Skip to content
An open-plan office with several workstations and screens
  1. Conglomerado
  2. Calaa Mind
  3. Consulting
  4. Lab and Robotics

01.2

What to measure, with which instrument and with how much coverage.

The instrument is decided before the algorithm

Which field the device reads, how often it reads it and what falls outside its reach are decided at the same time. An instrument that measures what was not needed is not corrected later in software.

01.2.1

Two instruments of our own, and a method fixed when they are built

Robotics builds two objects. The Cal-1 Link, which sits inside the vehicle and reads what the vehicle publishes through its diagnostic port, and KRN-01, the first satellite of the observation programme, which covers from orbit what the device cannot reach. Lab writes the software that collects what both produce, trains the models and maintains the infrastructure that serves them.

What comes out of that enters the corpus without negotiating method, because the method was fixed at manufacture. In exchange it reaches only as far as the instrumented fleet reaches, and that boundary moves with the units leaving the factory, not with the analysis.

What is instrumented

The most reliable source is the one that covers the least ground

The field is the automotive sector and the two zones are what the group's instruments read. It orders, it does not measure. It says which part is instrumented and which is not, not how much each one takes up.

Cal-1 Linkthe on-board device
KRN-01the observation satellite

The sector. Drag to rotate, or tap an object. KRN-01 is drawn faint because it is not in orbit yet.

What is not instrumented. It is resolved with an external source, and it is the list Mind returns to Lab and Robotics to decide what is worth building next.

What falls outside is not left unanswered. It is resolved through the order of preference of sources, and every figure enters flagged with the tier it came in on.

01.2.2

All three decisions are taken while the instrument is being built

Two requests from Mind, and what Mind puts in return.

  1. 01

    To Robotics

    The instrument

    Which field to read and how often, so the series serves to compare and not only to observe. A measurement taken every week and one taken every quarter cannot be cross-referenced, and that is decided at manufacture, not at analysis.

  2. 02

    To Lab

    The path of the data

    That the source, the date and the method travel with the figure at every hop. A figure that loses its provenance on aggregation stops being verifiable, and no later review gives it back.

  3. 03

    What Mind returns

    Where coverage is missing

    Which part of the sector is not instrumented and has to be resolved with an external source. That list decides what is worth building next, and it only appears once someone has tried to measure it and could not.

The two cadences

Two series with different steps almost never coincide

The same stretch of time read by two instruments at different frequencies. The links mark the moments when both have a figure at once. It orders the calendar, it does not say how much information each mark holds.

One measurement a weekShort step
Another every quarterLong step
24 of 27Where only one does
3 of 27Where both have a figure

Each dial turns once per reading. The marker passes the index and that reading drops onto its stack. Tap a series, or drag to turn the model.

Decided at manufacture, not at analysis.

Frequency is decided when the instrument is built, not when analysing what it produces. A series taken at the wrong step is not corrected afterwards, it is taken again.

Three people review an analysis on a laptop in a room overlooking the city

01.2At design

What the instrument does not read, the analysis does not recover.

That is why Mind sits in the specification of the Cal-1 Link and of KRN, and not in the reading of what they produce.

01.2.3

What an engineering team receives has to be quotable without the report

Three pieces, and the third decides whether the figure survives outside the report.

  1. The answer, first

    Whoever reads only the first page has to be able to decide.

    The question answered in one page, with the recommendation and the level of confidence. The rest of the report exists so that page can be checked if they choose to.

  2. The reasoning, with what was ruled out

    A report that shows only the path that worked cannot be argued with.

    How you get from the data to the recommendation, with the alternatives left out and the reason for each. That is where an engineering team finds whether its case was covered.

  3. A separate data sheet

    A figure that cannot be copied with its backing gets copied without it.

    Figures that can be cited separately, each with its source behind it. Without that sheet the backing stays in the report, and the figure arrives alone at the first deck that cites it.

01.2.4

Design fixes what can be measured for years

KRN-01 is fixed on the day it launches and the Cal-1 Link is fixed in its manufacturing run. What was not asked for at design is not added in the next release of the software, and from then on the sector is measured with what is available instead of with what was needed. That is why the question of what to measure comes before the instrument.

How long it can still be changed

The three pieces close at different moments

When each piece of the instrument is fixed and whether it allows a change afterwards. It compares moments of decision, not schedules or cost.

The softwareevery release
Cal-1 Linkthe manufacturing run
KRN-01the launch date
the first to close

What is sealed can no longer be touched. Only the software can be replaced. Tap a part, or drag to turn the model.

The question of what to measure has to arrive before the first gate that closes. After that, the sector is measured with what is available instead of with what was needed.