Ethos Declaration

The cheerful decline Link to heading

Maya has had a streak of bad luck. In the last sixty days she’s returned three different things: winter boots that didn’t fit (approved), a blender that arrived cracked (approved), and now a made‑to‑order dining table (declined). The app’s message is upbeat: “Heads up—custom builds are tricky.” On a good day that tone might soften the blow. Today it lands wrong. She already knows the policy. She doesn’t need pastel empathy; she needs a reason she can act on and a clean path to appeal.

That mismatch isn’t just tone. It’s a failure of stance. Same app, different situation, and the product behaved as if context didn’t matter. That gap has a name here: ethos.

An app doesn’t have a face. It has a character—learned over time by how it behaves when stakes rise. In philosophy, Paul Ricoeur called this narrative identity: who we are is the story we keep together through our kept promises 1 . In psychology, narrative coherence and future‑self continuity show up in whether people trust a system and follow through on decisions 2 . Translate that to software and the question becomes straightforward: does the product explain itself when it matters, match tone to the situation, and give real recourse? That’s how character shows up in code.

What I’m proposing is that we declare that character up front and wire it into the system stack so components consult it at runtime to adapt.

Why an ethos declaration Link to heading

Without a declared ethos, products drift. The marketing page promises care; the support flow delivers chirpiness; the moderation tool hides the reason your post lost reach. Teams didn’t mean to lie, they shipped speed. An Ethos Declaration is how you bind fast UX to slow commitments so the product behaves like the same “person” when stakes rise. It turns a few promises into rules the runtime can actually keep: when novelty and irreversibility spike, explain first; when outcomes disappoint, show criteria and a channel for voice; when ranking changes, disclose the lens and one alternative 1 2 [7].

This isn’t a design system (form) or a feature flag (on/off). A design system chooses typography; a feature flag toggles a component; an ethos chooses manners under conditions. The conditions are ordinary: first‑time versus routine actions, low versus consequential risk, novice versus expert, different jurisdictions. In ecommerce that means a high‑ticket first purchase gets a slower tempo and a short rationale next to the button; a routine reorder stays fast. In social it means that if a post is age‑gated or demoted, the interface shows the policy label, the top signals, and an appeal path with a timeline. The goal isn’t to slow everything; it’s to match pace to stakes using the simple gearshift you already know—uncertainty × novelty × reversibility—so judgment shows up exactly where it pays off.

Manners at the edges Link to heading

People judge systems by how they behave when things don’t go their way. Law and psychology have plain names for this. Contextual integrity says behavior should fit the room: who is involved, what’s being exchanged, and the norms of that situation 3 . Fair process says disappointment feels legitimate when the path is clear: what rule applied, whether you had a voice, whether the tone stayed respectful [6]. These aren’t soft ideas; they’re how legitimacy is felt in practice.

Two effects make the edges heavy. Negativity bias means a single mishandled denial can outweigh a week of smooth checkouts [4]. Affect labeling shows that naming the moment plainly—“We don’t refund made‑to‑order pieces once they ship”—reduces distress and helps people act [5]. Put together, the design rule is modest and strict at once: be serious when it’s serious. Not dour—proportionate.

In ecommerce, that looks like this: a first‑time, consequential denial uses one clear line, the two reasons that matter, and a visible appeal path with an ETA. If history shows several recent returns, spare the pep; surface remediation and appeals quickly, and keep the tone neutral. In social, when reach drops or a post is restricted, the interface names the policy, shows the top signals that fired, offers the edit that would help, and provides an appeal route you can actually find. None of this is theater. It is how you stop a product from performing friendliness while violating the manners that make trust stick.

What an ethos declaration is (and what it isn’t) Link to heading

Start with the claim. An ethos is a short list of kept promises and a set of contexts that turn those promises into live rules. Everything else—copy tone, visual hierarchy, what gets hidden or prominent, how errors speak, how notifications behave, even button names—flows from that center.

It is not a brand book. A brand book can tell you you’re “warm.” An ethos tells you that in refund denials you’ll be candid, not cute, and that you’ll always show criteria and an appeal link. It is not a pile of feature flags. Flags flip things on; ethos governs how the thing behaves: tempo for first‑time versus routine, explanation depth for consequential versus routine, visibility for reversal paths, disclosure for ranking and summarization [7]. It is not a virtue list. “Candor” isn’t a policy. “Show the top two reasons before non‑reversible actions” is.

If you want scaffolding, use these words, but keep them concrete. Character gives designers a north star (“competence over cuteness when stakes rise”) [9][10]. Commitments are statements you can test in code and in research. Contexts are rooms (checkout, refund, share, appeal) with overlays (region, age). Policies map “if this, then that” for tempo, tone, explanation, visibility, and recourse. Micro‑patterns are the tiny, repeatable moves—error copy that blames the system first, appeal links where the loss happens, reversal paths next to the button. Governance is how you keep yourself honest: a small scenario suite, telemetry for adherence and exceptions, and a change log you actually write.

The point of all this is not control. It’s coherence—the feeling that the same product met you again and kept its word. Coherence builds memory; memory builds trust 1 . And trust, in turn, buys you permission to be fast where it’s right to be fast.

The artifact you can ship Link to heading

A compact file at /.well-known/ethos.yaml that components consult at runtime, plus a public Ethos Card rendered for humans. Together they make your stance testable and visible.

Field notes for two domains Link to heading

These examples show how the same ethos commitments—explain before non‑reversible, disclose the lens, offer recourse—change shape by context and history. The goal isn’t to script copy; it’s to make manners legible so teams can keep character under pressure.

Ecommerce: refund and checkout rooms Link to heading

A first‑time, consequential denial earns a different register than a routine return. Begin with one clear line that names the policy, add the two reasons that actually matter, and make the appeal path visible with a real timeline. This satisfies the basic fairness test: people can see the rule, understand its basis, and speak back if needed [6]. If a customer has had several returns in sixty days, tighten the prose and surface remediation first (photo checklist, alternate payment path, callback window). It isn’t colder; it’s proportionate—and it respects that over‑soothing can feel like theater when the pattern is familiar [4][5].

At checkout, put reversibility next to the call‑to‑action on high‑ticket items (“Cancel within 24 hours”). Use the gearshift: first‑time + consequential → measured tempo and a short rationale; routine + low risk → stay fast. Error copy should be neutral, blame the system first (“We expected MM/YY”), and include the fix. These are small moves, but together they teach users what to expect in serious moments—and they keep your marketing promises honest. :contentReference[oaicite:1]{index=1}

Social platforms: moderation, ranking, sharing Link to heading

When a post loses reach or is age‑gated, people need legibility and recourse more than pep. Name the policy, show the top signals that fired, and highlight the at‑issue excerpt so the fix is not a guessing game. Offer a one‑tap appeal with an ETA. Disclose the active lens (“freshness,” “quality,” “safety”) and offer one meaningful alternate; swapping a wall of toggles for one real choice keeps agency usable. As creators demonstrate mastery, hints can recede, but the fairness trio—criteria, voice, respectful tone—stays put [6]. This isn’t a tone guideline; it’s procedure that makes authority feel earned rather than arbitrary 3 .

Micro‑patterns you can actually ship Link to heading

Micro‑patterns are the repeatable moves that carry your ethos through ordinary screens. They keep the everyday texture aligned with your promises without turning every decision into a meeting.

Use these as patterns, not templates: adjust tone by novelty, risk, and the person’s recent history; keep the fairness pieces visible [4][5][6].

Measurement that makes identity testable Link to heading

If you only optimize for conversion and time‑to‑complete, speed wins every meeting. Add counters that reveal whether you’re keeping your promises, then read them in context—not as trophies, but as coherence checks.

Start with four:

1. Commitment adherence. For each declared promise (e.g., “explain before non‑reversible”), calculate the percentage of eligible events that complied. Define eligible precisely to avoid false wins.
2. Exceptions with justification. Track exception rate and review a stratified sample monthly. Require a short justification string; trend by team and surface that in the release notes.
3. Fair‑process pulse. After consequential outcomes, ask three questions once: Did you understand the criteria? Did you have a channel for voice? Was the tone respectful? Keep it optional, randomize who sees it, and analyze by context and cohort [6].
4. 24‑hour reversals. Measure cancellations/edits within a day on hard‑to‑reverse actions. Falling reversals coupled with stable conversion is the signature of serious‑when‑serious working.

Two supporting reads help you steer rather than posture. The narrative‑coherence delta checks whether the same commitments fire in similar rooms across releases (it’s how you detect ethos drift). The ledger optics metric watches how often people open or export the Decision Ledger after consequential events and whether “Did this help you be confident?” nudges up. Align the whole program with Guidelines for Human‑AI Interaction (set expectations, support correction, make failure legible) and keep a light thread to the NIST AI RMF so these numbers sit inside a governance frame, not a dashboard island [7][8].

Edge cases (where ethos usually cracks) Link to heading

Marketing vs. terms. If the homepage promises “easy refunds” and the terms say “no refunds on anything,” you’re teaching people not to trust you. Let the Ethos Card force the merger: change the copy or change the policy. Coherence is the feature.

Tone as habit. “Playful everywhere” collapses under grief, loss, or money‑at‑risk. Encode tone escalation rules so serious rooms (denials, lockouts, appeals) default to candid and proportionate, not cute [4][5].

Repeat‑case fatigue. Ten returns in two months? Skip the pep talk. Offer remediation first and keep the path to appeal clean. Respect doesn’t always sound warm; it reads as clarity and recourse [6].

Cross‑region overlays. Teens in the EU mean tighter defaults (profiling off, explicit consent). Don’t bury changes in a policy page; put them in the declaration and make the behavior visible where it matters.

Model updates. Ranking or moderation model changes can break narrative coherence overnight. Run a scenario probe suite on release candidates and compare adherence before and after so the app’s character doesn’t shift without telling anyone [7][8].

Quick start (ship this first) Link to heading

You don’t need a giant program to make character visible. Ship these three pieces and you’ll feel the difference:

1. A one‑page narrative of the XYZ way for five rooms (checkout, high‑ticket checkout, refund denial, share seam, appeal). Keep it specific enough that a designer and an engineer can make the same call.
2. Two gated commitments in your build pipeline—e.g., “explain before non‑reversible” and “disclose lens + one alternate.” Break the build if they fail eligible scenarios.
3. A 12‑scenario probe suite that runs on every release candidate and a short, public change log that says what moved and why. Tie one paragraph of results to the fair‑process pulse so the numbers carry meaning, not theater [6][7][8].

[4]: https://assets.csom.umn.edu/assets/71516.pdf"Baumeister , R. F., et al. (2001). Bad is stronger than good. Review of General Psychology." [5]: https://sanlab.psych.ucla.edu/wp-content/uploads/sites/31/2015/05/Lieberman_AL-2007.pdf “Lieberman, M. D., et al. (2007). Putting feelings into words. Psychological Science.” [6]: https://press.princeton.edu/books/paperback/9780691126739/why-people-obey-the-law “Tyler, T. R. (2006). Why People Obey the Law (rev. ed.). Princeton University Press.” [7]: https://dl.acm.org/doi/10.1145/3290605.3300233 “Amershi, S., et al. (2019). Guidelines for Human‑AI Interaction. CHI 2019.” [8]: https://www.nist.gov/itl/ai-risk-management-framework “NIST (2023). AI Risk Management Framework 1.0.” [9]: https://www.jstor.org/stable/3151897 “Aaker, J. L. (1997). Dimensions of brand personality. Journal of Marketing Research.” [10]: https://www.researchgate.net/publication/37705092_The_Media_Equation_How_People_Treat_Computers_Television_and_New_Media_Like_Real_People_and_Pla “Reeves, B., & Nass, C. (1996). The Media Equation. Cambridge University Press.”