Standards
Metrology measures doubt. An AI answer hides it.
Metrology spent centuries working out how to trust a number: an unbroken chain of calibrations, and a stated doubt on every result. An AI answer about your own documents has neither of those things.
Cognatum Team · Nov 4, 2026 · 6 min read
A quality manager asks the assistant which torque figure applies to the current revision. The answer arrives in a second. On the bench beside her is the torque wrench itself, with a small sticker on the handle. Who calibrated it. Against what. On what date. When it is due again. The wrench carries more provenance than the answer does.
Cognatum governs the entry
source · version · approver · permissions
Measurement always has a reference
Metrology is the science of measurement and its application. NIST, the US measurement institute, puts it in those words. The discipline is unglamorous and it is everywhere. It sits behind the food you buy, the tests a lab runs and the parts in an aircraft.
One idea sits underneath all of it. A measured property has a size and a reference that gives that size meaning. The mass of the old international prototype kilogram was one kilogram, by definition. To measure a thing is to compare it with a standard, directly or through other steps. If the instrument has been calibrated, its readings mean something in relation to that standard.
Now look for the reference
An assistant answers a question about your own documents. Ask what that answer was compared against and there is nothing to point at. No approved version was named as the reference. The system pulled several documents, ranked them, and wrote one smooth reply.
Traceability is a chain with a number on every link
Traceability has a tight meaning here. A result is traceable when it links back to national or international standards through an unbroken chain of calibrations, each one with a known uncertainty. Your pressure calibrator was checked against a reference. That reference was checked against another, and so on up to a national institute.
The kilogram shows how seriously the chain is taken. In 1799 it was a platinum artifact held in Paris. In 1889, seventeen countries agreed on a new metal prototype as the shared reference. Metal objects degrade, so in 2019 the definition moved to a constant of nature.
A citation is not a chain
An answer with a link under it looks traceable. It is not. The link names a document. It does not say which revision was read, who approved that revision, when it took effect, or whether a second document in scope said something else.
Metrology states its doubt out loud
Here is the part the rest of us skipped. Every measurement carries doubt, and metrology makes you put a number on it. Uncertainty is that quantified doubt. It tells you how good the measurement is. The guide that governs the practice says uncertainty reflects the lack of exact knowledge of the value being measured.
The wording is chosen with care. The National Physical Laboratory in the UK explains that accuracy is a qualitative idea. Using it as a number assumes a true value can be known perfectly. It cannot be. Uncertainty makes no such assumption. It gives a range, with a stated level of confidence.
Why sales copy prefers the other word
The same laboratory adds a human note. An accurate device sounds more impressive than an uncertain one, which is probably why much sales literature reaches for accuracy. AI copy does the same thing. Researchers ran the first preregistered study of retrieval based legal research tools, against vendor claims of eliminating hallucinations. The tools they tested made things up between 17% and 33% of the time.
A calibration lab may say it cannot tell
Pass and fail are not the whole story in calibration either. Guidance on compliance statements says a result passes only when the error plus the uncertainty stays inside the tolerance. It fails only when the error, with the uncertainty added or taken off, lands outside. In between there is a third verdict.
When the result sits closer to the limit than half the uncertainty, the honest call is undefined. State neither pass nor fail. An assistant has no third verdict. It writes a confident sentence, and the reader cannot tell a settled answer from a guess.
The AI Act reached for metrology, then stopped
Article 15 of the EU AI Act asks high-risk systems to reach a suitable level of accuracy and to hold it across their life. The levels of accuracy, and the metrics behind them, have to be declared in the instructions for use. The article goes further. It asks the Commission to work with metrology and benchmarking authorities on benchmarks and ways to measure.
Read the published text with care, because the page carries a notice that the Act has been amended by the Digital Omnibus on AI. Either way, a declared accuracy figure describes the system. It says nothing about the document the system read this morning, or about who approved it.
What a certificate carries
A calibration certificate is a short document that does a lot of work. It records the instrument, the reference standards used, the conditions at the time, and the readings as found and as left. Then a traceability statement, the uncertainty, a pass or fail, and the names and signatures of the people responsible.
Note what it leaves out. Certificates do not expire. The owner of the instrument sets the interval, based on how it is used, how critical it is, and what the quality system asks for. Expiry is a decision somebody makes and records, not a date the paperwork hands you.
Reverse traceability
There is one more move, and knowledge work has no answer to it. Say a reference calibrator is found out of tolerance. You work out how far off it was and how long it had been wrong. Then you check which instruments it calibrated in that window, and decide whether those results still stand. Some get done again. In a regulated plant the review can reach the product.
Now try the same thing with a procedure that turned out to have been wrong since March. Which answers came from it. Who acted on them. Nobody can produce that list, so nobody runs the review.
Calibrating knowledge, in practice
The Cognatum Knowledge Loop runs eight steps across four phases, with one human gate. AI runs seven of the eight. It finds, gathers, cleans, enriches and surfaces. Approve is the step it does not run. A named person decides what is true, reworks the wording where it needs reworking, and signs.
The fields it carries
What comes out carries the fields a certificate would carry. A named approver. A date. The source it was drawn from, and where that knowledge stood at the moment the answer was given. Where two sources disagree, Cognatum puts both in front of a person and shows where the disagreement sits, rather than picking a winner.
It is also honest when nothing is a confident match. That is the undefined verdict a calibration lab is allowed to give, and the one an assistant never gives.
Nothing has to move
Your documents stay where they are. Cognatum works bidirectionally with the systems that already hold them, so nothing migrates and the departmental split stays as it is. When a source changes, it tells you what has gone out of date. It notifies. It does not quietly rewrite your knowledge, and it is not current by itself.
Your company's knowledge isn't missing. It's unusable. Cognatum changes that.
Where this lands
A caliper that has drifted gets a sticker, a review and a list of everything it touched. An answer that shaped a decision gets a thumbs up. Metrology worked out long ago that trust in a number comes from a chain and a stated doubt. It does not come from the confidence of the voice reading it out.
Sources
- Metrology (nist.gov)
- Measurement Uncertainty (nist.gov)
- Is there a difference between measurement accuracy and uncertainty? (npl.co.uk)
- What Is Metrology? Definition, How It's Used, and More (fluke.com)
- Calibration uncertainty for dummies (blog.beamex.com)
- What Is a Calibration Certificate? (blog.beamex.com)
- AI Act Article 15: Accuracy, robustness and cybersecurity (ai-act-service-desk.ec.europa.eu)
- Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools (arxiv.org)
- The Cognatum Knowledge Loop (cognatum.ai)
Common questions
Questions this raises.
Can you calibrate an AI answer?
Not in the metrological sense. Calibration is a documented comparison against a traceable reference, and an answer has no reference above it. What you can do is govern the material it answers from, so every answer carries a named approver, a date, the source it came from, and where that source stood at the time.
Is a citation under an AI answer the same as traceability?
No. Traceability means a result links back to national or international standards through an unbroken chain of calibrations, each with a known uncertainty. A citation only names a document. It does not say which revision was read, who approved it, when it took effect, or whether another document in scope disagreed.
Does the EU AI Act require a stated accuracy figure?
For high-risk systems, yes. Article 15 says the levels of accuracy and the relevant accuracy metrics shall be declared in the accompanying instructions of use, and it asks the Commission to work with metrology and benchmarking authorities on measurement methods. The published text carries a notice about amendments made by the Digital Omnibus on AI. Either way, a declared figure describes the system rather than the document it read.
Can software make us compliant with the AI Act or a measurement standard?
No. No software confers compliance with either. A tool can align with a framework and produce the evidence an auditor or an inspector asks for, which is a different and more honest claim. The obligation stays with your organisation.
Do we have to move our documents into Cognatum?
No. They stay in the repositories that already hold them. Cognatum works bidirectionally with those systems and governs the knowledge above them, so the departmental split you have today is fine. When a source document changes, it notifies you about what has gone out of date rather than editing anything on its own.