Caris Life Sciences, Inc. [CAI] — Valuation, Factor Scorecard and Trade Construction

Phase Space Research · investment-memo v1.4.2 · Tasks 3, 5, 6 · 2026-07-28 · spot $15.59

TRADE TICKET

The one-sentence version. Caris is the cheapest growth name in diagnostics (3.9x FY2026E revenue against a 7.8x peer median) with real 90%+ underlying growth, real cash generation, $441m of net cash and clustered insider buying two weeks ago — and 74% of its first-ever operating profit is a prior-period accounting true-up, its ASP just fell 25.4% sequentially, and it retired the ASP disclosure in the same quarter that ASP peaked. Those two facts very nearly cancel: expected return is +5.3% against a 4.7% hurdle, and the verdict flips on 0.6 percentage points of a probability weight nobody can verify. The honest answer is that the evidence does not determine the answer at $15.59. It does at $13.13.

1. Consensus bridge — required, and it runs the opposite way from usual

The Street

Consensus FY2026 revenue $1.00bn; FY2026 EPS $0.13; FY2027 EPS $0.36.

The stock trades 32% below the LOWEST published target, and every one of those five targets was published this month, after the Q1 print and into the decline. That is an unusually wide and unusually fresh gap.

The required numbers-vs-multiple decomposition

The gap is entirely the multiple, and it is proved rather than asserted. On the fundamentals the house sits slightly above the Street — +1.3% on revenue, and above on EPS. Applying the Street's own ~$27 target to the house's own revenue forecast requires a ~7.2x FY2026E EV/revenue exit, against 3.9x today.

This is calibration item B8 appearing for a fifth time, and the fifth consecutive healthcare name. ISRG (+0.2% above Street), NTRA (+3.9%), TXG (+0.5%), TWST (+0.4%) — and now CAI (+1.3% on revenue, +62% on EPS). In every case the house forecast is at or above consensus and the name is not bought. See §8 for why CAI is the discriminating case B8 has been waiting for — and why the answer it gives is not the one B8 predicted.

The real risk if the Street is right

The Street is underwriting Caris Detect and MRD as genuine option value on a platform with a defensible transcriptome moat, and treating the FY2025 profitability inflection as a durable structural change. If ASP stabilises at ~$4,000 and volume re-accelerates to the 56,000-case run-rate management claims it exited Q1 at, the Base case is too conservative and $27 is reachable. That is not a fringe possibility — it is roughly the Bull case, and it carries a 20% weight. The house is not confident the Street is wrong; the house is saying the price does not pay enough to take the bet.

2. Transcript Mention-Frequency (required table)

Corpus: SEC 8-K EX-99.1 earnings releases, narrative section only — one source across the series, named. Normalised per 10,000 words. Window: four quarters = the company's entire public history (IPO 2025-06-18). No first-ever claims are made.

Prepared-remarks share: 100%. These are company-authored releases with no analyst Q&A, so every mention is unprompted — the stronger signal class, at the cost of losing the Q&A dimension entirely. Stated as a limitation, not hidden.

Decaying → the ASP disclosure table itself was retired (verified: "Summary Financial Results" appears 1, 1, 0, 0). Emerging → the Caris Detect / whole-genome / ACHIEVE cluster, corroborated independently by the 2026-07-01 commercial launch and by ClinicalTrials.gov NCT07680868 (ChromoSeq, registered 2026-07).

3. Factor & Anomaly Scorecard (required)

Synthesis — and the finding that matters

The scorecard SPLITS, and it splits along a clean fault line: every price-based factor is a headwind and every quality-based factor is a tailwind.

That is not noise. It is the market pricing something the accounting-quality measures cannot see.

And here the CIEN "S5" problem appears in mirror image. Sloan accruals read −20.6%, i.e. excellent earnings quality — because Sloan measures (net income − operating cash flow) ÷ assets, and Caris's prior-period true-ups are cash. Over-collections make CFO larger, which makes the accruals ratio more negative, which the anomaly literature reads as higher quality. The single largest earnings-quality issue in this name makes the headline earnings-quality factor look better. Sloan is constructed to detect accrual-based earnings management; it is structurally blind to revenue-timing shifts that arrive in cash.

Gate 1 corroboration must therefore be read with that caveat explicit. This is logged as a new calibration item.

4. Valuation

DCF

Cash taxes are zero across the entire forecast — a $2.5bn accumulated deficit with a full valuation allowance. Stock-based compensation is NOT added back to free cash flow (7.5% of revenue in FY2026E); adding it back would overstate value by roughly $2.20/share.

Sensitivity (WACC × terminal growth) spans $9.66 to $19.83. Spot sits inside that band, which is the honest summary of what a DCF can say about a name whose terminal value is three-quarters of its enterprise value.

Comps

* TEM share count web-sourced and one quarter absent from XBRL — flagged as approximate.

CAI at 4.37x TTM / 3.93x FY2026E is the cheapest growth name in the set, and it has the best operating margin of any loss-making peer (−5.2%). Applying the 7.82x peer median to FY2026E revenue gives $29.14/share; the growth-diagnostics median gives $48.83.

This is uncomfortable and must be stated plainly. This system underwrote TXG at 8.95x and TWST at 13.25x within the last 48 hours and concluded both were 51–55% overvalued. CAI is growing faster than both, has better margins than both, generates cash unlike either, and trades at half TXG's multiple and a third of TWST's. Consistency demands the memo say so: on relative value CAI is the most attractive name in this cluster by a wide margin. The reason it is still not a buy is not the multiple — it is Gate 4 arithmetic on an absolute basis, and the earnings-quality overhang on the profitability that would justify a re-rating.

Scenarios

Probability-weighted value: $16.41 → E[R] +5.3%.

Base-rate check (Chan, Karceski & Lakonishok 2003). The Base 15.5% five-year CAGR from a ~$1bn base is roughly top-quartile, not top-decile — a deliberate choice: Caris grew 97% in FY2025 and guides 23–26% for FY2026, so the Base already assumes sharp deceleration rather than extrapolating. The Bull 23.5% CAGR IS a top-decile outcome at this revenue scale and is named as such; it also requires Caris Detect to reach reimbursed scale, which is unproven, and it is given 20%. The Bear 9.0% assumes ASP mean-reverts as the true-ups stop flattering the base.

Exit multiples cross-checked on earnings, not left as bare revenue multiples: Base 3.6x revenue = 21.8x EBIT for a mid-teens grower, which is fair rather than punitive. This addresses B8 directly — the exit multiples here are not the mechanism of rejection (see §8).

5. THE SIX GATES

Gate 1 — Causal mechanism with quantitative corroboration → PASS (long-side)

Long-side mechanism: a real, cash-generative molecular-profiling franchise growing volume ~15–20% with structurally improving gross margin (~65% run-rate vs 43.4% two years ago), $441m net cash, and a defensible whole-exome + whole-transcriptome moat evidenced by accelerating independent PubMed citation volume (143 → 201 → 281). Quantitatively corroborated: GP/A 0.479, F-score 6/9 (three failures being IPO artefacts), CFO +$83.2m.

Short-side mechanism: rejected. The bearish story (74% of operating profit is a true-up; ASP −25.4%; ASP disclosure retired) is real, but the quantitative scorecard contradicts it — Sloan accruals read −20.6%, i.e. excellent quality, precisely because the true-ups arrive in cash. Per the S5 precedent set on CIEN, a gate should not be passed on a mechanism its own corroboration contradicts unless the contradiction is explained; here it is explained (Sloan is blind to cash-settled revenue timing), but the price has also already fallen 60%, so the short-side mechanism has no remaining edge. PASS long-side, FAIL short-side.

Gate 2 — Negative variant vs consensus → FAIL (both branches)

2A (estimate variant): FAIL. House FY2026E revenue $1,012.9m vs consensus $1.00bn = +1.3%. House EPS $0.21 vs $0.13 = above. There is no negative variant; the house is marginally more optimistic than the Street on the numbers and materially less optimistic on the multiple. A multiple disagreement is not a variant view — that is the founding NET correction, and it applies here in full.

2B (duration/optionality variant): NOT CLAIMED. The four-leg test fails at leg 4. The $33.6m true-up is disclosed in the audited 10-K; management flagged it in the FY2026 guidance; any analyst reading the revenue note finds it. "Consensus has not emphasised it" is not "consensus is wrong by X%." Leg 4 is not satisfiable and is not claimed.

Gate 3 — Catalyst that resolves the specific disagreement → PASS

Q2-2026 earnings, ~6–12 August 2026 (≤2 weeks away). It resolves all three live questions cleanly and on a date: 1. Did ASP recover? Q1-2026 $3,992 vs Q4-2025 $5,353. Reconstructible from MP revenue ÷ case volume even though the table is retired. 2. Did the prior-period true-up normalise? Q1-2026 booked $10.3m. The 10-Q revenue note discloses it. 3. Did volume reach 56,000? Management's own stated Feb/Mar exit run-rate.

Plus first commercial commentary on Caris Detect (launched 2026-07-01). This is precise, dated, and falsifiable — the standard the gate demands.

Gate 4 — Positive scenario-weighted expected return → FAIL (BINDING)

E[R] RANGE: −11.3% to +23.0%. The 4.7% cash hurdle sits INSIDE the range.

FLIP POINT: E[R] equals the hurdle at p_bear = 30.6%, against the 30.0% used — a gap of 0.6 percentage points.

This is the starkest instance of calibration item C4 in the entire coverage. GOOGL — the first name to trigger C4 — flipped at an 18% bear weight against the 30% used, a 12pp gap. Here the gap is 0.6pp. The probability weights are judgement inputs that cannot be estimated to anything like 0.6pp precision. The framework is therefore not measuring a signal at this price; it is measuring rounding error in its own inputs.

The correct reading is not "E[R] is +5.3%, therefore marginally attractive." It is: at $15.59 the evidence does not determine the answer. Gate 4 fails not because expected return is negative but because it does not clear the hurdle by any margin that survives the precision of the inputs. Recorded as a FAIL, with the reasoning stated rather than the number asserted.

Prices at which the question becomes answerable: $14.27 → E[R] +15%; $13.13 → E[R] +25%.

Short side, for completeness: a short at spot earns −5.3% gross before borrow. It fails outright.

Gate 5 — Implementation feasibility → PASS (equity) / FAIL (options)

Options: FAIL, decisively. 166 listed contracts across 7 expirations exist (the Alpaca discovery endpoint returns zero, but it also returns zero for NTRA — the endpoint is broken, not the entitlement; the snapshots endpoint returns full chains).

• Median bid-ask spread on 50 near-the-money contracts (K $12.5–$20): 85% of mid. Worst observed 173%.
• Daily volumes 0–20 contracts across most strikes; open interest not reported.
• Put-call parity violations up to $0.95 — direct evidence the quotes are stale/indicative, not executable.

Per the skill's documented gotcha, IV was recomputed by Black-Scholes from mid prices rather than trusting Alpaca's impliedVolatility field. The two agree on the tighter long-dated strikes and diverge sharply on the wide near-dated ones (Aug-21 $15: Alpaca call 0.436 / put 0.782; recomputed 0.578 / 0.711) — exactly the unreliability the gotcha describes. ATM implied ≈ 86% vs realised 62.7% (252d) / 74.9% (63d): a normal 1.2–1.4x variance risk premium. But the premium is irrelevant — a defined-risk spread paying 85% of mid in friction on each of two legs is dead on arrival.

Equity: PASS. ADV ~3.5m shares ≈ $55m/day. Short interest 4.7% of float. A 1% book position is trivially executable.

Per the CIEN precedent, "this options trade is bad" is not "this position is uninvestable" — the two are separated here explicitly. Gate 5 does not bind; it constrains the vehicle.

Gate 6 — Momentum / tape → FAIL (long-side)

For a long this is a falling knife, and Gate 6 fails without qualification. (For a short it would pass — but Gate 4 kills the short.)

This name is the deliberate B1 test the calibration file has been asking for — see §8.

6. Position sizing (if converted)

• Conviction: LOW. Composite ~0.42 — Gate 2 fails, Gate 6 fails, Gate 4 clears by 0.6pp.
• Volatility tier: HIGH — 62.7% realised (252d), beta 1.39.
• Grid: LOW conviction × HIGH volatility → 1.0% of book, ceiling 1.5%, against a 5% single-name cap.
• Vehicle: equity only. Not options, at any strike or expiry.
• Maximum book loss at 1.0% sizing with a $10.50 invalidation: −0.33%, inside the −2.0% per-position limit.

Correlation and cluster

No breach of the book's 0.60 disclosure limit — unlike the NTRA–GH pair (+0.603 at 126 days). CAI would be the fifth name in the life-science tools & diagnostics cluster (GH, NTRA, TXG, TWST, CAI). All five are currently Watchlist/no-position, so the 25% sector-concentration cap does not bind today, but if more than one converts they must be sized as a cluster.

7. Catalyst calendar (Task 6)

Monitoring items that resolve without an earnings call:

8. Calibration — what this name actually tests

B1 — the deliberate test case, and the answer is not what B1 predicted

CALIBRATION_WATCH.md item B1 states plainly: "B1 requires a deliberately-selected sample of names actually trading BELOW their 200-day — it cannot be tested opportunistically." TXG (+102.8% above its 200-day) and TWST (+68.3% above) did not test it; on both, Gate 6 was a tailwind.

CAI is 28.7% BELOW its 200-day, with −31.6% 12-1 momentum and a 0.397 52-week-high ratio. This is the first name in fifteen where Gate 6 is a genuine long-side headwind on a cheap, cash-generative, insider-bought name — precisely the configuration B1 says the framework mishandles.

And the result is informative: Gate 6 is NOT the binding gate. Delete Gate 6 entirely and CAI is still a Watchlist, because Gate 4 fails (+0.6pp) and Gate 2 fails independently. B1's concern — that momentum blocks otherwise-buyable value names — is not what happened here. The name was stopped by arithmetic, not by the tape.

That is one observation, not a refutation. But it is the first real evidence on B1 in either direction, and it points against B1 being the binding structural flaw. Logged as such.

B8 — the discriminating case, and it discriminates

B8 was escalated at four instances (ISRG, NTRA, TXG, TWST — all healthcare, all house-at-or-above-Street, all rejected on the multiple), with the explicit note: "Needs a healthcare name with clean earnings quality to discriminate."

CAI is the fifth instance and it is the test case — but it resolves B8 by failing the precondition: CAI does not have clean earnings quality. 74% of FY2025 operating income is a prior-period true-up. So CAI cannot be the clean-earnings discriminator either.

However, CAI does discriminate on the other axis, and this is the more useful finding. The B8 worry is that rejections are driven by systematically conservative exit multiples. Here:

• The Base exit multiple is 3.6x FY2031E revenue — BELOW the 3.9x at which CAI trades today, and cross-checked at 21.8x EBIT, which is generous rather than punitive for a mid-teens grower.
• Even so, the Base case returns +0.7% and the blended E[R] is +5.3%.
• Applying the current peer median 7.8x instead would give ~$29/share — but that is assuming Caris re-rates to the multiple of peers this same system just called 51–55% overvalued.

So on CAI the rejection is NOT a conservative-multiple artefact. The exit multiple is below the current trading multiple and it still does not produce a clearing return, because the absolute level of forecast cash flow is modest relative to a $4.0bn EV. That is genuine evidence that B8's mechanism is not universal. It does not clear B8 — four prior instances stand — but it removes CAI from the pattern rather than adding to it.

NEW — C9: Sloan accruals are structurally blind to cash-settled revenue timing

The mirror of CIEN's S5. Sloan accruals = (NI − CFO)/assets. Caris's prior-period true-ups increase CFO, making the ratio more negative, which the literature reads as higher earnings quality. The single largest earnings-quality defect in this name causes the headline earnings-quality factor to improve. Gate 1's quantitative corroboration limb cannot see this class of defect at all.

Where S5 said the scorecard measures the wrong object for forward-disclosure theses, C9 says it measures the wrong object for revenue-timing theses — and worse, it moves in the wrong direction. Suggested (not acted on): a revenue-quality limb — prior-period revenue adjustments as a % of operating income, and DSO trend — alongside accruals. For CAI those read 74.5% and 11 → 38 days, both severe, while accruals read "excellent."

NEW — C10: the Watchlist verdict is under-determined when the flip point is inside input precision

A 0.6pp flip point means the verdict is not a finding. The framework currently records this as "Gate 4 FAIL," which reads as a substantive conclusion when it is closer to "insufficient resolution to decide." Suggested (not acted on): a distinct verdict class — UNDETERMINED — for flip points inside ~2pp, so the ledger can later score whether these differ from genuine rejections. GOOGL (12pp) and CAI (0.6pp) are currently recorded identically and are not the same epistemic object.

B6 — is this Watchlist honest?

Yes, and by a wide margin. P(conversion within 12 months) ≈ 78%: - T1 (price ≤ $13.13): barrier-touch probability at 62.7% vol = 84.6% naive, ~72% after a haircut for positive drift. - T2 (Q2-2026 evidence): all three of ASP ≥ $4,200, true-up ≤ $6m, volume ≥ 56,000 — ~22%, resolving in under two weeks.

Far above the ~10% "no position in a watchlist costume" floor.

The screener

CAI did not appear in Screen_2026-07-27 — it would have been filtered at the momentum/trend stage (−31.6% 12-1). Honest base-rate note: this name has no screen support, which per B3 is exactly the population the funnel is designed to exclude. It arrived as a user-named ticker.

9. Conversion triggers

To LONG — DISJUNCTIVE (either suffices):

• T1 (price, independently converts): ≤ $13.13 with no fundamental deterioration — defined as clinical case volume still growing ≥ 12% y/y and blended ASP not falling y/y. At $13.13, E[R] = +25%, clearing the hurdle by 20.3pp — a margin that survives input precision. P ≈ 72% / 12m.
• T2 (evidence, independently converts): Q2-2026 print delivers ALL THREE of (i) blended ASP ≥ $4,200, (ii) prior-period true-up ≤ $6m in the quarter, (iii) clinical volume ≥ 56,000. P ≈ 22%. This would raise the Base case materially and cut p_bear below the 25% level at which E[R] reaches +11.5%.

To SHORT — NOT A LIVE TRIGGER. Gate 4 fails outright for a short (−5.3% gross before borrow), Gate 6 is already exhausted (−60.3% drawdown), and the mechanism's own quantitative corroboration contradicts it. Classified honestly as no short at any price currently foreseeable, rather than left on the book as a two-sided costume.

Invalidation of the long bias (→ Avoid at any price): - Blended ASP below $3,600 in any quarter - Prior-period true-up ≥ $12m for two consecutive quarters - Payer 1 concentration above 40%, or any adverse MolDX rate action on the primary WES/WTS codes - FY2026 revenue guidance cut below $1.00bn - Any discretionary (non-10b5-1) insider selling - Clinical case volume growth below +8% y/y

Model: CAI_Model.xlsx. Verified by independent formula-graph recalculation: tie-out sum of absolute errors to the filed 10-K = 0; balance-sheet check = 0.0000 in every forecast year. Two real bugs were found and fixed by that verification (a revenue unit mismatch and three off-by-one row references) — the read-back is not ceremonial.