blended clinical ASP. This construction was validated exactly against Q3-2025 (50,763 cases × $4,089 = $207.57m vs $207.587m reported), which is also what makes the retired ASP disclosure recoverable.

The prior-period true-up is an explicit, separate row (Assumptions row 42), not buried in revenue:

FY2023–FY2025 are the filed 10-K figures. Forecast years are house estimates and are labelled as such: Q1-2026 alone booked $10.3m, so FY2026E of $20m assumes the run-rate roughly halves through the year, then normalises toward a ~$5m steady state as the payer mix matures. This row drives the memo block showing operating income and gross margin ex true-up.

Zero cash taxes across the whole forecast. A $2,548.7m accumulated deficit with a full valuation allowance; FY2023/24/25 income tax expense was $0 in each year. This is worth real money in the DCF and is not an aggressive assumption — it is the filed position.

Stock-based compensation is NOT added back to free cash flow. At 7.5% of FY2026E revenue it is a real cost of employing this workforce. Adding it back would overstate DCF value by roughly $2.20/share.

WACC 11.52% — beta 1.39 measured over 277 daily observations against SPY since the IPO (the full available history), risk-free 4.70% (the same 3M T-bill rate as the book's cash hurdle, so the hurdle and the discount rate are consistent), ERP 5.0%, pre-tax cost of debt 10.0% post the Q1-2026 refinancing with no tax shield because of the valuation allowance. FY2025 cash interest was $44.0m on ~$378m ≈ 11.6%, so 10.0% reflects the refinancing rather than history.

The Balance Sheet forecast plug is labelled, not hidden. Row 22 (Other long-term liabilities) is a residual in forecast years so the sheet balances by construction; FY2024A and FY2025A are the filed figures. The plug's size as a percentage of total assets is displayed on row 36 so it can be seen rather than discovered.

Terminal value is 75.2% of DCF enterprise value. High, and disclosed rather than smoothed. It is why the DCF carries less weight in the final verdict than the scenario analysis — the same treatment applied to GH and TWST.

Scenario construction and the base-rate check

Base rate (Chan, Karceski & Lakonishok 2003 — growth persistence beyond chance is near-nonexistent). The Base 15.5% five-year CAGR from a ~$1bn base is roughly top-quartile, not top-decile. That is deliberate: Caris grew 97% in FY2025 and guides 23–26% for FY2026, so the Base already assumes sharp deceleration rather than extrapolating the recent rate — which is the modal modelling error the literature identifies.

The Bull 23.5% CAGR IS a top-decile outcome at this revenue scale and is named as such. It additionally requires Caris Detect — launched 2026-07-01, unreimbursed — to reach scale. It carries 20%.

Exit multiples are cross-checked on earnings rather than left as bare revenue multiples. The Base 3.6x revenue equals 21.8x EBIT for a mid-teens grower, which is fair rather than punitive — and it is below the 3.9x FY2026E multiple at which CAI trades today. This matters for calibration item B8: on this name the rejection is demonstrably not an artefact of a conservative exit multiple.

What the model does not do

• No Caris Detect revenue in the Base case. The MCED product has no reimbursement and launched three weeks before this memo. It is unpriced optionality in the Bull case only.
• No M&A, no capital raise. Net cash is +$441m and CFO is positive, so neither is required in the forecast.
• Pharma R&D revenue is taken at the guided $75–85m for FY2026 despite that segment having fallen 28.2% in FY2025 ($63.1m → $45.3m). The guide implies +65% to +88% growth off a shrinking base. The model follows guidance rather than substituting a house view, but this is flagged as the most aggressive single component of the FY2026 guide and is stress-tested in the Bear case.
• No quarterly build. The model is annual; quarterly analysis (ASP, DSO, true-up series) is done in the research document from the filings directly.