idea-003 · feasibility deck
Every slide states a condition that would have to be true, then reports where it stands using the candidate file's own claim and its own confidence tag. A deck never upgrades a tag, invents a number, or recommends anything, and there is no ask slide.
The chair: engineering. The people who have to make a pooled-donor-plasma material with known tumor-derived molecules in it at levels no dPCR (digital polymerase chain reaction) can measure directly, make it the same way three times, and produce the evidence FDA (US Food and Drug Administration)'s special control and a laboratory's inspector will ask for.
How to read this: every slide is a condition, not a conclusion. A bracketed claim reference points into knowledge-base/candidates/idea-003.md: the number is the claim's position in its ## Claims list, and the tag is copied from that line and never adjusted here. A [no claim] marker means nothing in the file speaks to the condition at all. See docs/deck-spec.md.
What changed since the 2026-09-15 build: that deck's load-bearing condition was the Poisson limit at the lowest levels. The P1 derived-value bottom levels mechanism and the P2 clinical QC material at launch intended use answer it by design rather than by measurement. Levels at or above 0.1% are dPCR-quantified. Levels below 0.1% are value-assigned by gravimetric dilution into a dPCR-quantified diluent, with an uncertainty budget including a limit-of-blank term, and are labelled as derived. The regulatory route is now 866.5910 / SCU (subcutaneous), which brings design controls with it. The file has also gained a detailed set of design conditions from the clinical, technology and operations chairs (claims 133-173). The engineering question has moved from "can it be measured" to "can the derivation be defended, and can it be built three times". Rebuilt from the file, not patched.
Would have to be true: The product would have to deliver two things. The first is a multi-level panel: an unspiked blank plus at least three tumor-fraction levels, each with a stated expanded uncertainty. The second is a daily control at a mean per-variant VAF (variant allele frequency) of about 0.03% in single-use vials of at least 30 ng, with enough certified variants that a customer's assay expects at least 3 mutant molecules per extraction, so a zero-copy run happens no more than 5% of the time. Both have to ship with tumor-proxy and normal-proxy DNA so a tumor-informed assay can be designed to see them.
Where it stands: Every element is a design condition on a product that does not exist. The panel configuration is part of claim 149unconfirmed claim 149unconfirmed. The daily control's level, vial mass and single-step derivation from the 0.1% level are claim 166unconfirmed, which the Verifier also flags for an arithmetic slip: 30 ng is about 9,091 genome equivalents, not "at least 9,100" claim 166unconfirmed. The at-least-3-molecules rule and its labelled zero-copy probability are claim 160unconfirmed claim 160unconfirmed. The qualitative-only restriction below 100 molecules is claim 161unconfirmed claim 161unconfirmed. The tumor-proxy and normal-proxy supply is claim 162unconfirmed claim 162unconfirmed.
What would settle it: A written design input specification carrying these numbers under design control, then pilot-lot data against it.
If it's false: If a tumor-informed assay cannot be designed against the control, the daily control is invisible to the assays it is meant to monitor. See desirability slide 4.
Would have to be true: dPCR would have to anchor the value-assignment chain at or above 0.1% VAF. Gravimetric dilution would have to carry it below that level with an expanded uncertainty (k=2) of no more than 35% relative, including dilution, between-unit heterogeneity and a characterized limit-of-blank term. At realistic inputs, the Poisson floor would have to be handled by the label rather than denied.
Where it stands: The anchor is the best-evidenced part of the whole file. In an interlaboratory study, ddPCR agreed with the gravimetric value from 53.9% down to 0.1%, with an LoB, LoD and LoQ of 0.01%, 0.02% and 0.1% claim 100verified. A dPCR primary reference procedure had an LoQ of 0.2% and a false-positive rate of 0-0.003% claim 99verified. Read together, as the file's Verified Composite reads them, that is why sub-0.1% levels can be derived but not independently measured. The step below the anchor is not demonstrated. The 35% budget is claim 134unconfirmed claim 134unconfirmed. The Poisson arithmetic is internally correct and rests on unverified inputs: at 10-30 ng, a single variant at 0.01% is absent from roughly 40-74% of inputs claim 133unconfirmed claim 16unconfirmed. Vial-to-vial heterogeneity of no more than 10% needs at least 100 expected molecules per vial claim 138unconfirmed. An independent sequencing check of the derived levels on the pilot lot is a separate condition claim 169unconfirmed. The published warning sign is that every assay in the FNIH QC (quality control) study dropped out at 0.1% claim 96unconfirmed.
What would settle it: A dPCR replicate series of at least 20 replicates per level at the panel's stated input claim 62unconfirmed, a written uncertainty budget for the pilot lot, and the orthogonal sequencing comparison in claim 169unconfirmed. All of this is bench work, and no connector reaches it.
If it's false: The levels below 0.1% lose their stated uncertainty, the product retreats to the range incumbents already occupy (viability slide 6), and the mechanism factor's 3 on verified evidence has nothing below the anchor.
Would have to be true: The 510(k) (premarket notification) exemption at 866.5910 would have to survive all three tests in 866.9: the neoplastic-disease limb, the different-intended-use limb, and the "reasonably foreseeable characteristics" preamble, for a native-plasma matrix. That would mean the build is design controls plus a special-controls study set, not a premarket submission. FDA would also have to accept performance testing for that study set when no cleared MRD (molecular residual disease) assay exists to test on.
Where it stands: The code is verified. SCU is Class II (FDA device class II) under 866.5910, 510(k)-exempt, and defined as a QC material for NGS (next-generation sequencing)-based IVD (in vitro diagnostic) test systems, including synthetic DNA and RNA controls claim 131verified. The engineering obligation that comes with it is verified: an ISO (International Organization for Standardization) 13485 quality system with Clause 7.3 design and development, UDI (unique device identifier) and complaint reporting claim 159verified. Whether the exemption applies is four open FDA judgments: the neoplastic limb claim 52unconfirmed, the different-intended-use limb claim 53unconfirmed, whether "assay validation" falls within 866.5910(a) claim 156unconfirmed, and whether a native-plasma matrix is a reasonably foreseeable characteristic when the regulation names only recombinant, synthetic and cell-line controls claim 157unconfirmed. The special-controls guidance is claimed to recommend matrix and extraction evaluation, multi-lot stability, reproducibility at 3 or more sites, and performance testing on FDA-cleared assays or an accepted reference method claim 158unconfirmed. No tumor-informed MRD assay is claimed to hold FDA clearance claim 163unconfirmed. So whether FDA would accept LDT (laboratory-developed test) assays or dPCR in their place is a condition this deck put on record claim 189unconfirmed. The nearest precedent, a third-party oncology-monitoring control distributed exempt, is unchecked on its load-bearing half claim 155unconfirmed.
What would settle it: A 513(g) request or a pre-submission to FDA, and a human attaching the 2007 special-controls guidance PDF. No connector reaches either.
If it's false: A 510(k) is required. The design-verification set grows into a submission package, and the capital and calendar on viability slide 4 both move.
Would have to be true: A single pooled-donor-plasma lot would have to be commutable and clean at the same time. Its background at each certified position (donor germline or clonal-hematopoiesis variants plus method error) would have to be no more than one-tenth of the lowest assigned level, with a 95% upper bound no higher than 0.001%. The spike would also have to be recovered through extraction at the same efficiency as the donors' own cfDNA (cell-free DNA).
Where it stands: The joint condition is claim 63unconfirmed claim 63unconfirmed. The background thresholds are claims 139 and 167 claim 139unconfirmed claim 167unconfirmed, and extraction equivalence across three methods is claim 135unconfirmed claim 135unconfirmed. The verified evidence here is all hazard. Some extraction kits preferentially recover spiked mononucleosomal DNA over endogenous cfDNA claim 92verified. The matrix and the digestion method both change dPCR output, though that study used fetal bovine serum rather than human plasma claim 110verified. Fragmentation improved commutability for one viral standard and not another, which is an analogy at best and a mixed one claim 93verified. Commercial materials from two vendors are claimed to carry more low-VAF variants than patient samples from 0.01% upward claim 89unconfirmed.
What would settle it: Duplex sequencing of an unspiked candidate pool at the certified positions, and a spike-recovery study through three routine extraction kits. Both are bench work.
If it's false: At the lowest levels the assay sees the donors or its own error rather than the spike. The limit-of-blank term in slide 2's budget swamps the signal, and the commutability premise the product is named for fails.
Would have to be true: Before any vial ships, the file's own design conditions would have to be met on real material. That means a dPCR replicate series, spike recovery across extraction methods, duplex background on the pool, orthogonal agreement on the pilot lot, three consecutive lots within 20% of nominal with overlapping uncertainty intervals, one external laboratory designing a panel against a lot, and a commutability demonstration against patient samples.
Where it stands: None of this has been run, and every line is unconfirmed: the replicate series claim 62unconfirmed, extraction recovery claim 135unconfirmed, duplex background on at least 3 µg of pool cfDNA claim 167unconfirmed, pilot-lot agreement claim 169unconfirmed, three-lot consistency claim 170unconfirmed, an external panel design claim 162unconfirmed, and cross-chemistry commutability claim 31unconfirmed. The functional characterization study BLOODPAC is claimed to expect of contrived samples is also unread claim 164unconfirmed. The background measurement alone consumes plasma: at 5-20 ng of cfDNA per mL, 3 µg needs 150-600 mL of pool per measurement, and a 0.001% floor needs litres claim 173unconfirmed. The concentration input to that arithmetic has published counter-evidence on record claim 141unconfirmed.
What would settle it: The bench programme itself, run under design control, with its outputs attached to the candidate so the Verifier can read them.
If it's false: Any one of these failing means the P2 clinical QC material at launch label cannot be written as specified. Slide 1's specification has to change, and so does the panel's claimed range.
Would have to be true: The inputs would have to be obtainable on terms that permit a commercial product. That means a pooled plasma lot of at least 10 L, tested per 21 CFR (Title 21 of the Code of Federal Regulations) 610.40 with donor-level records, under consent terms that allow genetic characterization; a matched tumor and normal cell-line pair under terms that allow redistribution and variant disclosure, banked with at least 1,200 certified clonal variants; stability through freeze-thaw, shipping and shelf life; and an ISO 17034 calendar that starts only once a first lot exists.
Where it stands: Donor screening is a verified requirement for plasma used to manufacture a device claim 61verified, and the file names no supplier for any input claim 59verified. Everything else is open: the plasma supplier and pool size claim 172unconfirmed, plasma consent terms claim 148unconfirmed, spike source supply terms claim 143unconfirmed, cell-line licence and consent claim 176unconfirmed, the banked master stock claim 168unconfirmed, stability at the lowest levels claim 32unconfirmed, and the ISO 17034 sequencing claim 142unconfirmed. The nearest device family's recall record shows the failure modes to design against: value-assignment error, stability drift and vial-to-vial imprecision claim 60unconfirmed. That tag covers the whole claim. Its historical half is corroborated inside the claim's own note.
What would settle it: Named suppliers and signed terms for plasma and cell lines, attached by a human, and a real-time and shipping-stress stability study on the pilot lot.
If it's false: A consent or licence term that forbids commercial redistribution or variant disclosure stops the product outright, whatever the bench shows. A plasma pool too small for one background characterization multiplies slide 5's work per lot.
Would have to be true: The spike-production route would have to avoid whatever every independent claim of US 11,215,622 requires, which is claimed to be formaldehyde fixation followed by acoustic shearing. The product would have to fall outside the claims of the two Natera testing-standards patents. And no incumbent patent could read on the plasma matrix, on sub-0.1% gravimetric value assignment, or on run-by-run monitoring with a spiked control.
Where it stands: Three live, in-term US patents in this exact space are verified, but only on title and dates claim 102verified claim 103verified claim 104verified. What their claims cover is unread: the claim-1 readings claim 105unconfirmed claim 106unconfirmed and the independent-claim scope readings claim 145unconfirmed claim 146unconfirmed claim 147unconfirmed. The broader clearance is an FTO (freedom to operate) negative nobody has searched claim 174unconfirmed. A design-around route exists in the literature. Micrococcal-nuclease and enzymatic digestion both produce nucleosome-sized fragments without the fixation and shearing steps claim 107verified claim 109verified. That route is worth something only if claim 145unconfirmed's reading holds.
What would settle it: The granted claims of all three patents attached from USPTO (United States Patent and Trademark Office), and a commissioned FTO opinion. The patent connector returns bibliography only.
If it's false: The spike-production step has to change or be licensed, and the enzymatic-digestion route may be the one that is blocked instead.
Every condition above with nothing verified behind the condition itself. Slides 2, 3, 4, 6 and 7 carry verified lines: the dPCR anchor, the SCU code and its quality-system obligation, three extraction and matrix hazards, the donor-screening rule, and the three patents' existence. But the condition each slide turns on rests on unconfirmed or unverified lines, and slide 4's verified lines are all adverse. Seven of seven conditions.
No condition on this deck is [no claim]. The operations, technology and clinical chairs already put every design condition on record. What the deck adds is claim 189unconfirmed, the gap between what the special control asks for and what exists to test against.
If only one thing from this chair could be checked: whether each panel level below 0.1% VAF can be value-assigned by gravimetric dilution from a dPCR-quantified parent into a dPCR-quantified diluent with an expanded uncertainty (k=2) of no more than 35% relative that includes dilution, between-unit heterogeneity and a characterized limit-of-blank term, and be labelled as derived rather than measured claim 134unconfirmed.
This one rather than the others because it is P2 clinical QC material at launch's answer to the Poisson wall, and it is the product thesis. The dPCR anchor above it is verified. Everything that makes it hard sits inside its budget: the vial heterogeneity on slide 2 and the background on slide 4 are its terms. Naming it is not a recommendation, a gate, or a kill. It is the answer to "check what first."
_shared/plain-language.md