Pieza independiente
Provenance or nothing
Movement 5 · The Laboratory That Watches Itself — September 2026
I Two ways of being wrong
Pipetting by hand at twenty to four in the afternoon, anyone skips a column. You know half a second later, from the movement, before you look. The body tells you: that was the whole of quality control for centuries.
An automated station does not skip columns. It also notices nothing. There is no movement. If the reagent in reservoir three has degraded, it distributes it with admirable precision across all three hundred and eighty-four positions [four full plates], and the report will come out clean, with tight error bars.
An error repeated with precision stops looking like an error. It looks like a law.
II What has to be kept
| Field | Why | Cost of not having it |
|---|---|---|
| Instrument and calibration | locates the bias | repeating everything |
| Reagent lot | groups the failures | false findings by lot |
| Model version | explains a change of judgement | incomparable results |
| Operator or agent | reconstructs the chain | nobody answers |
| Failed attempts | corrects the optimism | biased literature |
This is called provenance. It sounds like filing. It is. It is the part of the work nobody presents at a conference, and without it a suspect result cannot be traced back to its origin: you have to repeat everything.
“Labs that do this well do not publish faster. They retract less.”
III The arithmetic of bias
- e
- systematic error per measurement
- n
- number of measurements the system produces
By hand, n is small and the error shows. It shows in the movement. Nobody writes it down. With an autonomous lab, n is enormous and e gets multiplied by it before anyone looks at a single chart. Automation introduces no new errors. It scales the ones that were already there. It scales them well.
IV Minimum protocol [safe]
-
01
Every datum travels with its origin, its timestamp and its instrument.
-
02
Every change of model version or protocol is dated and signed.
-
03
Negative results are kept with the same care as positive ones.
-
04
A human reproduces one measurement in every so many by hand, chosen at random.
-
05
No conclusion leaves the system without being able to point at the physical experiment holding it up.
V Risks
Documenting everything costs, and it is paid in time. A lab that records well moves more slowly than one that records nothing; over a quarter it looks worse.
That comparison is the trap. The careless lab moves fast until somebody finds the bias; then it loses, all at once, everything built on top, which is usually two years of work and several careers.
VI Coda
Scientific automation is not decided by the speed of the loop. It is decided by whether the loop can retrace its steps, by whether someone can point six months later at the exact bottle [with its lot and its date] the number came out of.
A system that is wrong coherently does not shout. It publishes.