Rules that survive a change of owner
Protections that survive a change of management, a bad quarter, a new owner, or a decision that the program should start paying for itself.
THE CASE
Why this needs to be a standard.
A single company could build a private version of this without asking anyone. This page answers it directly, and sets out what a participating business might get, what a reward is allowed to be, and what survives if the portable version never clears review.
THE NECESSITY OBJECTION
Any company could build this alone.
The objection is correct, and it is the one worth taking most seriously. A hotel group with its own staff, its own records, and its own loyalty platform could start scoring its own customers without permission from anyone. Nothing in this proposal is needed to do that, and some companies already hold the data to try.
The reason to write a standard is that the private version keeps none of the protections. A company scoring its own customers has no reason to publish what counts, exclude spending, hide the worker's answer from that worker's manager, show the customer the record, or stop a low result from quietly shaping service. Each of those constraints costs the company something and returns it nothing, so a private system will not choose them. The standard makes them a condition of taking part.
THE COMPARISON THAT MATTERS
The alternative already exists, and it infers.
Customer conduct is already being scored, and not by anything like this. Call-center platforms are marketed as classifying emotion from voice, face, and word choice in real time, with industry coverage describing coverage of every interaction rather than a sample, and some products naming the specific emotions they claim to detect. Whether those products measure what they claim is a separate question from whether they are sold and deployed. Nobody asked the customer, no worker made a judgment, and the person being classified is not told the classification exists.
That is the actual competing design. The choice a business faces is not between this proposal and nothing. It is between a bounded question a person answers and a model that reads a face. The comparison is what makes a standard worth writing, because the second option requires no consent, no disclosure, and no standard at all.
Regulators have reached the same conclusion about the mechanism, though more narrowly than a headline would suggest. Article 5(1)(f) of the EU AI Act, Regulation (EU) 2024/1689, prohibits AI systems that infer the emotions of a person in the workplace and in education institutions, except for medical or safety reasons. It has applied since 2 February 2025 and sits in the instrument's highest penalty tier. Emotion inference aimed at customers rather than workers was not banned but reclassified as high-risk on 2 August 2026, with a transparency duty owed to the person it is applied to.
Those rules leave more unregulated than they reach, in three specific ways. The prohibition reaches emotion inferred from biometric data, so inference drawn from word choice or a transcript falls outside it. Enforcement is early, with no public action against a prohibited practice announced as of March 2026. And the instrument is European, with no federal equivalent in the United States, which is where a first pilot would most likely sit.
If regulators are already restricting emotion inference, a standard looks redundant. The two instruments do different work. A prohibition removes a mechanism. It does not publish what may be counted, require that a worker's answer stay hidden from that worker's manager, give a person a record they can inspect and contest, or put anything in the consumer's hands that they choose whether to offer. Law can forbid the bad version. It cannot specify the good one, and it does not appear to be trying.
What this proposal does is refuse the mechanism in its own rules. The NCS specification requires that every response be a deliberate act by a person who was present, and prohibits facial analysis, voice and tone analysis, sentiment classification, affect detection, and biometric signals as inputs, including as suggestions offered to a worker. Responses may not train a model that infers emotion or conduct about anyone. That is a constitutional prohibition rather than a preference, and it is the specific thing a private system would have no reason to adopt.
This section describes what published instruments say and what products are marketed as doing. It is not a claim that this proposal prevents emotion inference, or that a standard would slow its adoption. Neither has been demonstrated.
WHAT A PRIVATE VERSION CANNOT HOLD
Five things only a standard holds.
A worker's answer kept from their employer
No company writes that rule for itself. Without it, the response becomes a management instrument and the worker learns to answer accordingly.
A record the customer can contest
Correction, review, and escalation matter most in the cases a company would prefer not to reopen.
One definition of conduct everywhere
Without a shared definition, a person is judged by a different private rule at every business, and none of those rules is inspectable.
History that can travel between companies
History that moves between companies requires a common definition and an independent referee. This is the part that cannot exist without a standard at all.
WHAT A PARTICIPATING BUSINESS MIGHT GET
Six hypotheses a pilot would need to test.
Section 17 of the working specification records these as value hypotheses. Public materials cannot promise reduced turnover, lower support cost, higher retention, or revenue lift without evidence, and no evidence exists.
Recognition beyond spend
A way to notice customers who use a service responsibly without being high spenders, who are currently invisible to every loyalty tier.
Better working conditions at the counter
Lower frequency or severity of harmful interactions, and a better working experience for the people who absorb them.
A starting point with a brand-new customer
A first interaction currently starts from nothing, which pushes businesses toward treating every unknown person as a risk.
One shared definition instead of many private deals
One shared definition rather than a separate private arrangement negotiated with every partner and platform.
A clearer view of the friction the business causes
Business-caused friction has to be recorded to be excluded, which surfaces problems that customer-value systems currently hide.
Human relationships matter more as service automates
As agents handle more of the transaction, the remaining human relationship becomes the part worth distinguishing.
The specification also settles who is allowed to capture that value. Pricing and governance must not let a platform or the largest participating brand take the gains while consumers, workers, and smaller businesses carry the cost and the risk.
If a pilot measures these hypotheses and finds nothing, that is a result rather than a setback. The stop conditions exist for exactly that outcome.
WHAT A REWARD CAN BE
A reward can be almost anything.
A company should be able to decide how generously to recognize the customers its own people most want to work with. The catalog is deliberately open. A handwritten thank-you, a waived deposit, a late checkout, an invitation, or something a company invents are all candidates. Anything scarce, an upgrade or a queue position among them, carries the higher bar set out below. Each one is approved or refused on its own terms rather than by one blanket rule, because a thank-you card and a reduced deposit raise completely different questions.
The first benefit in a pilot will still be modest, because a first pilot tests whether the mechanism is safe, fair, and understandable rather than whether the reward is appealing. That is a property of the first step rather than a permanent ceiling on what recognition can become.
Every reward is classified before it is offered
Whether it is a privilege or advantage of a public accommodation, whether it touches price, deposit, credit, or contract, whether withholding it could amount to adverse treatment, and whether it screens out people who need an accessible or assisted path.
A scarce reward needs independent evidence first
Anything drawn from limited inventory takes something from another customer, which raises fairness and competition questions rather than design preferences. A queue position, an upgrade, held inventory, and priority allocation are all scarce, and none of them can be reclassified as merely additional, whether or not any baseline service was reduced to create them.
A scarce reward opens only on published evidence that non-participants were not made materially worse off, reviewed by someone independent of the business offering it and of this project. A completed pilot is not that evidence. Neither is the absence of complaints.
Baseline service holds while a reward sits on top
Degrading the ordinary experience and then selling its restoration as recognition of good conduct is the failure mode this rule exists to catch.
THE PERMANENT BOUNDARY
Rights are never rewards.
One category never opens, at any tier, for any customer, however generous a program becomes. Complaints, refunds, corrections, disputes, appeals, accommodations, and safety responses are rights rather than rewards. Conduct cannot speed them up for one person, slow them down for another, or change who is entitled to them.
Baseline service holds on the same terms. If the standard line gets worse so that the recognized line can look better, the program has failed regardless of how much eligible customers enjoy the benefit. That is the difference between rewarding conduct and running a market in rights.
THE QUESTION THAT COULD END THE PORTABLE VERSION
The hardest objection to this.
Carrying history between companies is the part that most needs a standard and the part most likely to fail review. Cross-context transfer is where social-scoring law bites hardest, where relevance is hardest to demonstrate, and where the fairness questions are worst. Transfer already starts at zero for that reason, and an overall score sits outside the first pilot entirely.
Read those two facts together and an objection follows that this page has to answer rather than dodge. If the first pilot is company-specific, and portability may never clear, then the pilot scope is the version that needs no standard at all. A single company could run it alone.
THE ANSWER TO THAT OBJECTION
Only a standard publishes its rules.
What a standard produces, even for one company scoring only its own customers, is a published and inspectable set of rules about how a person may be scored at all. What counts and what is excluded. Whose answer stays hidden from whom. What the customer can see and contest. What the score may never be used for. Which questions remain unanswered, named as unanswered.
No private system publishes any of that, because publishing it creates obligations and returns nothing. That is the asymmetry the whole proposal rests on, and it holds whether or not history ever travels between companies. Portability would be the larger prize. The rules exist either way.
If portability never clears, Net Conduct Score still stands on its own. The bounded question, the four responses, the safe abstention, the exclusions, and the worker protections do not depend on a portable score existing. A business, an agency, a client relationship, or an event could use them to give people a structured voice about how they were treated.
That is a smaller project than a trust network. It would still be worth publishing, and it would still be a standard rather than a feature, because the rules are what make it one.
HELP TEST THIS CASE
Where this case needs a harder look.
Two things would sharpen this page most: a reason a business would decline to take part, and a way to reach the same result without a standard. Both are most visible to people who have built loyalty or trust systems before.
Add to the case