Monday, March 2, 2026

refrigerator mysteries per AI. Stops cooling. Fixes itself.

 We have a household appliance puzzle for you. We have a 25 year-old extra refrigerator, which is a large refrigerator. We use it mostly to store sodas in the top and store extra frozen goods in the bottom. A week ago, the refrigerator part stopped cooling, it would be at least 50°, however, the freezer part on the bottom seem to be working perfectly fine. 

This went on for several days. Also, we are absolutely certain that both doors were completely closed, we double and triple check that. This went on for several days. 

Then, the refrigerator part quietly started working again in refrigerating to 37°.What do you think was going on?

LInked In May Conflate 2 Caris Tests (MRD MCED)

Linked in article says CARIS MCED may be 80% sensitive in stage 1, instead of 15% like competitors.

https://www.linkedin.com/posts/alexgdickinson_wtf-caris-life-sciences-launching-an-mced-share-7434033461000302592-LxLn/

Chat GPT thinks this is conflating two studies (one commenter on Linked In seemed to raise this too).

I don't know the correct answer but a jump to 80% sensitivity in Stage 1 seems "unlikely" in a multi cancer MCED test.

###

 What you’re seeing on LinkedIn is (mostly) a test-name / study-design mix-up.

Two different Caris blood tests are getting conflated

1) Caris Assure (commercial today; liquid biopsy for people with known cancer—therapy selection / monitoring)

Caris published a 2025 Scientific Reports paper describing Caris Assure + an AI layer (ABCDai) and reported stage-by-stage “MCED” sensitivity figures including 83.1% in Stage I at 99.6% specificity—but this work is based on case-control / known-cancer samples, not a true “average-risk screening” setting.

That’s where the viral “83% Stage I” number is coming from.

2) Caris Detect (the proposed MCED screening product; not yet broadly launched)

Caris Detect is the company’s forthcoming blood-based screening MCED. Caris says it uses whole genome sequencing (WGS) + AI.

Caris has publicly described an Achieve 1 study interim readout for Detect, with:

  • Stage I sensitivity: 56.8% (n=266)

  • Stage II sensitivity: 70.1%

  • Stage I–II combined sensitivity: 61.3%

  • Specificity: 99.1% in an “asymptomatic screening population” (n=121) and 95.3% in the broader “undiagnosed” cohort (n=1,505)

Caris also announced a partnership with Everlywell to distribute/launch Caris Detect, and stated an expected launch in the first half of 2026.

So, what’s the “proposed new Caris MCED test” in plain English?

Caris Detect is Caris’ screening-intended MCED: a blood draw, WGS-based signal extraction, and AI classification to call cancer presence (and they’ve shown some stage/cancer-type breakdowns in the interim Achieve 1 readout).

Why the LinkedIn table is “WTF?!” but not necessarily “fraud”

The LinkedIn post is effectively comparing:

  • Assure paper (case-control / known-cancer) → yields eye-popping Stage I 83.1% figures
    vs

  • Galleri / other MCEDs (often reported in screening-like or intended-use frameworks)

Case-control designs usually inflate apparent sensitivity compared to true screening settings (spectrum effects, cancer mix, sample ascertainment, etc.). That doesn’t make the Assure results “fake”—it makes them not directly comparable to real-world screening performance.

What we do know about pricing and launch timing

  • Caris has said Caris Detect is expected to launch 1H 2026 and will be offered through Everlywell’s platform.

  • I did not find an official public list price for Caris Detect in Caris’ releases; the $3,500 figure being circulated looks like commentary rather than a formal published price (and may be getting tangled with “Assure” economics).

Practical takeaways for you (policy / investor lens)

  • The headline “83% Stage I MCED” is best interpreted as Assure+ABCDai performance in a case-control context, not proven screening performance.

  • The more relevant current “MCED screening” signal for Caris is Detect / Achieve 1 interim: ~57% Stage I, ~61% Stage I–II, with specificity ranging ~95%–99% depending on cohort definition.

  • The key due-diligence questions now are intended-use population, prospective blinded validation results (the ~865 held-out set Caris mentions), and how they’ll manage PPV / downstream workups at population scale.

If you want, I can turn this into a one-page “what’s real vs what’s hype” memo suitable for an investor update (with the two-test taxonomy front and center).

##


##

Author link (BQ)

https://chatgpt.com/c/69a5c7eb-f568-8327-bdfd-f58cb3444b57

Public link

https://chatgpt.com/share/69a5c977-897c-8005-9f54-a81b6079d361

###

###

Here is a clean one-page investor-style brief you could circulate.


CARIS MCED: What’s Real vs What’s Hype

March 2026 | Bruce Quinn–style investor brief

Executive Summary

Recent LinkedIn commentary claiming “83% Stage I detection” for a new Caris MCED test is based on a conflation of two different Caris products:

  • Caris Assure → liquid biopsy for known cancer (therapy selection/monitoring).

  • Caris Detect → the screening-intended MCED product.

The 83% Stage I figure comes from Assure in a case-control study, not from the population screening product.

The relevant MCED screening data for investors is from Caris Detect (Achieve 1 interim) showing:

  • Stage I sensitivity: ~57%

  • Stage I–II combined: ~61%

  • Specificity: 95–99% (cohort-dependent)

Still strong, but not “83% Stage I screening.”


The Two Caris Tests (Do Not Confuse)

1️⃣ Caris Assure

Intended use: Patients with known cancer
Platform: Multiomic profiling (WES/WTS database + AI)
Study design: Case-control
Reported Stage I sensitivity: ~83%
Specificity: ~99.6%

This is not a screening test in average-risk individuals.
Case-control design typically inflates apparent sensitivity relative to screening settings due to spectrum effects.


2️⃣ Caris Detect (MCED Screening Product)

Intended use: Asymptomatic screening population
Technology: Whole genome sequencing + AI classifier
Study: Achieve 1 (interim readout)

Interim performance:

  • Stage I: 56.8%

  • Stage II: 70.1%

  • Stage I–II combined: 61.3%

  • Specificity: 99.1% in asymptomatic subset; 95.3% broader cohort

Planned commercial launch: 1H 2026 (Everlywell distribution partnership)

This is the relevant dataset for MCED comparison.


Competitive Context

TestStage I Sensitivity (Reported)Study Context
GRAIL Galleri~15–20%Case-control but screening-intended
Exact Cancerguard~15%Case-control
Caris Detect~57%Interim, mixed cohort
Caris Assure~83%Known-cancer case-control

The viral narrative compares Assure vs Galleri, which is not apples-to-apples.


Is Caris Differentiated?

Potential advantages:

  • Massive tissue multiomic training dataset (~376k historical cases)

  • Multi-signal detection (mutations, CNVs, expression, etc.) vs methylation-only competitors

  • WGS-based platform

Open questions:

  • Will prospective, blinded validation replicate interim numbers?

  • Performance in true average-risk screening (low tumor fraction)?

  • Positive predictive value at population prevalence (~0.5–1%)

  • Cost structure (rumored ~$3,500) vs competitors ($700–$950)


Key Investor Questions

  1. Is Detect’s 56–60% Stage I signal durable in held-out validation?

  2. What is cancer-type mix? (Stage I pancreas ≠ Stage I thyroid.)

  3. What is PPV at screening prevalence?

  4. How does assay cost scale if WGS depth must increase?

  5. Will FDA demand prospective interventional validation before reimbursement?


Bottom Line

The 83% Stage I claim is real—but not for screening.
The actual MCED screening candidate (Detect) currently shows ~57% Stage I sensitivity, which is meaningfully better than 15% competitors, but not revolutionary.

Caris may indeed represent the first serious multiomic alternative to methylation-only MCED—but validation and economics will determine whether this becomes a commercial inflection point or another over-interpreted interim dataset.


If you’d like, I can also produce:

  • A buy / hold / sell framing

  • A regulatory pathway risk analysis

  • Or a PPV modeling table at different cancer prevalence rates (more useful than stage sensitivity headlines)