Skip to content
CTS Field Notes

CTS Field Notes

Overlooked problems, assumptions worth questioning, and things worth knowing.

Practical technology judgment for business leaders. Fresh perspectives for curious minds.

Two Batches, One Assembly Order

Consider an illustrative manufacturing job: a small company builds 40 control panels, each containing one circuit board, using boards from two incoming batches. Twenty-four boards come from batch A and sixteen from batch B. The purchasing system shows both deliveries. The assembly record shows 40 finished panels.

Then the board supplier asks which panels received batch B.

In this example, the invoices, packing slips, and assembly record contain aggregate quantities but no mapping from boards to individual panels. Those documents could carry that mapping; these do not. Identifying the sixteen panels requires the connection between each finished unit and the board actually installed in it.

This is a useful place to begin a traceability discussion because the answer has a practical consequence. In this example, a reliable connection would let the company identify a specific group for further investigation. Without it, the uncertainty covers more of the production run.

Preserve the relationship at assembly

NIST’s manufacturing traceability report, finalized in September 2026, describes an approach for connecting product histories across organizations that use different systems. The September announcement emphasizes exchanging and verifying information while allowing existing industry standards to remain in use. This guidance for designing traceability does not prescribe purchasing a particular platform. NIST announcement.

For our example, the first design choice is the level of detail needed. If the customer needs an answer for individual panels, a record saying that both batches were used somewhere in the order will be insufficient. The local production record needs to preserve which board went into which panel, or a sufficiently precise grouping that supports the customer’s question.

The supplier inquiry and internal panel mapping are CTS’s operational example. NIST focuses on traceability between organizations; section 4.3.2 supports backward links to preceding events and deliberately avoids forward tracking. The framework does not deliver the downstream panel lookup imagined here. The manufacturer still has to decide how to capture and query its internal work accurately.

A practical trial could use one completed order. Have production and quality staff reconstruct the component-to-panel relationships from the records they actually retain. Notice where they rely on memory, assumptions about installation order, or labels that no longer accompany the part. Those are specific capture problems to resolve before expanding the trial.

Ask for a bounded supplier record

Now extend the example upstream. The supplier may be able to identify the shipment and manufacturing batch while keeping its process settings private. The customer needs to decide which additional claim it wants supported: the batch’s origin, a particular test result, or conformity with an agreed specification.

Those requests call for different evidence. A delivery record establishes a different fact from a test report. Before asking for another document, state what it must establish and which components it covers.

NIST’s report allows for selective disclosure and alternative evidence, including digitized paper records. It also distinguishes links to preceding supply-chain events from supporting material such as test reports. The buyer must judge whether that evidence provides sufficient assurance for the item’s risk. NIST IR 8536, sections 3.4 and 4.2.3.

For the panel maker, a sensible pilot request might identify batch B, the supplier’s shipment reference, and the specific test documentation needed for the investigation. Requesting the supplier’s entire production database would add a much larger disclosure question without necessarily resolving this one.

Leave the gaps visible

Suppose the supplier’s older test report cannot be retrieved. Our editorial recommendation is to record that limitation alongside the batch relationship. A missing report and a report showing an unacceptable result are different findings; the response should reflect what is actually known.

NIST explicitly recommends distinguishing gaps such as unknown, withheld, and unreachable information. Its scope also matters: the core framework follows products through acquisition or initial deployment, with broader later-life applications reserved for further work. It should not be presented as a complete recall-management system. NIST IR 8536, sections 1.3 and 3.5.

The pilot can end with a concrete result: a list of panels associated with batch B, the evidence supporting that list, and any units whose component history remains uncertain. That gives the quality team a bounded set of facts to assess with the supplier. It also shows exactly which information the next assembly order needs to retain.

Permalink

Patient Data in the Business Applications

iRhythm's October breach update puts two facts next to each other: the company reported no identified impact to its clinical or medical device systems, and patient information had been downloaded from business applications hosted by third parties.

Both can be true. A device can remain unaffected while information about the person using it leaves through another system.

According to the company's October 2, 2026 notice, unauthorized access and downloading occurred between June 3 and June 8, 2026. The affected information included patient names and contact details, dates of birth and service, insurance and patient account numbers, and device serial numbers. Notifications began October 2 for affected individuals whose contact information the company held. Source: iRhythm's notice.

Its June filing attributed the incident to social engineering and reported an extortion demand. It did not name the affected applications or describe the precise deception. The October update said the company had found no evidence of identity theft. Neither the June filing nor the October notice discloses whether a ransom was paid. Sources: June SEC filing and October 2 announcement.

For a smaller business, this suggests a concrete exercise: choose one customer record and follow the copies. Where does it go after the main system—into a support tool, a billing service, or an export used by another team? Those are examples to investigate, not claims about iRhythm's setup.

A business application can hold sensitive information without operating the product the customer bought. Include those copies when deciding what data an incident could expose. Start the review with the fields that travel: the customer's name, account number, service history, and any other details sent along with the record.

Permalink

Three Missions, One Launch

On October 1, three NASA-supported small spacecraft launched on SpaceX’s Transporter-18 commercial rideshare. One will test propulsion systems. One will study ultraviolet light from galaxies. One is scheduled to begin inspecting inactive satellites in 2027.

They shared a rocket, not a mission.

NASA describes commercial rideshare as a way to create more opportunities for science and technology demonstrations by sharing a launch with other government and commercial spacecraft. Here, that shared journey serves three very different kinds of work: propulsion engineering, astronomy, and satellite inspection.

The launch is their common milestone. What comes afterward is specific to each spacecraft. For the inspection mission, the close-up work is still ahead.

NASA: SmallSats launch to advance science, technology, orbital operations

Permalink

Security Evidence Has a Delivery Schedule

On September 24, NIST added three practical resources to its DevSecOps project: a mapping to the Secure Software Development Framework, a CI/CD and container-deployment example, and scenarios showing activities across the software lifecycle. The live reference is open for public comment through November 9.

The project connects security work to development, builds, packaging, distribution, and deployment. For a technology leader deciding whether a change can ship, that raises a concrete question: can the evidence be traced to the exact version being released?

The release record below is an editorial recommendation for a smaller team, rather than a NIST-mandated form. Its purpose is to connect existing evidence to the decision that uses it.

A release is a chain of handoffs

Most software changes pass through more people than the final approval suggests. A developer changes code. A build system assembles it. Tests produce results. A security tool raises findings with varying severity and confidence. An operations person checks a deployment condition. A product or business owner decides whether a delay is acceptable.

Each handoff can lose context. A finding without the affected version is hard to judge. A passed test without its scope can be falsely reassuring. An exception without an owner or an expiration date can quietly become permanent. A deployment note without a rollback condition is not much help at 4:45 on a Friday.

The answer is not to make every small change endure a ceremonial review board. It is to decide, before the release rush, what information must accompany a change and who is allowed to make the call when that information is incomplete.

Build a release record people can actually use

For a modest internal application or a vendor-managed customization, the record may be a ticket, pull request, or short release note. The format matters less than the questions it answers:

  • What is being released? Name the version, environment, and material components or configuration changes.
  • What evidence was produced? Link the relevant test, review, dependency, or deployment results, and state their scope.
  • What remains open? Describe material findings, known limitations, and the conditions under which an exception is acceptable.
  • Who accepted the remaining risk? Record the person or role with authority to decide, plus any expiry or follow-up date.
  • What happens if the change misbehaves? Identify the monitoring signal, rollback or recovery path, and the person who will act.

These are not five new pieces of paperwork if the team already has useful evidence. They are a way to connect information that otherwise lives in separate tools, inboxes, and memories. The release record should point to the original evidence, not become a handwritten copy that immediately goes stale.

As an illustrative record, a release ticket might link version 2.4 to its test run and identify one unresolved dependency finding. The acceptance entry would name the authorized reviewer, the compensating condition, and the date for reconsideration. A successful run for version 2.3 would not establish what was checked in 2.4. Nor would a signed exception prove that its stated condition actually holds.

Exceptions need an arrival time and a departure time

Exceptions are where a delivery schedule becomes visible. A team may have a legitimate reason to ship with a known limitation: a compensating control exists, a fix is being tested, or the affected path is not enabled. What matters is that the explanation reaches the accountable decision-maker before the change goes live.

It also needs an ending. An exception should identify the condition it covers, its owner, and the event or date that forces a new decision. Otherwise, the organization has not accepted a bounded risk; it has created an undocumented feature of its operating model.

This is distinct from declaring a security problem “fixed” after the fact. Post-incident remediation asks whether a patch, cleanup, and investigation are complete. A release record asks a different, earlier question: based on what we know today, can this exact change proceed under these conditions?

Make the decision legible

Small organizations often have an advantage here. The people who understand the customer impact, the technical change, and the operational constraint may be close enough to have a useful conversation. But that only helps if the relevant evidence reaches them in time.

Before the next release, choose one change and trace the information it needs from code or configuration through deployment. See whether a person outside the immediate technical team could identify the build, the checks, the open condition, the decision-maker, and the recovery plan. If not, the security work may be happening. Its evidence just has not arrived where the decision is made.

Permalink

Assorted Links — October 5, 2026

  1. Cloudflare adds account-level views for investigating abuse — Early Access dashboard links login and signup activity; vendor claims.
  2. GAO maps federal support for manufacturing innovation — A new visual overview of Manufacturing USA and performance measurement.
  3. FTC and states challenge Lens.com’s advertised prices — Their complaint alleges mandatory fees hidden during checkout.
  4. Microsoft introduces a Business Premium security partner promotion — The 7.5% partner offer does not guarantee a customer discount.
  5. Commerce award recipients have an eRA transition date — NIST’s April notice sets October 1 for required use.
  6. NASA’s October guide covers the Orionids and Pleiades — Moonlight affects meteor viewing; the Moon meets the cluster October 27–28.
  7. HERMES spots network slowdowns that outage monitors can miss — Researchers combine speed tests and traffic paths; reachability alone misses performance problems.
  8. RFC Editor adds subject tags, shared reading sets, and notifications — Some features require an account; useful for following standards and corrections.
Permalink

Three Ways Small Businesses Can Approach AI

For many small businesses, AI feels both important and frustratingly vague.

Owners and managers hear constantly about what AI is capable of. They see demonstrations, read headlines, and watch larger companies invest heavily in it. But the practical question is much harder:

How do we actually figure out where AI fits in our business?

For a small business, that usually requires bringing together three things:

Knowledge of the business. Knowledge of AI. The ability to implement.

The problem is that those three things rarely exist in the same place.

Broadly, I think small businesses have three ways to approach the problem.

1. Figure it out internally

The first option is to learn about AI yourself and look for opportunities inside the business.

This approach has one major advantage: you know your business better than anyone else.

You know which processes are inefficient, what employees spend too much time doing, where customers get frustrated, and which tasks have always been harder than they should be. An outside person may need weeks of conversations just to uncover things that are obvious to you.

The challenge is AI knowledge.

Someone inside the company has to understand enough about what AI can do to connect those business problems with possible solutions. Then someone has to actually build, configure, integrate, and maintain whatever you decide to use.

For a 20-, 50-, or even 100-person company, that can be difficult. There may not be anyone whose job it is to experiment with AI, and the people capable of doing it usually already have plenty of other responsibilities.

Advantages: Deep knowledge of your own business, greater flexibility and control, and the potential to create something highly specific to the way you operate.

Tradeoffs: Requires time, curiosity, AI knowledge, technical ability, and someone willing to own the project through implementation.

2. Bring in outside AI expertise

The second approach is to hire someone to help.

A good consultant or outside firm can bring something most small businesses don’t have internally: a broader understanding of what AI can currently do.

They may have seen similar problems in other companies and can help identify opportunities you wouldn’t have recognized yourself.

But the knowledge imbalance works in the opposite direction.

They understand AI. You understand your business.

So a significant part of the process becomes transferring enough knowledge about your company to the outside firm for them to make useful recommendations.

There is also an important distinction between identifying an opportunity and implementing it.

It is relatively easy to produce a list of ideas for how a company might use AI. It is much harder to redesign a workflow, connect systems, train employees, handle security concerns, and make the new process actually work every day.

That doesn’t make consulting a bad approach. It just means businesses should understand what they’re buying. A strategy report and a working solution are very different things.

Advantages: Access to AI expertise, outside perspective, and potentially much faster identification of useful opportunities.

Tradeoffs: Cost, the time required to teach someone your business, and the possibility that implementation becomes a much larger project than identifying the idea.

3. Use something someone else has already figured out

The third path is increasingly interesting: use a packaged AI-enabled product or service.

This could be entirely new software, an AI-enabled service, or simply a new capability inside software you already use.

In this case, someone else has already done much of the work.

They identified a specific problem. They determined how AI could help solve it. And, importantly, they built a product or service you can evaluate. You still have to check its fit, data handling, integrations, and implementation requirements.

Instead of asking:

“How could we use AI to solve this?”

you’re asking:

“Is this already-solved problem important to our business?”

That’s a very different question.

Imagine a product built specifically to automate one narrow accounting, customer service, sales, legal, or administrative workflow. The company behind it doesn’t need to understand everything about your organization. It needs to understand that particular problem extremely well.

If the problem matches yours and the product works as claimed, much of the difficult AI work may already have been done.

Potential advantages: Faster adoption, easier implementation, lower internal expertise requirements, and lower cost than building something custom, depending on the product and workflow.

Tradeoffs: Less customization and control. You’re largely accepting someone else’s definition of the problem and their way of solving it. And, of course, the solution is only valuable if the problem it solves is actually important to your business.

There isn’t a universally right answer

These three approaches aren’t mutually exclusive.

A company might use packaged AI tools for several straightforward functions, work internally on something unique to its business, and bring in outside expertise for a particularly complicated opportunity.

And each approach involves tradeoffs.

The table is a rough comparison, not a promise about any particular project. Integration, sensitive data, and unusual workflows can change the cost and effort.

Approach Business Knowledge AI Knowledge Implementation Effort Customization Typical Speed
1. Figure it out internally High Must develop internally High High Slower
2. Bring in outside expertise Must be transferred High Medium to high High Medium
3. Use a packaged solution Needed mainly to judge fit Built into the solution Lower Lower Faster

The ideal situation would be someone who understands your company deeply, understands the rapidly changing capabilities of AI, and can implement the solution economically.

For most small businesses, that person or team doesn’t exist.

So the more useful question may not be:

“What should our AI strategy be?”

It may be:

“Which parts of this do we want to figure out ourselves, which parts should we get help with, and which problems has someone else already solved for us?”

For many small businesses, answering that question is probably a much more practical place to start.

Permalink

The 75-Cent Spy

A 75-cent discrepancy sent astronomer Clifford Stoll looking for a computer user. The printer records soon pointed far beyond the laboratory he was trying to protect.

Read post : The 75-Cent Spy

Before You Change HVAC Contractors

Before You Change HVAC Contractors

When you replace an HVAC contractor, put the building’s login credentials on the handover list alongside maintenance records, warranties, and the date of the next service visit.

Ask the outgoing company to identify every way it can reach the system. That might include a manufacturer’s portal, a remote connection to a computer in the building, or an administrator account used by its technicians. Find out who controls those accounts and what has to happen for the incoming contractor to take over.

Do this while everyone is still answering the phone.

A contractor change is a useful time to work through these details because the people involved have an immediate reason to cooperate. Someone needs to accept responsibility for the equipment. Someone needs to demonstrate that it works. Access belongs in that conversation.

Find out whose account it is

Consider a possible arrangement: a contractor installed a heating-control system several years ago and registered the management portal using an employee’s email address. The customer pays for the service and owns the equipment, but password resets go to the contractor’s employee.

The arrangement may have worked perfectly well. Trouble begins when the employee leaves, the contractor changes, or the customer needs to make an adjustment without calling for help.

Before the handover, establish whether the business has its own administrator access, who receives recovery messages, and whether the manufacturer requires a formal transfer. Ask the incoming contractor to explain any account or subscription it expects to control.

A leased building adds another conversation. The property manager may own the system, while the tenant depends on its operation. In that case, the tenant needs a named contact, an escalation route, and a clear account of who can authorize changes. Buying another service contract will not settle those responsibilities.

The same exercise applies to badge readers, cameras, alarm notifications, and environmental monitors. Each may have a different installer, portal, and support arrangement. Start with the systems whose failure would interrupt work or affect access to the building.

Test the handover before closing it

A list of usernames is a poor substitute for watching the new arrangement work.

Have the incoming provider demonstrate an authorized routine task with the person responsible for the building. Depending on the equipment, that could mean reviewing a schedule, retrieving a configuration backup, or checking where a notification goes. Confirm that the business can reach support and that recovery messages arrive at an address it controls.

Then review the old access. Individual accounts, shared credentials, remote-support software, and manufacturer permissions may need different treatment. A password change in one portal does not tell you whether a separate remote connection remains available.

Coordinate removals with the people responsible for operating the equipment. Abruptly disabling an account can create its own problems if it supports an integration or an essential service. Record what was removed, what remains, and the reason for any temporary exception.

NIST’s September 2026 draft of its Guide to Operational Technology Security provides context for this work. It includes building automation and physical-access systems within operational technology and addresses security alongside the equipment’s reliability and safety requirements. The draft also expands guidance on asset management and protection of system-management functions. It remains a draft open for public comment.

For a smaller business, a contractor handover is a manageable opportunity to put that thinking into practice. Facilities knows how the equipment behaves. The contractor knows how to service it. IT can help examine accounts, authentication, and network access. Bring those people together around a specific system and a specific change.

Keep the resulting record with the service agreement. Include the equipment and location, account owner, support contacts, access methods, and the date the transfer was tested. Store passwords in the organization’s approved password manager.

Before closing the job, ask the person responsible for the building to explain what they would do if the system stopped responding tomorrow morning. They should be able to find the right contact, reach the necessary records, and describe any approved manual fallback. If the answer still depends on calling the former contractor, there is another item to finish.

NIST SP 800-82 Rev. 4, Guide to Operational Technology Security, initial public draft

Permalink