Gates before money.
Phases, the raise, early hires and how it earns.
Summary
The cheapest tests are also the ones that could kill the company, so they run first, on public data, before any money is raised. Then raise about $1.8M for 18 months, the deck’s pre-seed, to the first paying members and the seed: a leader, an ML engineer, three partner labs and the benchmark playground, a $884k plan with an equal contingency. The seed round after that is raised on proof that labs share and members pay, not on the science alone.
All dollar figures here are proposals derived from public benchmarks and salary norms, not quotes; market medians are cited where used.
01Phases and gates
Phase 0 costs a salary and compute, and it is the phase most likely to change the plan: if brightfield alone cannot place a lab, the check needs one cheap stain; e8 and e10 now show that answers move with position, so the pitch is not QC alone.
| Phase | Months | Proves | Kills or narrows it |
|---|---|---|---|
| 0 · Kill tests | 0–3 | brightfield carries the lab signal; position changes the drug answer (e8); the signal is not only instrument | brightfield near 9% chance; e8 Spearman near zero; variance nearly all microscope model |
| 1 · Three-site proof | 3–9 | three labs upload the same line and their checks agree with independent ones | labs will not upload; checks disagree with ground truth |
| 2 · Built to break | 9–18 | the planted-failure reference catches each failure and grades correction methods | failures go undetected; a second line does not transfer |
| 3 · Network | 18+ | many labs per lot make bad-lot alerts possible | too few labs share a lot |
02The raise
Recommendation: about $1.8M, the deck’s pre-seed, to fund 18 months to the first paying members and the seed. The kill tests run first on public data; the round buys three partner labs, a paid pilot with a screening lab by month 9, the benchmark playground and a model for any cell. The product is a membership to the repository: labs join free by uploading, screening teams and AI teams subscribe; biotech companies come in through a free, private backup of their raw images.
Why that size: the $1–2.5M middle of the pre-seed market is thinning, down from 24% of rounds in Q1 2023 to 18% in Q1 2026 [1], and once a financing reaches about $3M, priced rounds become as common as SAFEs [2]. So $1.8M sits in the squeezed middle and is raised on what the public map already shows; about $3M is the size to choose if the kill tests land before the round. For scale, aggregators of Carta data put the mid-2026 median US priced seed at $4.1M on a $24.3M post-money, about 18% dilution [4].
| Option | Size | When | For | Against |
|---|---|---|---|---|
| A · Pre-seed now, the deck’s ask | ~$1.8M | now | 18 months to the first paying members | lands in the thinning middle [1] |
| B · Pre-seed after kill tests | ~$3M | month 2–3 | priced on evidence; covers phases 1–2 | 2–3 months of runway from savings or a grant |
| C · Priced seed now | ~$4M+ | now | matches the seed median [4] | priced before brightfield is tested, so valued as a hypothesis |
The next round is a seed of roughly the median size, raised around months 15–18 on phase 1 proof (section 6). Series A comes after the network works; the Q2 2026 median Series A was $14.7M [3].
03Use of funds
About 71% of the money is people; the benchmark playground and the partner labs are the lines that buy the moat. Each line is the plan; the contingency below it doubles the plan (estimates, not offers).
| Line | 18 months | Share | What it buys |
|---|---|---|---|
| leader, to run it | $180k | 20% | 18 months |
| scientific lead (me) | $180k | 20% | 18 months |
| ML engineer | $160k | 18% | from month 7 |
| payroll overhead | $104k | 12% | 20% of salaries |
| benchmark playground | $120k | 14% | ~60 plates · planted faults |
| partner-lab bench time | $30k | 3% | 3 labs · controls they already image · reagent top-ups · shipping |
| compute and storage | $45k | 5% | images · models · hosting · API |
| legal, IP, insurance, accounting | $45k | 5% | incorporation to first contract |
| travel and partner visits | $20k | 2% | 2 benchmark partners · 3 labs |
| contingency | $884k | +100% | most things take two to three times the time and money planned; we plan for 2× |
| Total | ~$1.8M | $884k plan + $884k contingency |
The order matters more than the split: nothing is spent on bench time until phase 0 passes, and the planted-failure set is sized by what phase 1 shows.
04Early hires
Four hires, each staged to the phase that needs it. A leader comes before any of them, someone to run the company. The pre-seed pays for the leader and the first hire, the ML engineer from month 7; the other three come with the seed. You hold the science.
| # | Role | Starts | Profile to look for | Why then |
|---|---|---|---|---|
| 1 | Founding ML engineer | month 7 | image representation learning plus distribution shift; evaluates rigorously (grouped cross-validation, permutation nulls); fluent in the bioimage stack (CellProfiler, JUMP tooling); has shipped a benchmark or a paper someone reran | phase 0 is entirely computational |
| 2 | Assay and imaging scientist | with the seed | high-content imaging in brightfield and Cell Painting; has transferred an assay between sites; has run plates through cores and CROs; can design plates that fail on purpose | phase 1 needs three sites run the same way |
| 3 | Data platform engineer | with the seed | moves terabytes of images reliably; keeps each lab's data separate by design; watches cloud cost | uploads from outside labs start in earnest |
| 4 | Partnerships lead | with the seed (fractional first) | has sold to core facilities or pharma screening groups, or worked as a field scientist for an imaging or cell vendor; scientists trust them | converting pilots to paid lines before the seed |
Where to look. ML: alumni of cell-profiling groups and of phenomics or ML-for-biology companies, and academic groups working on distribution shift. Assay: pharma screening groups, CROs, and core facility staff. Partnerships: field application scientists at imaging and cell companies.
Advisors, before hire 4. A bioimage-informatics academic from the JUMP circle (credibility and the reference); a head of screening at a pharma company (the buyer's view); a quality and standards expert who has run identity testing to published standards [5]; someone from a cell bank or vendor (the channel). Plus the employment counsel you already use, to keep the company clear of any prior-employer IP question.
Not yet. A sales team or a wet lab of your own: cores and CROs cover bench work until the planted-failure set proves what an in-house lab would add.
05Membership and revenue
Identity testing already has a price: $217 per human STR sample at ATCC [5], $255 with a database search at the Johns Hopkins core [6]. That buys one identity check. Membership buys a behaviour check on every batch and the repository behind it, so it is priced per cell line per quarter; the quarter is a starting cadence, and the drift measured so far argues for more often.
| Member | Price (proposed) | What’s at stake for them | Role |
|---|---|---|---|
| Academic labs and cores | free for labs that share their control images; ~$500–1,000 per line-quarter for a private check | the paper | grows the map |
| Pharma and CRO screening teams | ~$5–15k per line-quarter, or a site licence | the screen; one late-caught line cost ~$2.6M | first revenue |
| AI labs and model builders | by API, priced by use; the most | a model that fails on another lab’s cells | main revenue |
| Companies that buy cell data | ~$10–50k per dataset placed on the map | data they did not make and must trust | data revenue |
| Cell banks and vendors | a membership for each line they sell: every lot checked against the field before it ships | the lot and every lab that bought it | the channel |
| Later: bad-lot alerts | part of membership | every screen run on a drifting lot | only the network can make it |
Unit economics (derived, not measured). One pharma site with 10 active lines at $5k per line-quarter is ~$200k a year; twenty such sites is ~$4M a year before AI, dataset or vendor memberships. Against a single ~$2.6M failure, $50k a year is under 2% of the loss it prevents. For the size of the whole problem, about $28B a year of US preclinical research is irreproducible, with reagents and reference materials the largest share [7].
What never gets sold. Rankings of instruments, or the right for a vendor to grade its own lot. The benchmark can be licensed; the verdict cannot.
06What the seed is raised on
The seed pitch is traction, not the 99.9%. By about month 15, these should be true:
| Evidence | Target by month 15 | Why an investor cares |
|---|---|---|
| Kill tests | all three reported, including any that narrowed the product | the risk that was science is now known |
| Labs uploading | 3+ institutions on one line, uploading every quarter | the cold start is solved |
| Checks that hold up | agreement with independent checks at all three sites | the product works outside your hands |
| Paying members | 2+ paying, at least one pharma or CRO | someone pays at the proposed price |
| Second line | the reference extended past U2OS | the method is not one cell line |
| Planted-failure set | designed and started | the moat is under construction |
07Trade-offs to decide
| Decision | Short-term cost | What it buys | Recommendation |
|---|---|---|---|
| Free membership for contributing labs | academic revenue | the reference and, later, bad-lot detection | ✓ |
| Price per line per quarter, not per plate | revenue that tracks usage | repeated uploads over time, which build the reference | per line |
| Raise after the kill tests, not before | 2–3 months without outside money | a round priced on evidence | wait |
| Instrument makers as a channel | the easiest distribution | none that survives: neutrality is the moat | license the benchmark only |
| Cores and CROs instead of an own lab | less control of the bench | 18 months of runway | outsource until phase 2 |
| Who leads it | the lead, and its equity | someone whose job is to run it, while you build the science | a leader, or a home that already has one |
References
Checked for this doc on 28 September 2026. Reference 4 is a secondary summary of Carta data; the others are primary. Salaries, prices and budgets not cited here are proposals.
- Carta. State of Pre-Seed: Q1 2026. carta.com
- Carta. A disappearing middle: why the pre-seed market is growing increasingly barbell-shaped. March 2026. carta.com
- Carta. Series A funding: how to raise your first priced round (Q2 2026 benchmarks). carta.com
- Waveup. Seed round benchmarks 2026, summarising Carta medians. waveup.com
- ATCC. FTA Sample Collection Kit for Human Cell Authentication Service (STR), 135-XV. atcc.org · service standards: atcc.org
- Johns Hopkins Genetic Resources Core Facility. Cell line authentication. grcf.jhmi.edu
- Freedman LP, Cockburn IM, Simcoe TS. The economics of reproducibility in preclinical research. PLoS Biology 13(6):e1002165, 2015. PMC