Does disclosure need to happen at the start of every interaction, or only when asked?

Proactive disclosure is the more defensible position for interactions above the Disclosure Threshold, because waiting to be asked places the burden of discovering the truth on the customer rather than on the business that has the information. For interactions below the threshold, waiting to be asked, or not disclosing at all, is reasonable, since the customer's actual concern is the outcome and a proactive disclosure there mostly adds friction without adding value the customer was seeking.

How does the Disclosure Threshold interact with Deterministic Outcome classification?

They can share the same underlying work. The classification process that determines whether a decision class has a verifiable, evaluable outcome — and is therefore a candidate for autonomous promotion per the criteria in The Last Approval — naturally also indicates its disclosure requirement. A Deterministic Outcome decision class typically sits below the Disclosure Threshold because the customer can verify the result themselves regardless of who produced it; a decision class requiring genuine judgment typically sits above it, because the customer cannot fully verify fairness on their own and is relying on trust in the process instead.

What is the actual risk of disclosing too much, rather than too little?

Unnecessary friction and, in some cases, an undeserved erosion of confidence in interactions that were actually handled correctly. Disclosing autonomous operation on every routine, low-stakes interaction can create a background hum of doubt that adds no value, since the customer's concern in those interactions was never about who handled it. The risk is real but asymmetric with the risk of under-disclosure: over-disclosure costs some friction; under-disclosure at a consequential moment costs trust that is far harder to rebuild.

Can a business set its Disclosure Threshold differently across different customer segments?

The consequentiality test should apply consistently regardless of which customer is asking, because the risk of a badly discovered concealment doesn't diminish for any particular segment — if anything, a business's highest-value or most vocal customers are the ones most likely to notice and publicise an undisclosed autonomous decision that went badly. What can reasonably vary by segment is the form the disclosure takes, not whether it happens: an enterprise customer evaluating vendor risk as part of procurement may want a more detailed disclosure than a consumer customer who is satisfied with a brief, clear statement, but both are owed disclosure once the interaction crosses the same consequentiality threshold.

How should a business communicate its Disclosure Threshold policy externally, if at all?

A general, public statement of the principle — that consequential decisions involve disclosed review processes while routine ones may not require explicit notice — is more defensible than either silence on the topic or an attempt to enumerate every specific interaction type. This mirrors the same logic behind stating a Redundancy Dividend claim only once it has been tested: a general, honest policy statement that matches actual practice is a durable trust asset, while an overly specific public commitment risks becoming a claim the business cannot consistently honour as its product and decision classes evolve.

Does the Customer Need to Know?

Concealing autonomy feels safer today. A badly discovered truth is more damaging than an early disclosure.

Neither Steward Experience nor Retention Reflex takes a position on a question every autonomous business eventually faces directly: should the customer be told that no human is involved in the interaction they're having. The instinct to avoid the question — to let the interaction speak for itself and hope the customer doesn't ask — is understandable and, on its own, not a strategy. This memo takes the position the rest of the corpus has left open.

The customer cares about the outcome, not the actor — until they don't

For a large share of interactions, disclosure is genuinely unnecessary because the customer's actual concern is resolved regardless of who or what resolved it. A password reset, a delivery status check, a routine billing question — these have a Deterministic Outcome: a verifiable, correct answer that either happened or didn't. A customer whose password was reset correctly and quickly has received exactly what they wanted; whether a human or an agent performed the reset adds nothing to their assessment of the interaction, and forcing a disclosure here — "a bot just did that" — creates friction and a moment of doubt with no corresponding benefit to the customer.

The threshold shifts, and shifts hard, once the interaction involves genuine judgment, empathy, or a consequential decision the customer cannot fully verify themselves. A customer disputing a charge, negotiating a refund, seeking advice on a decision with real financial or personal stakes, or receiving a denial of some kind is not just seeking an outcome — they are implicitly trusting that a reasonable, accountable process produced that outcome. Concealing that no human weighed the specific circumstances of their case, in this category of interaction, is not a neutral omission. It is withholding information the customer would reasonably want in order to assess whether the outcome was actually fair to them.

Why concealment fails specifically at the moment it matters most

The risk asymmetry is the strongest argument for taking a clear position rather than deciding case by case under pressure. If a business conceals autonomous operation in low-consequence interactions and the concealment is eventually discovered, the cost is minor — a slightly awkward realization with no real stakes attached. If a business conceals autonomous operation in a high-consequence interaction — a denied claim, a disputed charge, a significant account decision — and that concealment surfaces later, through a regulatory inquiry, a lawsuit, or simply a customer comparing notes with someone who received a different-seeming explanation, the damage compounds specifically because the stakes were real. The customer does not just learn they spoke to an agent. They learn the business let them believe a person had weighed their specific situation, at exactly the moment that belief mattered most.

This connects to the discipline established in Redundancy as a Feature, Not Just Insurance: a claim discovered to be false at the worst possible moment is far more damaging than no claim at all, because the failure converts an unearned assumption into a broken promise. Concealed autonomy at a consequential decision point is the same structural risk applied to trust rather than to an uptime SLA.

A practical test, not a blanket rule

The Disclosure Threshold is not "disclose everything" or "disclose nothing" — both are wrong in different directions. The practical test: if the customer, on learning after the fact that no human was involved, would reasonably feel misled about something that mattered to them, the interaction sits above the threshold and requires disclosure. If the customer, on learning the same fact, would reasonably shrug — because the outcome was correct and verifiable regardless of who produced it — the interaction sits below the threshold and disclosure is optional, driven by brand voice rather than trust obligation.

This test also gives Full-System Design a concrete specification point: the same Decision Execution Autonomy classification work that determines whether a decision class is promoted to autonomous operation can simultaneously determine its disclosure requirement. A decision class with a Deterministic Outcome, verified and promoted per the criteria in The Last Approval, sits naturally below the Disclosure Threshold. A decision class that remains at genuine human judgment, or one whose outcome is not fully verifiable by the customer themselves, sits above it.

The Operator's Verdict

Deciding whether to disclose autonomous operation should not be a live judgment call made under the pressure of an individual interaction. It should be specified once, at the same design stage as the Intervention Threshold and the decision-class promotion criteria, using consequentiality as the test rather than convenience or a general preference for concealment. A business that discloses correctly at the moments that matter builds trust it can rely on later. A business that conceals correctly at the moments that don't waste no customer's attention on a distinction they never needed to care about.

Technology changes what the interaction can resolve without a person. Disclosure determines whether the customer trusts that it was resolved fairly.

KEY TAKEAWAY

When should an autonomous business disclose to a customer that no human is involved in their interaction?

The Disclosure Threshold is the point in a customer interaction, determined by the consequentiality of the decision rather than the channel it occurs on, at which failing to disclose autonomous operation becomes a trust liability rather than a neutral design choice. For interactions with a Deterministic Outcome — a password reset, a delivery status check — the customer cares about the correct outcome, not who or what produced it, and disclosure is optional. For interactions involving genuine judgment, empathy, or consequential decisions the customer cannot fully verify themselves — a disputed charge, a denied claim, a significant account decision — concealing autonomous operation withholds information the customer would reasonably want to assess whether the outcome was fair, and the risk of that concealment surfacing later is asymmetric: low-stakes concealment discovered later is minor, high-stakes concealment discovered later compounds because the customer learns the business let them believe a person had weighed their specific situation at exactly the moment that mattered. The practical test: if the customer would feel reasonably misled upon learning the truth, disclose; if they would reasonably shrug, disclosure is optional. This should be specified once, using the same Deterministic Outcome classification already used to determine decision-class promotion, not decided live under pressure. Source: Arco Venture Studio.