The short version
- What Theranos claimed: more than 200 tests from a few drops of fingertip blood, run on its own small analyzer in Walgreens pharmacies. What it had: an analyzer that could complete about 12 tests; the rest ran on commercial Siemens analyzers that engineers modified to take diluted fingerstick samples. Regulators found "immediate jeopardy" at its lab in 2016, it voided two years of results, and Elizabeth Holmes was convicted of investor fraud in 2022. Her projected release date is February 2030.
- What others have honestly built since: Babson Diagnostics with BD collects fingertip blood in Austin pharmacies for 68 tests, but runs them in a central lab. Truvian has an FDA-cleared benchtop that combines chemistry, immunoassay and blood counts, but only on venous blood so far. Nobody does the whole thing from capillary blood in minutes.
- What Thimble Dx proposes: a counter-top analyzer that combines four standard measurement methods, splits the sample at collection, measures interference on every sample, refuses to report what it cannot stand behind, and sends anything else to a central "hub" laboratory the same day.
- Status: a paper design, a bench plan and a business plan. No prototype exists yet.
Can one sample run many tests at once?
Both of the obvious answers are half right.
The skeptic's half. Every quantitative result needs three physical things: a reaction or property specific to the target (an enzyme, an antibody, an electrode, a dye, a mass-to-charge ratio), enough sample for that reaction, and a detector. Software only does the last step, turning a calibrated signal into a number. It cannot create specificity the chemistry does not provide. That is why one undivided drop cannot run hundreds of tests, why "scan the blood with light and let software read everything" (the Raman approach) has not worked for blood, and why some tests need a different tube: EDTA preserves cells for counting, but it binds calcium and its potassium salt contaminates potassium results, which ruins those chemistry tests.
The optimist's half. Many tests from one sample is real and routine when the sample is divided into many small parallel reactions read by shared detectors. Software scales across all of them easily; the chemistry and the volume are what limit the count.
| Approach | Analytes from one sample | Example | Limit |
|---|---|---|---|
| Parallel reaction chambers, one reagent each | About 10 to 30 per consumable | Abaxis Piccolo: up to 14 analytes from 100 µL on one disc; the Chem Disc design has 24 reagent cuvettes | Microliters per reaction; one reagent per chamber |
| One tube feeding several measurement modules | 30-plus | Truvian: 34 tests from one lithium-heparin tube, including a CBC (FDA-cleared 2025 and 2026, venous) | Venous volume; capillary blood not yet cleared |
| Sensor arrays, each spot coated with its own antibody | Dozens | Genalyte: more than 100 photonic rings per chip, four per analyte | Each analyte still needs its own antibody and validation |
| Mass spectrometry | Hundreds of small molecules per injection | Forensic toxicology screens | Large instrument, sample preparation and specialists; quantitation still needs standards for each analyte |
| Optical "scan everything" (Raman) | Unproven in blood | Haemanthus's claim | Weak signal; overlapping spectra; hemoglobin absorption and fluorescence swamp the targets |
Where that leaves the Station: about 40 assays plus a 19-parameter CBC from about 170 µL used in the store, by splitting the draw into two tubes (EDTA and heparin), and those tubes into about 30 parallel reactions and a few thousand imaged cells. Not infinite, and not one or two: a few dozen, each with its own chemistry, its own share of the sample and its own validation.
Theranos would not have survived a Daubert hearing
The five Daubert factors that courts use to screen scientific testimony make a clean test of any diagnostic claim. Theranos failed every one. The Thimble plan is written to pass them before anything is sold.
| Daubert factor | Theranos | Thimble plan |
|---|---|---|
| Can it be, and has it been, tested? | CMS inspectors found its own analyzer often missed the company's own precision standards | Every assay is compared with a cleared reference method (CLSI EP09) before it is offered |
| Peer review and publication | Two papers with original data from company scientists by August 2016 | Method comparisons published before any marketing claim relies on them |
| Known or potential error rate | Unknown; two years of results voided | Each analyte held to its CLIA acceptance limit with a sigma metric of 3 or better, and the measured error published |
| Standards controlling the technique | Lab cited for immediate jeopardy; license revoked | CAP accreditation, CLIA, an ISO 13485 quality system, proficiency testing |
| General acceptance | Proprietary methods kept secret | Textbook methods: enzymatic photometry, image cytometry, immunoassay, ion-selective electrodes |
How it works, in laboratory terms
1. Collection and traceability
The analog of chain of custody. A barcoded kit timestamps the draw. Blood goes from a cleared lancet or an upper-arm collector into two microtubes with the right additive for each kind of test: EDTA for cell counts, lithium heparin for chemistry and immunoassays. The order of draw follows the clinical standard for capillary samples (CLSI GP42). The instrument rejects a sample that is underfilled, too old, clotted or not traceable, the way an evidence intake desk refuses an unlabeled item. Every result carries its sample age, reagent lot, calibration, QC status and operator in an audit trail.
2. Four methods, all textbook
| Engine | Method | Measures | Where you have seen it |
|---|---|---|---|
| Chem Disc | UV-visible absorbance, enzymatic endpoint and kinetic assays (NADH at 340 nm), Beer-Lambert quantitation, on a spinning disc that separates plasma | Metabolic panel, lipids, liver and kidney tests | Any clinical chemistry analyzer; spectrophotometric methods |
| Heme Card | Image cytometry: cells counted and classified in chambers of known depth, with a fluorescent nucleic-acid stain (acridine orange) and a trained image classifier | Complete blood count with a 5-part white-cell differential; HbA1c by boronate affinity | Microscopy and automated image analysis |
| Immuno Pack | Sandwich and competitive immunoassays on strips, read by time-resolved europium fluorescence | TSH, free T4, ferritin, CRP, vitamin D, PSA, hCG; cardiac and sepsis markers for urgent care | Immunoassay screens such as ELISA and EMIT in toxicology |
| EC Card (option) | Ion-selective electrodes and enzyme amperometry, calibrated on the card before the sample arrives | Sodium, potassium, chloride, ionized calcium, glucose, lactate | pH meters and blood gas analyzers |
The counting statistics are ordinary Poisson arithmetic. A chamber 20 µm deep with blood diluted 1:200 shows about 500 red cells per square millimeter; imaging 14 mm² counts about 7,000 cells, a counting error near 1.2 percent, inside the CLIA limit of 4 percent for red cell counts. The same scan counts about 350 platelets (5.4 percent against a limit of 25 percent). White cells, in a separate 100 µm chamber at 1:10 with the red cells lysed, give about 970 cells (3.2 percent against 10 percent).
3. Presumptive and confirmatory
Forensic toxicology screens with immunoassay and confirms with mass spectrometry. Thimble borrows that structure without borrowing its weakness: the Station's results are quantitative and must meet CLIA limits on their own, but anything flagged (abnormal cells, results near a decision limit, interference) and anything outside the in-store menu goes to the hub laboratory's high-complexity analyzers, with images or the reserve sample.
4. Interference: the postmortem potassium problem, in the living
Hemolysis releases potassium, LDH and AST from red cells. That is why postmortem blood potassium is uninterpretable, and it is the main reason fingerstick chemistry is hard: capillary samples hemolyze in about 17.7 percent of draws versus 10.6 percent for venous, and potassium is the analyte most affected. Theranos diluted its samples and reported anyway. The Station measures a hemolysis index (and lipemia and icterus indices) spectrophotometrically on every sample and, above a validated threshold, does not report potassium at all. The patient is offered a venous draw instead. Suppress, never "correct."
5. Quality control
The same tools a forensic laboratory uses under ISO/IEC 17025: control materials charted on Levey-Jennings plots with Westgard rules, external proficiency testing, accreditation inspections, and a laboratory director who signs off. Two additions suit a fleet of small instruments: patient-based moving averages that catch slow drift, and automatic remote lockout of any analyte at any site that fails QC until service and director review.
Why it is possible
Every method has been cleared or CE-marked before. Centrifugal chemistry discs (Abaxis Piccolo, 1995; 11 of its discs are CLIA-waived), handheld electrochemistry (i-STAT, 1992), fingerstick image cytometry (Sight OLO cleared in 2019; PixCell HemoScreen cleared for direct fingerstick in 2023), fluorescence immunoassay readers (about 60,000 Boditech readers installed), boronate-affinity HbA1c (Abbott Afinion), cleared capillary collection (Tasso+ in 2022, BD MiniDraw in 2023).
The volume arithmetic works. About 370 µL is collected (16 to 18 fingertip drops, or one upper-arm draw); about 170 µL is used in the store for a typical panel; the rest is reserve. Theranos's "200 tests from a few drops" never could: each assay class needs its own additive and its own aliquot.
Regulators clear this kind of instrument. FDA cleared Truvian's multi-modal benchtop in 2025 and 2026, 26 assays across chemistry and hematology, including the first CBC from a lithium-heparin sample. Babson's 44 clinical studies on nearly a million tests showed pharmacy fingertip collection can match venous testing.
The limits are known and planned for. Capillary potassium and total CO2 may end up venous-only; capillary platelets run lower than venous; coagulation and blood gases are excluded; lateral-flow immunoassays cannot reach third-generation TSH or high-sensitivity troponin, so those go to the hub. The plan says in advance what gets dropped if a stage gate fails.
What a real prototype costs
| Stage | What it proves | People | Time | Cost (planning estimate) |
|---|---|---|---|---|
| Bench modules | The physics works on control material and the founder's own blood | Founder, with advisors | 2 to 4 months part-time | $800 to $1,650 in parts |
| Engine prototypes | Each of the four engines meets precision targets on commercial controls; a first pilot comparison against a reference lab | 3 to 5 (two engineers, an assay scientist, a part-time consulting lab director) | about 6 months | about $0.6M to $0.9M |
| Integrated alpha Station | All engines in one chassis; venous method comparison for most of the menu | about 25 | about 12 months | about $10M (the seed round) |
| Beta Stations, hub lab, capillary study | Real-world capillary performance under IRB; first revenue; first FDA submissions | about 55 | about 18 months | about $40M (Series A) |
Engine prototypes, itemized:
| Item | Estimate |
|---|---|
| Salaries: two engineers and an assay scientist for six months, plus a consulting laboratory director | $350k to $500k |
| Shared wet-lab space | $30k to $60k |
| Used lab equipment: spectrophotometer, centrifuge, research microscope, pipettes, incubator, biosafety cabinet, refrigerator and freezer | $60k to $120k |
| Prototype hardware: circuit boards, machining, 3D printing, optics, motors | $40k to $80k |
| Reagents, controls, calibrators and reference-lab testing | $25k to $50k |
| Legal, patent filings and a freedom-to-operate search | $40k to $80k |
| Regulatory and quality consulting | $20k to $40k |
| Total | about $0.6M to $0.9M |
For scale: Theranos raised more than $700 million (some reports say about $945 million). Truvian raised roughly $206 to $294 million over about ten years before its first clearances. The full Thimble plan raises $130.75 million in four stages, each released against data.
Can an honest prototype attract investors?
The money exists, for companies with data
- Function Health raised $298 million in November 2025 and $450 million in July 2026 for consumer lab testing.
- Truvian raised a $74 million round in July 2024 on the way to its clearances.
- Vitestro (a blood-drawing robot) raised a $70 million Series B; Biolinq raised a $100 million Series C.
- Early rounds are smaller: SiPhox Health's seed was $10 million (Khosla Ventures and Y Combinator) and its Series A $17 million (Intel Capital); Babson's Series A was $13.7 million and its Series B $31 million, with Siemens Healthineers and BD as strategic investors.
- Big diagnostics companies buy working technology early: bioMérieux paid €138 million for SpinChip at the CE-mark stage, before US clearance.
- Grants fund diagnostics too: BARDA gave Visby $12.3 million for its at-home PCR work, and the Gates Foundation gave LumiraDx $14.2 million for a tuberculosis test.
The scrutiny is higher than in Holmes's day
Theranos's largest checks came largely from investors outside health care, and some health-care investors reportedly passed after diligence. Today any fingerstick company is asked about Theranos in the first meeting. Haemanthus, the new company led by Holmes's partner, raised about $18 to $20 million with no product and in October 2026 drew a Fortune investigation headlined "patents, investors, and a missing CTO, but no product." That scrutiny is the opportunity: a company that invites it, with data, stands out.
What Holmes showed versus what Thimble would show
| Theranos | Thimble | |
|---|---|---|
| Demonstrations | Staged; software hid failures; some samples were run elsewhere | Live, on production software, with blinded samples the investor's own expert brings, scored like a proficiency test |
| Data | No peer-reviewed validation; pharmaceutical logos added to reports | Method comparisons against a reference lab, posted publicly |
| Validation | Claimed outside validations that did not exist | CLSI EP05 precision and EP09 comparison studies, reviewed by independent clinical chemists |
| Board | Statesmen and generals | At least two laboratory-medicine experts with authority over quality |
| Laboratory leadership | Credentialed directors sidelined | A named, board-certified laboratory director who can halt testing |
| The ask | More than $700 million on claims | Four staged rounds, each released when the previous gate passes |
A realistic path
This is judgment, not data:
- Bench modules plus a technical co-founder with IVD experience: enough to approach medical-device angel investors and grant programs for a pre-seed round of $0.5 million to $1 million. Feasible, competitive, and the co-founder matters more than the bench.
- Engine prototypes that meet precision targets, a pilot comparison against a reference laboratory, a consulting laboratory director and a regulatory strategy memo: the normal basis for a seed round near $10 million in diagnostics, comparable to SiPhox's seed and Babson's Series A.
- An integrated alpha with venous comparison data for most of the menu: the basis for a Series A, where strategic investors such as diagnostics manufacturers often join.
What makes it fail: a team without IVD experience, claims that run ahead of data, and running out of money between gates. The plan addresses the first by hiring, the second by the honesty charter, and the third by keeping each stage small.
Questions a forensic scientist should ask
Can't a computer just read all the tests at once? No. Software turns signals into numbers, and that part scales almost without limit. Each number still needs its own chemical reaction, its own share of the blood and its own validation; that is the part that limits how many tests one sample can give.
What is your error rate? Not known yet. The targets are the CLIA acceptance limits (for example ±0.3 mmol/L for potassium, ±4 percent for hemoglobin, ±8 percent for HbA1c) with a sigma metric of 3 or better. An analyte that misses is not offered from capillary blood.
Has any of it been peer reviewed? Not yet; nothing has been built. The plan forbids marketing a claim before its method comparison is public.
How do you know capillary results equal venous? The plan requires a paired capillary-versus-venous study of at least 2,000 adults under IRB approval before any capillary claim. Until then, results are venous claims only.
What is your chain of custody? A barcoded, timestamped kit; additive-specific tubes; rejection criteria; an audit trail recording who, what and when for every result.
What happens when QC fails? The affected analyte locks at that site automatically and stays locked until service and the laboratory director sign off.
Who is accountable for results? A named, board-certified laboratory director with authority to stop testing, reporting to an independent quality committee of the board rather than to the CEO.
How is this not Theranos again? Theranos hid its methods, its error rates and its failures. This plan publishes methods and data before claims, invites blinded testing, empowers the people who can say no, and writes those rules into the bylaws.
What would make you stop? Each stage has a written gate. If an engine cannot meet its precision target, it is redesigned before money is spent on integration; if capillary results fail for an analyte, that analyte becomes venous-only or hub-only; if the hub does not reach 100 requisitions a day by month 36, the go-to-market plan is reworked before raising more.
Read further
- One drop, many tests: what Theranos claimed, how it quietly retreated, and every real route to many results from one small sample.
- The Theranos autopsy: what was real, what was fraud and why the science failed.
- Who got there: 69 companies scored on the six parts of the 2013 vision.
- The device: optics, counting statistics, electronics and the validation plan.
- Build it: the bench modules and their parts list.
- Regulatory and quality and the business plan.