Separate Facts From Inference

Two records sit side by side in GTM Lab's review queue. Both display the same line: Annual revenue: above threshold.

On the first, that figure came from the account record, where it was entered from a filing eleven days ago. On the second, there was no revenue figure at all. The decision estimated it, as the decision card permits, from the industry and the size of the company's web presence. The estimate is recorded in the decision output, exactly as Make the Decision requires.

It just is not recorded on the screen. Both records show the same sentence in the same weight, and the reviewer clears both in the same four seconds.

An inference that looks like an observation is a fabrication

That is a strong word and it is the right one. The system did nothing wrong; it estimated when permitted and recorded that it estimated. The display is what converted a marked inference into an apparent fact, and from the reviewer's side there is no difference between that and the system having made it up.

This is the point where the earlier guides' work either pays off or is wasted. Define the Decision required that an estimate announce itself. Make the Decision gave it a field. If the explanation renders that field the same as an observed one, all of that was bookkeeping.

Three levels of provenance, visibly different

Every displayed value carries one of three states, and they must be distinguishable at a glance rather than on inspection. Observed: a value that exists on the record, from a stated source, at a stated time. Inferred: a value the decision derived, with what it was derived from, displayed differently, marked, and never shown without its basis. Absent: no value, and the decision proceeded without it, shown explicitly rather than omitted, because a missing row and a row nobody scrolled to look identical.

GTM Lab's second record now reads: Annual revenue: estimated above threshold, inferred from industry and web presence. No figure on record. The reviewer's four seconds become a real judgment, because the thing they need to judge is now on the screen.

Do not let an inference become a field

The display rule has a storage counterpart worth restating, because the two failures reinforce each other. An estimated value must not be written into the field that holds observed values. Once it is there, it is indistinguishable from an observation to every system that reads it afterward, including this one on its next run, which will then treat its own guess as evidence. That is how a single inference becomes a permanent fact about a company that nobody ever measured.

Carry this into your business

Look at a screen your team uses that mixes CRM fields with anything generated or derived. Ask somebody to point at which values were observed. Most people point at whichever ones are formatted most confidently, and that is the whole problem: it is a presentation failure, not a data-model one.