Intended use, limitations and AI disclosure

Cala is a desktop application for hardware engineers troubleshooting complex equipment. You load the machine’s OEM manuals, service bulletins and your own notes. Cala builds a model of that machine on your computer, then works the issue with you: what else the problem could be, and evidence quoted from your documents with page citations. It orders the checks by what each one rules out, and writes a service report at the end. It runs on macOS and Windows.

It is built for the issues nobody has written up. An issue with a ticket behind it belongs in your wiki or your ticket system, and Cala is not the expensive way to look something up.

The candidates are spread deliberately across different kinds of causes:

• thermal
• mechanical
• electrical
• contamination
• control and software
• alignment and metrology

Spreading them is the method. It enforces the discipline not to chase the first idea. It is not a claim that the right cause is always among them.

The output is an argument, not an instruction

Cala produces reasoning, citations, labelled assumptions, an ordered diagnostic sequence and a report file. It produces no command, no setpoint and no signal. Everything it produces is intended to be read, checked and either accepted or rejected by a person before anything is done to a machine.

Cala is intended for qualified engineers and technicians who are competent to troubleshoot the equipment in front of them. They must also be authorised to work on it, and trained in the site and OEM procedures that govern that work. It is written on the assumption that you could reach the same answer without it, given enough time and enough reading.

Cala is not intended for, and must not be relied on by:

• anyone who is not qualified to work on the equipment, or not authorised to work on it;
• an operator being asked to act on guidance they are not trained to evaluate;
• anyone using it as a substitute for OEM service, OEM training or a required qualification;
• anyone using it to justify departing from an OEM procedure, a site safety rule or a permit-to-work system.

The competence to recognise a wrong answer is a precondition for using Cala, not a benefit it provides. If you cannot tell whether a cited value has been read correctly off the page it points at, the product is not usable safely by you on that machine.

Cala reasons from two things and nothing else: the documents you have loaded, and what you have typed into the investigation. Every limit below follows from that.

• It has no connection to your equipment. No sensor feed, no telemetry, no historian, no fieldbus, no tool control system. It cannot read a live value and it cannot write one.
• It does not know the current state of your machine. It cannot see wear history, recent maintenance, site-specific modifications, or out-of-spec parts fitted years ago. Anything not written in a document you loaded or a message you typed is invisible to it.
• It is bounded by the documentation you gave it. If a subsystem has no documentation loaded, Cala should say the evidence is missing rather than fill the gap with something that reads well.
• It does not know your document is current. Cala reads the revision you gave it. If a later revision, a service bulletin or a field change order supersedes it, Cala has no way to know unless you loaded that too.
• It has no view of your regulatory context. Qualification, validation, calibration status and any certification obligation on the equipment are outside what it models.
• Its ranking is a judgement. The order of candidate mechanisms is an argument about likelihood, not a measurement. It carries no calibrated probability and none is displayed.

The following uses are excluded. They are not unsupported features that may arrive later. They are outside the intended purpose of the product, and using Cala for them is a misuse of it.

Never in a control loop

Cala output must never be fed, manually or automatically, into anything that acts on equipment without a competent person deciding first. That includes an automated control loop, a machine controller, a recipe, a PLC and a supervisory control system. There is no interface for this and none is planned.

Never in an interlock or a safety function

Cala must never form part of an interlock, a permissive, an emergency shutdown path, a protective device, a functional-safety function or any instrumented safety system. It must not be used to defeat, bypass, override or justify overriding any of those.

Never as the sole basis for a safety-critical decision

Do not use Cala output as the only support for a decision that could injure a person, damage equipment, release energy or a hazardous substance, or invalidate a qualification. Those decisions need the OEM documentation, the site procedure and a qualified person, in that order.

Also excluded

• Any use as evidence of compliance, qualification, calibration or certification, or as a record required by a regulator or a quality system.
• Any medical, clinical or patient-facing use, and any use in the diagnosis, treatment or monitoring of a person.
• Any use where a wrong answer is acted on without a competent person having the opportunity, the information and the authority to catch it.

Cala is decision support for a qualified engineer, not for automated control, interlocks or safety functions.

Article 50 of the EU Artificial Intelligence Act requires that a person is told when they are interacting with an AI system, and that AI-generated content is marked as such. Cala meets that obligation as follows, and would meet it regardless, because a person reading a troubleshooting argument needs to know where it came from.

You are interacting with an AI system

Cala’s answers, its extracted knowledge, its mechanism rankings, its diagnostic sequences and its service reports are generated by a large language model operating on your documents. Cala is not a person, not a reviewed engineering opinion and not an OEM position. Inference runs at a third-party provider, named on the subprocessor register.

A persistent marker the model cannot remove

For as long as the desktop application is open, its status bar carries a marker stating that the content is AI-generated. That marker is drawn by the application, not written by the model. A label the model writes disappears the moment the model decides not to write it. This one cannot be dropped, suppressed or argued away by anything in a conversation.

A stamped footer on every exported deliverable

Every file Cala exports carries a verification footer, written by the application at export time and not by the model. It tells whoever opens the file that the content is AI-generated and must be verified against the OEM documentation and procedures before anyone touches the tool. It travels with the file, so the disclosure survives the report being attached to a ticket, forwarded to a vendor or read by someone who has never seen Cala.

The stamped wording is the same sentence used everywhere else on this site and in the application:

Cala's answers and extracted knowledge are AI-generated. Verify against the source documents and OEM procedures before acting.

If you remove or edit that footer when you republish the content, you take on the duty of telling the reader yourself. Do not present Cala output as your own reviewed engineering conclusion unless you have actually reviewed it.

These are the ways Cala is known to go wrong. They are listed so you can look for them, not so the list can be pointed at afterwards.

• It can misread a value. A number can be read correctly as a shape and wrongly as a fact. That happens in tables, multi-column layouts, footnotes, units in a header several rows up, scanned pages and phone photographs.
• It can cite the wrong place. A value may be attributed to the wrong page, the wrong table or the wrong revision of a document. This is why the citation names a document and a page: so the claim is cheap to check.
• It can produce something that is not in your documents. Where Cala is reasoning past what it has read, it should label the line as an assumption. The label tells you the line is unsourced. It does not tell you the assumption is sound, and an assumption can be missed and presented as though it were read.
• It can miss a cause entirely. Spreading candidates across different kinds of causes reduces the chance that every candidate is the same idea in different words. It does not guarantee the right one is on the list, particularly where the subsystem involved has no documentation loaded.
• It can call evidence missing when it is not. A value present under a name, symbol or unit that Cala did not connect to your question can be reported as absent.
• It can rank a correct set in the wrong order. Every cited value can be right and the favourite still be wrong. Read the excluded mechanisms and the stated reasons, and disagree with them where you have grounds to.
• It is not repeatable. The same question asked twice can produce differently worded reasoning and, occasionally, a different ranking.
• It has no memory of your site. Local practice, undocumented modifications and the history of that particular tool are known to you and not to Cala.

The Windows build is a preview and is not code-signed yet, so you may see a SmartScreen warning on first launch. That is a product limitation too, and it is better learned here than at 2 AM with the tool down.

Before you act on anything Cala tells you, verify it against the original OEM documentation and the procedures in force at your site.

• Open the citation. Treat every cited value as a pointer to a page, not as a fact. Read the page.
• Check the revision. Confirm the document Cala read is the one that currently applies to your machine, and that no bulletin or field change order supersedes it.
• Separate evidence from reasoning. Anything labelled as an assumption is something you still have to test.
• Follow the procedure, not the report. Isolation, lockout, permits, personal protective equipment and requalification are governed by your site and the OEM. Cala has no view of any of them.
• Where they disagree, the OEM document wins. Every time, without exception.

Cala's answers and extracted knowledge are AI-generated. Verify against the source documents and OEM procedures before acting.

This statement is published by Lumino Solution LLC, trading as Cala. Questions about intended use, limitations or the AI disclosure can be sent to the address below.

privacy@calalayer.com

This page is versioned. Every change to it, including the wording of the excluded uses and the failure modes, is recorded in the legal changelog, and material changes are notified by email to registered account holders.

Related pages: what Cala does, how Cala protects your data, terms and conditions.