Why more molecules.

months median survival on daraxonrasib, the first RAS drug for pancreatic cancer (FDA, Aug 2026). The next drugs aim at single mutations, so doctors need to read the tumour's DNA.

RASolute 302, metastatic pancreatic cancer (ASCO 2026, NEJM).

tests last quarter, from four US companies alone. Up 70% in a year.

Q3 2026 company reports: Natera, Guardant (incl. Shield), GRAIL, Tempus.

what the best relapse tests now claim to see. Screening tests, with no tumour known, stop at about 1 in 2,000.

2026 figures are company claims. Screening: Guardant Shield 0.05% (FDA SSED P230009); GRAIL 0.07–0.17% (PLoS One 2023).

usable genome copies in a typical blood sample. Every dot is one.

Median 6,000 (range 400–14,000), 194 patients. Zviran et al., Nature Medicine 2020.

tumour copies of a given mutation in that tube, at one in a million. Usually zero. No sequencer finds a molecule that isn't there.

6,000 copies × 1/1,000,000.

more genome copies: 225,000 to 900,000. About a hundred blood draws in one cup. Red: one draw. Yellow: the cup.

300 mL is device capacity, scaled from our 100 mL data (250–1,000 ng cfDNA per 100 mL, 3.3 pg per copy).

If urine carries the same tumour share as blood. For pancreatic cancer, that's the first thing we measure.

At 1 in 100,000: 1 − e^{−copies × fraction}. Same tumour-share assumption.

the DNA a urine kit gets. HFX concentrates a whole cup, 100–500 mL, into one pellet. Collected at home, no needle.

Half are found after it has spread.

SEER: localised 44%, distant 3%.

people die of pancreatic cancer. No screening test is recommended for the public.

ACS 2026.

Veracyte for Convergent, a urine-DNA company with no revenue yet. Quest paid $300M+ for Haystack before launch. At scale: Tempus–Personalis ~$1.5B, Abbott–Exact ~$23B. Big labs buy their way in.

Veracyte 8-K (Sep 2026, $150M + up to $30M); Quest (Jun 2023, $300M + up to $150M); Tempus (Jul 2026); Abbott–Exact closed Mar 2026.