Overview

Caris is the cheapest growth name in diagnostics — 3.9x FY2026E revenue against a 7.8x peer median — with genuinely real underlying growth (+90.6% in FY2025 excluding the true-up), real cash generation (+$83.2m CFO), $441m of net cash and clustered, non-routine insider buying as recently as 2026-07-13 at $14.56. Against that: 74.5% of its first-ever positive GAAP operating income is a prior-period revenue true-up, its blended ASP fell 25.4% sequentially in Q1-2026 on flat volume, and it retired the ASP disclosure table in the very quarter ASP peaked. Those two sides very nearly cancel. Probability-weighted value is $16.41 against a $15.59 spot: expected return +5.3% versus a 4.7% cash hurdle, a margin of 0.57 percentage points, with the flip point at a 30.6% bear weight against the 30.0% used. The honest conclusion is that at $15.59 the evidence does not determine the answer — the framework is measuring rounding error in its own judgement inputs. It does determine the answer at $13.13, where expected return is +25%.

The six gates

Key findings

• RESOLVED IDENTITY: CAI is Caris Life Sciences, Inc. (CIK 0002019410), a Texas-incorporated precision-oncology molecular-profiling laboratory that IPO'd on Nasdaq at $21.00 on 2025-06-18 and dual-listed on NYSE Texas on 2026-06-17. It is currently filing and currently trading. The ticker previously belonged to CAI International (container leasing) and was reassigned at the IPO.
• THE CENTRAL FINDING: 74.5% of FY2025's first-ever positive GAAP operating income is a prior-period revenue true-up. Reported operating income was +$45.1m; the audited 10-K revenue note discloses $33.6m of 'adjustments to revenue related to services delivered in prior periods'. Those carry ~100% incremental margin because the cases were costed in earlier periods. Clean FY2025 operating income is +$11.5m.
• The same adjustment made Q1-2026 loss-making on a clean basis: reported operating income +$5.3m less a $10.3m true-up = -$5.0m. The true-up is accelerating, not normalising: $(1.9)m (FY2023) to $3.9m (FY2024) to $33.6m (FY2025) to $10.3m in Q1-2026 alone, versus $3.9m in Q1-2025.
• The figure was derived TWICE, independently. Method 1: solving the company's own FY2026 guidance sentence ('excluding out-of-year revenue from over collections recorded in 2025, this range implies growth of approximately 26% to 28%') gives $30-36m. Method 2: the audited 10-K revenue note gives $33.6m. The phrase 'out-of-year' appears ZERO times in the 10-K — it exists only in one unquantified sentence in one press release.
• MENTION-FREQUENCY, run generatively BEFORE any view was formed, surfaced the thesis: 'ASP' mentions decayed 76% (46.2 to 11.2 per 10k words) and 'volume' 66% across the four-quarter window. Investigation showed this tracked an actual RETIREMENT OF DISCLOSURE — the 'Summary Financial Results' table carrying per-product ASP and volume appears in the Q2-2025 and Q3-2025 releases and is ABSENT from Q4-2025 and Q1-2026 (verified 1, 1, 0, 0 by string search).
• Caris retired the ASP disclosure in the same quarter ASP peaked. Reconstructing ASP from molecular profiling revenue divided by case volume (method validated exactly against Q3-2025: 50,763 x $4,089 = $207.57m vs $207.587m reported) shows blended ASP rose $2,138 to $5,353 then fell 25.4% sequentially to $3,992 in Q1-2026 — on essentially FLAT volume (52,700 to 52,800 cases).
• Clean gross margin corroborates independently from two directions: FY2025 reported gross margin 66.4%, ex the $33.6m true-up 64.9%; Q1-2026 reported gross margin 65.0%. The run-rate margin of this business is ~65% and the reported FY2025 figure is flattered.
• AND YET THE NAME IS CHEAP. EV is $3.97bn (market cap $4.41bn less $822m cash plus $381m debt) = 4.37x TTM revenue and 3.93x FY2026E guidance. Peer median is 7.82x; the growth-diagnostics median is 13.26x. CAI trades at half TXG's multiple and a third of TWST's, while growing faster than both, with better margins than both, and generating cash unlike either.
• INSIDER BUYING is clustered, non-routine and recent. Jon Halbert bought 68,000 shares at $14.56 on 2026-07-13 (~$990k) and Jeff Vacirca 31,050 at $16.15 on 2026-05-12 (~$501k), neither under a 10b5-1 plan. The only open-market sale across all 39 Form 4s since the IPO is a single 10b5-1 disposal; everything else is tax withholding on vesting. Under Cohen/Malloy/Pomorski 2012 this is the textbook positive configuration and it is the strongest argument against the cautious view. A $100m buyback was authorised 2026-06-08.
• GATE 4 IS THE STARKEST C4 CASE IN COVERAGE. Expected return is +5.3% against the 4.7% hurdle — a 0.57pp margin — with an E[R] range of -11.3% to +23.0% and a flip point at a 30.6% bear weight versus the 30.0% used. That is 0.6 percentage points. GOOGL, the first C4 name, flipped at 12pp. The framework is not measuring a signal at this price; it is measuring rounding error in an unverifiable judgement input.
• GATE 5: CAI OPTIONS ARE UNINVESTABLE. 166 contracts exist (Alpaca's discovery endpoint returns zero, but it also returns zero for NTRA — the endpoint is broken, not the entitlement; snapshots return full chains). Median bid-ask spread across 50 near-the-money contracts is 85% of mid, worst 173%, with near-zero volume and put-call parity violations up to $0.95. Equity is the only viable vehicle. Per the skill's gotcha, IV was recomputed by Black-Scholes from mid prices rather than trusting Alpaca's field — the two diverge sharply on the wide near-dated strikes exactly as documented.
• PAYER CONCENTRATION IS DETERIORATING SHARPLY: the top three payers rose from 58.1% to 70.1% of revenue in one year (Payer 1 37.4%, Payer 2 21.1%, Payer 3 11.6%). Payer 1 is almost certainly Medicare/MolDX. One rate action reprices the entire ASP series — this is the mechanism by which the Bear case happens.
• TAM SATURATION IS THE BURIED PROBLEM. Bottom-up, the US therapy-selection TAM is ~$1.86bn/yr (2.04m new diagnoses x 35% CGP-eligible x 65% profiled x $3,992 realised ASP). Caris ran ~211,000 clinical cases in the twelve months to Q1-2026, which is ALREADY ~45% penetration. The FY2031 Base case of $2.08bn therefore cannot be a penetration story — it is implicitly a bet on Caris Detect and MRD, products launched three weeks ago with no reimbursement.
• B1 IS FINALLY TESTED, AND THE ANSWER IS NOT WHAT B1 PREDICTED. CAI trades 28.7% BELOW its 200-day moving average with -31.6% 12-1 momentum — the deliberately-below-the-200-day sample the calibration file said was required and that TXG (+102.8% above) and TWST (+68.3% above) did not provide. Gate 6 is a genuine long-side headwind here for the first time in fifteen names. But Gate 6 is NOT the binding gate: delete it entirely and CAI is still a Watchlist, because Gate 4 and Gate 2 fail independently. First real evidence on B1, and it points against B1 being the binding structural flaw.
• B8 IS PARTIALLY RESOLVED — by exclusion. B8 needed 'a healthcare name with clean earnings quality to discriminate'; CAI does not have clean earnings quality, so it cannot be that name. But it discriminates on the other axis: the Base exit multiple of 3.6x FY2031E revenue is BELOW the 3.9x at which CAI trades today, and cross-checks at 21.8x EBIT. The rejection here is demonstrably NOT a conservative-multiple artefact, which removes CAI from the B8 pattern rather than adding a fifth instance to it.
• NEW CALIBRATION ITEM C9: Sloan accruals are structurally blind to cash-settled revenue timing, and move in the WRONG direction. Accruals read -20.6% (excellent quality) precisely BECAUSE the prior-period true-ups arrive as cash, inflating CFO. The single largest earnings-quality defect in this name makes the headline earnings-quality factor look better. This is the mirror of the CIEN S5 finding.
• NEW CALIBRATION ITEM C10: a 0.6pp flip point means the verdict is under-determined, not concluded. Recording it as 'Gate 4 FAIL' reads as a substantive finding when it is closer to 'insufficient resolution to decide'. GOOGL (12pp) and CAI (0.6pp) are currently recorded identically and are not the same epistemic object.
• THE STREET IS FAR ABOVE SPOT AND FRESHLY SO. All five most recent targets are Buy and dated July 2026: BofA $23, Evercore $25, Goldman $27, TD Cowen $30, BTIG $32; consensus ~$27 across 13 analysts. The stock at $15.59 trades 32% BELOW the lowest published target. The required decomposition shows the gap is entirely the multiple — the house is +1.3% above consensus on FY2026 revenue and +62% on EPS.
• MODEL VERIFICATION FOUND TWO REAL BUGS. Independent formula-graph recalculation (the `formulas` library, after Excel-via-AppleScript hung) caught a revenue unit mismatch that produced FY2026E revenue of $932,960 MILLION, and three off-by-one row references that left the balance check at -$1,013.6m in FY2031E. Both would have shipped silently; the balance check was unaffected by the first. Post-fix: tie-out sum of absolute errors to the filed 10-K = 0, balance check = 0.0000 in every forecast year.

Sections

Disclosed limitations

• MENTION-FREQUENCY WINDOW IS FOUR QUARTERS. Caris IPO'd 2025-06-18, so four quarterly releases are its entire public history. That is simultaneously 'all public history' and a very short series — four points cannot establish a trend with confidence. Every claim is stated as 'first in the four-quarter window examined'; NO first-ever claim is made anywhere in this memo.
• TRANSCRIPT CORPUS IS EARNINGS RELEASES, NOT CALL TRANSCRIPTS. Alpha Vantage's transcript endpoint was unavailable (shared 25/day quota exhausted at ~14:10 PDT 2026-07-28), so the pinned corpus is SEC 8-K EX-99.1 narrative sections — first-party, complete and near-constant length, which references/mention-frequency.md explicitly endorses. Consequence: prepared-remarks share is 100% and the analyst Q&A dimension is entirely absent. One source used across the whole series, never mixed.
• QUARTERLY SPLIT OF THE FY2025 TRUE-UP IS NOT PUBLISHABLE. The 10-K's annual $33.6m measures adjustments to services delivered in prior YEARS; the Q3-2025 10-Q's $38.3m measures services delivered before 1 July 2025 and therefore includes intra-2025 catch-up. The two bases are not additive, and naive subtraction produces a spurious negative Q2-2025. Attempted and deliberately withheld rather than published wrong.
• EARNINGS-SURPRISE (SUE) AND ESTIMATE-REVISION FACTORS COULD NOT BE COMPUTED. No free source gives point-in-time consensus as of the Q1-2026 print, and Alpha Vantage was exhausted. Two rows of the Factor & Anomaly Scorecard are flagged blank rather than silently omitted or guessed.
• PAYER 1/2/3 IDENTITIES ARE NOT DISCLOSED. The inference that Payer 1 (37.4% of revenue) is Medicare/MolDX is based on the coverage narrative in the 10-K and is labelled an inference, not a sourced fact.
• Q4-2025 AND Q1-2026 ASP FIGURES ARE DERIVED, NOT DISCLOSED — precisely because the company retired the ASP table in Q4-2025. The derivation method was validated exactly against Q3-2025, where the table still existed, but the derived figures carry reconstruction risk if the mix of cases within 'clinical therapy selection' changed.
• TEMPUS AI [TEM] COMPARABLE IS APPROXIMATE. Share count is web-sourced and one quarter is absent from its XBRL company-facts, so its EV/revenue of ~6.4x is a reasonable estimate rather than a computed figure. Flagged in the comps table itself.
• THE Q4-2025 TO Q1-2026 ASP COLLAPSE IS NOT FULLY EXPLAINED. The prior-period true-up was ~$10.2m in Q4-2025 and $10.3m in Q1-2026 — essentially identical — so it does NOT account for the 25.4% sequential ASP decline. Q1 seasonality (deductible resets) and the January sales-force re-alignment are real partial explanations offered by the company, but the residual is unexplained and is the single most important thing to resolve at the Q2-2026 print.
• THE DCF CARRIES LOW WEIGHT: terminal value is 75.2% of enterprise value, and the WACC-by-terminal-growth sensitivity spans $9.66 to $19.83 with spot inside that band. The scenario analysis, not the DCF, carries the verdict — the same treatment applied to GH and TWST.
• SCENARIO PROBABILITIES ARE JUDGEMENT AND, ON THIS NAME MORE THAN ANY OTHER, THEY CARRY THE ENTIRE RESULT. The flip point is 0.6pp from the weight used. This is calibration item B7 in its most acute form and is the reason the verdict is characterised as under-determined rather than concluded.
• CAI DID NOT CLEAR THE SCREEN. It does not appear in Screen_2026-07-27 and would have been filtered at the momentum/trend stage. Per calibration item B3 this is exactly the population the funnel is designed to exclude; the name arrived as a user-named ticker and has no base-rate support.
• NO POSITION IS RECOMMENDED, so no entry appears in trade_recommendations.jsonl.