Thimble Dx
The Theranos idea, done honestly.
Theranos promised hundreds of tests from a drop of blood and never had them. We traced exactly why, scored 69 companies that have built pieces of the idea since, and designed a station that only claims what it can prove.
Concept stage Not a medical device · Figures are targets, not results
The short version
Every part has been done.
Never all at once.
Theranos promised six things: a few drops, many tests, minutes, a pharmacy counter, a posted price and lab-grade accuracy. Each has been achieved by someone. Nobody has put all six in one box.
From first principles
Start with how blood tests actually work.
Six facts decide what any blood-testing device can do. Theranos ignored them. The Station is built on them.
Every test needs its own reaction.
A number comes from chemistry: an enzyme, an antibody or an electrode that responds to one substance. Software turns that signal into a result, and it can read as many signals as you like. It cannot create a signal the chemistry never made.
One drop, many testsEvery reaction needs its own share.
A drop is about 30 µL. A lab tube holds a few thousand. Hundreds of tests from one drop was arithmetic that never worked.
Some tests need a different tube.
Cell counts need one additive; chemistry needs another that would ruin them. So the blood is split the moment it is drawn.
Precision comes from counting.
Count 100 cells and you are off by about 10 percent. Count 7,000 and it is about 1. A camera counts thousands in seconds.
Squeezing breaks cells.
A squeezed finger bursts red cells, which spill potassium and fake a dangerous result. The Station measures that damage in every sample and will not report a number it cannot trust.
Trust is a number.
Every US lab test has a legal limit on how wrong it may be. The Station publishes its own error against that limit. Theranos hid its.
Split the draw. Run small reactions in parallel. Check every sample. Send the rest to a lab. Publish the data.
That is the whole design: about 40 tests and a full blood count from about 170 µL, using four methods that are already cleared in other products.
How it works
Four proven methods. One counter-top station.
Each bay does one kind of measurement that another company has already proven. Together they cover the tests most people actually get.
Heme Card
A camera counts and classifies your blood cells, the full blood count with the white-cell breakdown, and reads your three-month blood sugar.
20 µL · about 5 min · CBC + HbA1cChem Disc
Spins the blood to separate plasma, then reads color changes in 24 tiny wells: kidney, liver, salts, sugar and cholesterol.
100 µL · about 12 min · 25 testsImmuno Pack
Antibodies that glow find hormones and proteins: thyroid, iron stores, inflammation, vitamin D, PSA, pregnancy, heart markers.
15 µL each · 10 to 15 min · 12 testsEC Card
Tiny electrodes, like a pH meter, measure salts, calcium and lactate in about two minutes. An option for urgent care.
65 µL · about 2 min · 9 testsEverything else goes to a real lab, the same day.
Hormones, B12, ApoB, serology and drug levels travel in a third tube to the hub laboratory by courier.
One visit
About fifteen minutes, start to finish.
- Minute 0CollectA painless upper-arm collector or a fingertip draw by trained staff fills two small tubes.
- Minute 4LoadEach card or disc takes its own measured share. No pipettes, no robot arm.
- Minute 5 to 15MeasureAll bays run at once. The Station checks every sample for damage before it reports.
- Minute 15ReviewThe pharmacist reviews the results with you. Anything flagged goes to the hub lab.
Has anyone done this?
Yes, in pieces. Never the whole thing.
Yes, in pieces.
Babson and BD run 68 tests from fingertip blood collected in Austin pharmacies, in a central lab. Truvian's FDA-cleared benchtop runs chemistry, immunoassays and blood counts in minutes, from a vein. Small devices from Abbott, Roche and others have run a few tests in minutes for 30 years.
Never all at once.
Of the 69 companies and products we scored, none delivers a broad menu from a few capillary drops, in minutes, at a pharmacy counter, at a posted price, with published accuracy.
Many tried and failed.
Theranos, LumiraDx, Cue, Forward and Sight Diagnostics raised more than $2.7 billion between them. Most died of economics, not physics.
- Full
- Partial
- None
- Thimble target, not yet built
Scoring rules are in section 1. The full 69-entry scorecard is in section 15.
Show the data as a table
| Entry | Drop | Menu | Minutes | Retail | Price | Accuracy |
|---|---|---|---|---|---|---|
| Theranos (2013 claims) | partial | none | none | partial | full | none |
| Truvian TruVerus | none | full | full | partial | partial | full |
| Babson + BD MiniDraw | full | full | partial | full | full | full |
| biosurfit spinit | full | partial | full | partial | partial | partial |
| Abbott i-STAT | partial | partial | full | partial | partial | full |
| Abbott Piccolo Xpress | partial | partial | full | full | partial | full |
| Roche cobas b 101 | full | partial | full | full | partial | full |
| Genalyte Maverick | partial | partial | full | partial | partial | partial |
| Boditech ichroma | full | partial | full | partial | partial | partial |
| Sight Diagnostics OLO | full | partial | full | partial | partial | full |
| PixCell HemoScreen | full | partial | full | partial | partial | full |
| Function Health | none | full | none | full | full | full |
| Thimble Station (target) | target | target | target | target | target | target |
Why this could work where others failed
Built against each failure.
Not guaranteed. Unproven until a working instrument meets the legal accuracy limits on fingertip blood.
Two numbers decide it: accuracy on fingertip blood for a few hard tests (potassium, CO2, platelets), and patients per Station per day. The plan tests both before raising large money.
Honesty, built in
The opposite of Theranos, in writing.
Ten rules go into the company's bylaws and every investor agreement. Six of them:
The plan
Raised in stages. Released by data.
Stations are placed in urgent care clinics and pharmacies and paid for by the test supplies they use. Our own lab proves the system first.
- Revenue
- Operating expenses
Projections, not results. Base case: 8 patients per Station per day. Generated from site/model.py.
Show the data as a table
| Year | Revenue ($M) | Operating expenses ($M) | EBITDA ($M) | Stations at year end |
|---|---|---|---|---|
| 2027 | 0.0 | 3.5 | -3.5 | 0 |
| 2028 | 0.1 | 8.5 | -8.5 | 0 |
| 2029 | 1.7 | 16.0 | -15.2 | 0 |
| 2030 | 5.6 | 21.5 | -18.8 | 60 |
| 2031 | 17.7 | 27.5 | -17.4 | 400 |
| 2032 | 41.7 | 32.0 | -6.6 | 900 |
| 2033 | 73.8 | 36.0 | 11.1 | 1,600 |
Go deeper