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: the reference-material scientist who has to make, characterise and certify material at allele fractions where counting statistics run out.
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, the tag is copied from it 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 2026-08-24 build: the file now specifies the bar to clear [claim 45: unconfirmed], [claim 46: unconfirmed], the assay performance the panel would have to support [claim 34: unconfirmed] through [claim 37: unconfirmed], and — most importantly for this chair — an empirical measurement of where the field's reproducibility actually breaks down [claim 39: unconfirmed]. Rebuilt from the file, not patched.
Would have to be true: Pooled donor plasma would have to retain a native cfDNA fragmentomic profile, carry characterised variant fragments from roughly 0.001% to 0.5% VAF, and each level would have to be orthogonally quantified by digital PCR with a published uncertainty budget — as a multi-level panel plus a daily-use single-level control.
Where it stands: No performance specification is stated as a claim [no claim], so this deck states none. What the last build lacked and this one has is the comparator: both named incumbent product lines bottom out at 0.1% VAF, LGC/SeraCare's supplied in a vendor-described "plasma like matrix" [claim 45: unconfirmed] and Horizon/Revvity's manufactured from fragmented human cell-line genomic DNA at an average 160 bp spiked into a synthetic plasma matrix [claim 46: unconfirmed]. That is the specification bar stated in someone else's product literature, and it locates this candidate's two differentiators precisely: an order to two orders of magnitude lower in allele fraction, and a native rather than synthetic matrix. Both vendor pages are unconfirmed because no connector in this repo verifies vendor specifications.
What would settle it: The incumbent vendors' published specifications as the bar to clear — now named in the file, though not independently confirmed; CLSI guidance on commutability and on reference material characterisation.
If it's false: "Commutable, characterised, ultra-low" is a marketing sentence rather than a spec, and there is nothing to test against.
Would have to be true: The lowest panel levels would have to be makeable and measurable — not merely difficult.
Where it stands: This is where the new evidence bites hardest. The candidate already states the physics: producing material with a stated uncertainty budget at 0.001%-0.01% VAF is Poisson-limited at typical 10-30 ng cfDNA inputs [claim 16: unconfirmed]. The research lane has now added the empirical shadow of that limit, measured across the field rather than argued from first principles: in a multi-manufacturer study spanning eight NGS platforms, coefficient of variation increased as mutant allele fraction fell, with poor repetition below 0.5% [claim 39: unconfirmed]. The formal proficiency-testing programme in this field works at 0.5%-2.5% [claim 40: unconfirmed]. So the demonstrated working floor of the discipline sits roughly 50 to 500 times above this panel's target range. Against that, the assays the panel would serve claim limits of detection well inside it — 3.45 parts per million [claim 36: unconfirmed], 0.008% VAF at high input [claim 37: unconfirmed], sub-molecule-per-mL figures [claim 34: unconfirmed] — which is precisely the gap the product exists to fill, and also the reason nobody has filled it.
What would settle it: The arithmetic itself, done explicitly against the intended input mass and stated as part of the uncertainty budget rather than around it; then dPCR replicate studies at the lowest levels to show the observed variance matches the Poisson expectation — and a direct reconciliation with [claim 39: unconfirmed], which is the published counter-datum any customer will raise.
If it's false: Note that this one may be true and survivable — the honest response is not to abandon the low levels but to publish the floor, which is what a reference-material producer is for. A panel whose lowest level says "this is the counting limit, here is its distribution" is more useful than one that implies a precision it cannot have. That is arguably the strongest version of this product, and it is not the version the frontmatter describes.
Would have to be true: The RUO label would have to hold — and the file's own claims say that depends on marketing conduct rather than on the material.
Where it stands: The definitional position is the best-evidenced thing in the file and it is genuinely favourable: quality control material under 21 CFR 862.1660 is Class I [claim 4: verified], exempt from 510(k) except for named uses this product does not claim [claim 5: verified], with product codes existing under that regulation [claim 6: verified], and a Verifier-added finding that FDA classifies a genetics/DNA quality control material as Class I [claim 22: verified]. What is not settled is whether the RUO label survives the intended conduct: the RUO regulation itself [claim 1: unconfirmed], the 2013 FDA guidance [claim 2: unconfirmed], and the enforcement-risk reading [claim 3: unconfirmed] are all unchecked, and two of them are documents no connector in this repo can read. The LDT rule's status is a fourth moving part [claim 15: unconfirmed].
What would settle it: Regulatory counsel on which regulation applies given this intended use, and on whether the intended marketing conduct is compatible with an RUO label. The connectors cannot do this.
If it's false: The build changes from "documentation and a wet lab" to a Class II submission, and the capital claim [claim 20: unconfirmed] is wrong by a category.
Would have to be true: The material would have to be commutable — behaving like a patient sample across the assay chemistries that will use it — and that would have to be demonstrated, not inferred from the choice of matrix.
Where it stands: Unbacked, and still asserted in the product's own title [claim 31: unconfirmed]. The new vendor specifications make the argument for the approach sharper without evidencing it: if the incumbents really are built on fragmented cell-line genomic DNA in synthetic matrix [claim 46: unconfirmed] and a "plasma like matrix" [claim 45: unconfirmed], then a pooled-human-donor-plasma panel is a materially different proposition — but "different matrix" is a design choice, and commutability is a measured property. Inferring the second from the first is exactly the step this slide exists to refuse. One datum shows the field does use commercial standards for method benchmarking [claim 38: unconfirmed], which is adjacent to commutability, not evidence of it.
What would settle it: A commutability study in the CLSI sense — the panel and clinical patient samples run side by side across the major tumor-informed MRD chemistries, with the agreement between them reported.
If it's false: The panel is a spiked control that behaves like a spiked control, which is what the incumbent products already are [claim 17: unconfirmed], [claim 45: unconfirmed], [claim 46: unconfirmed], and the differentiation claim disappears.
Would have to be true: Orthogonal quantification, replicate precision at the lowest levels, commutability, and stability would all have to be established before a single panel is sold into a validation packet.
Where it stands: All four are on record as this deck's write-back and all four came back unconfirmed [claim 30: unconfirmed], [claim 31: unconfirmed], [claim 32: unconfirmed], [claim 33: unconfirmed] — bench and accreditation claims about material that does not yet exist, which no connector in this repo reaches. The order of work is set by [claim 39: unconfirmed]: replicate precision at the lowest levels is not one item among four, it is the item that determines whether the other three are worth doing.
What would settle it: dPCR replicate studies per level; freeze-thaw and accelerated stability studies at the lowest levels; lot-over-lot fragmentomic profiling of the donor pool. No human subjects, no regulator — this is bench work and documentation, which is the candidate's genuine advantage.
If it's false: A panel sold into other laboratories' validation packets, whose own performance was never characterised, is the failure mode with the longest blast radius on this deck — every customer's LoD claim inherits it.
Would have to be true: Donor plasma sourcing, cold chain, lot documentation and ISO 17034 accreditation would all have to work at the scale of the target customer set.
Where it stands: Sourcing consistency that preserves the native fragmentomic profile is on record and unchecked [claim 30: unconfirmed] — and it is the requirement that distinguishes this product from both incumbents, neither of which uses pooled donor plasma [claim 45: unconfirmed], [claim 46: unconfirmed], which means there is no incumbent supply chain to copy. ISO 17034 accreditation on the stated budget and timeline is likewise unchecked [claim 33: unconfirmed], as is stability at the lowest levels through freeze-thaw and shipping [claim 32: unconfirmed].
What would settle it: An accreditation body's cost and timeline; a plasma supplier agreement with lot-level fragmentomic acceptance criteria; a cold-chain validation.
If it's false: The material may be excellent and the documentation not credible, which for a reference material is the same as failing — nobody buys an uncertified certainty.
Would have to be true: Contrived cfDNA reference material composition would have to be clear, with the real exposure sitting in fragmentomic-profile and variant-spiking methods held by sequencing incumbents.
Where it stands: Nothing was run and no patent number is identified anywhere in the file [claim 18: unconfirmed] — this is unexamined, not clear. The connector performs no landscape or assignee search, so the question cannot currently be asked in this repo at all.
What would settle it: A patent landscape search on contrived cfDNA reference material, variant spiking at defined allele fractions, and fragmentomic profile preservation; then expiry checks on anything live; then counsel.
If it's false: Automatic kill under the Stage 5 gate. On a candidate whose material is conceded to be reproducible by any competent lab [claim 19: unconfirmed], the patent position is also the only structural barrier available.
Every condition above with nothing verified behind it.
Worth stating plainly, because it is unusual in this repo: this candidate has five verified claims and they are all regulatory or definitional [claim 4: verified], [claim 5: verified], [claim 6: verified], [claim 7: verified], [claim 22: verified]. A high verified count is not a strong position; it is a well-mapped one. Thirteen new claims since the last build added real technical substance and moved no tag, because vendor pages and literature scans cannot be verified by anything here.
If only one thing from this chair could be checked: whether the lowest panel levels can carry a stated uncertainty budget at all, given the Poisson limit at realistic cfDNA inputs [claim 16: unconfirmed].
The sub-0.01% VAF range is the entire differentiation claim — it is what the incumbents' own specifications stop short of [claim 45: unconfirmed], [claim 46: unconfirmed] and what the intended use is written around. If those levels are irreducibly imprecise, the product is a competent entry into a category that already has incumbents.
The week's evidence made this condition both more urgent and more concrete. It is no longer only a theoretical Poisson argument: there is now a published multi-manufacturer measurement of reproducibility collapsing below 0.5% VAF [claim 39: unconfirmed], which is where the panel's range begins to get interesting. And unlike every other condition on this deck, the first pass at it costs an afternoon: the arithmetic can be done before the wet lab exists.
Naming it is not a recommendation, a gate, or a kill.