The AI dialog — terms for designers and BAs

Who this is for: the designer who owns the page and the analyst who owns the number, sitting in a meeting where engineers say "it fails silently on the event bus" and everyone nods. This page is the context nobody hands you. Each entry: what the word means, plainly, and — where a term is a framework — the parameters it actually evaluates, because a framework without parameters is a vibe.

An honesty note: some of these terms are industry-standard (system of record, event bus, hydration). Some are ours (judgment-based loading, the helmet, chain pricing). Each entry says which. Definitions are written to be used in a meeting, not to survive a computer-science exam.

1. The moving parts

Fragment (ours) — a section of a page, authored once in Webflow, that other platforms display verbatim by fetching it. Not a copy: a feed. Edit the original, and every platform shows the edit within a minute.

Symbol (Adobe's design term; Webflow's legacy name for what it now calls Components; our cross-platform usage) — in the design tools (Illustrator, XD, Animate), a reusable element defined once: edit the symbol, every instance updates. Webflow adopted the word, then renamed its version Components — same idea, nav and footer defined once. Developers know the word differently: in Ruby, a symbol (:name) is an immutable, unique identifier — a name interned once, so every reference points to the same single thing and comparison is instant; JavaScript's Symbol primitive makes the same guarantee. The two senses are secretly one: a symbol is one canonical identity that is referenced everywhere and copied nowhere. That is exactly the cross-platform usage in this stack — the nav's DOM id is the symbol; AEM, Shopify, and every other consumer hold references to it, never copies of it.

Hydration (industry) — the moment JavaScript "wakes up" static HTML: the page paints first as an inert document, then scripts attach the working parts — the login avatar, the consent buttons, the live prices. "Conversions that hydrate" means the conversion signal exists only if that wake-up step runs and succeeds. Nothing about the visible page tells you whether it did — which is why the term matters to a BA: the page can look perfect while the number it was supposed to send never left the browser.

Event bus (industry) — the in-page announcement channel. When something meaningful happens — consent granted, user signed in, a purchase completed — an event is announced, and any listening component reacts. It replaces the old world where every signal was assumed from a page view. The catch: an announcement no one wires up is never made, and there is no error for silence.

Function, not SaaS-bound (ours, as a distinction) — a small piece of server-side logic that runs on the edge (a Cloudflare Worker) and belongs to you — as opposed to a feature that lives inside a SaaS vendor's product and works only while you subscribe. A SaaS feature is rented behavior; a function is owned behavior. The practical difference appears the day you leave: functions travel, features don't.

A function on the server vs. an if/then on the client (the distinction under everything) — an if/then on the client is a decision made inside the visitor's browser: it evaluates whatever the browser happens to know, it can be skipped by an ad blocker or a script that never loaded, and when it fails, it fails silently. A function on the server runs where the record lives: it cannot be skipped by the browser, it consults the authoritative row rather than a local guess, and when it fails, it fails loudly — a status code, a log line, an alarm. The division of labor follows: presentation may be decided on the client; authority, money, and consent must be decided by a server function. Every silent failure in §5 is, at root, an authority decision that was left to a client if/then.

The helmet (ours) — the first script on every page. It loads the protective layer before anything else: consent defaults (denied until granted), the event bus, and then — only as needed — everything else. Named for what it does: nothing else gets on the field without it.

2. The decision layer

Intent (framework — parameters required) — what a subject is trying to do, stated precisely enough to be judged. Its parameters: subject (which human or agent), target (which page, section, or record — usually an address like #segment-hero), scope (what may change if it succeeds), and horizon (when the intent expires). An intent without these four is a wish.

Judgment (framework — parameters required; ours) — a decision made at the moment of action instead of configured in advance. The six parameters every judgment in this architecture evaluates:

1. Consent state — does permission exist right now?
2. Entitlement row — is this subject granted this capability on this asset? (A grant is a database row: insert to grant, delete to revoke.)
3. Demonstrated need — does the page or task actually require what is about to load or run?
4. Mandate scope and cap — under whose authority, bounded by what spending or action limit?
5. Freshness — is the record being consulted still inside its truth window (a price older than a minute is not a price)?
6. Ledger obligation — will the outcome be recorded where an audit can find it?

Judgment-based loading (ours) — the helmet applying those parameters to JavaScript: each script loads only when the judgment passes, instead of firing because someone configured a trigger last quarter. The same idea applied to integrations is the AI tool runner: a tool call happens only when the row says yes, now, under this mandate.

Mandate (industry term, our parameters) — the signed permission an AI agent acts under. Parameters: scope (what it may do), cap (how much it may spend), rail (which payment method), expiry, and revocation (one row deleted, agent stopped).

3. The record

System of record (SoR) (industry) — the one place a fact is authoritative: writes land there first, disputes resolve there. Here, that is Xano — the rows for identity, consent, entitlement. Everything else — every page, cache, CRM copy, share card — is a projection: a display copy that can be regenerated at any time precisely because it is never the original.

Latent collision (ours) — a decision made now against a record that arrives later: the ad bid placed on last week's inferred conversion, the agent acting on a permission revoked an export ago. Latency turns records into liabilities at exactly the moment they're consulted. The cure is structural, not heroic: the record and the decision must share a clock.

4. The money

Chain pricing vs. SLA pricing vs. the ceiling tier (ours) — three ways to be charged for your own data. SLA pricing: the data sits in the vendor's silo and you pay for the storage plus the promise of access. The ceiling tier (Zapier, n8n, Airtable): data is metered per motion with a monthly ceiling — the busier your business, the more your own data costs, and the month you grow is the month the workflow pauses. Chain pricing: the record stays in your system of record, and fees attach only where operated work happens. The tell: ask what happens at the ceiling, and what happens when you leave.

Boundary before the fee (ours) — consent and entitlement are evaluated before any commercial event and independent of the plan. No fee unlocks a bypass; no lapsed invoice switches the gate off. Trust is not a tier.

5. The failure, and the fix

Silent failure (industry term, sharpened) — a failure with no alarm, because the metric everyone watches keeps moving. The canonical case: a client-side-only Shopify store. Checkout is Shopify-hosted, so revenue never stops — while the analytics events, plugin writes, and consent history quietly go dark. Orders complete; the data plane starves; the damage books as "market conditions" weeks later. Rule of thumb for any review: a failure that pays out at checkout has no alarm attached to it — go looking.

Privacy streaming (ours) — the course correction: logging where every event is consented before it fires, redacted at the source, streamed in real time, and ledgered on arrival. Four revenue effects: bidding algorithms get true conversions instead of inferences; budget follows observed signals; a ledgered stream that stops is visible (the alarm silent failures lack); and a consented stream is the only one that keeps flowing as browsers and regulators tighten. Unconsented logging is a signal with an expiry date.

Arrow, not line (ours) — every connection in the system points one way and says who moved, toward what, under whose judgment. Four verbs cover all of it: rails pull, planes hydrate, events emit, tools call. If a diagram shows a line instead of an arrow, ask which direction the data actually flows — the answer is the audit.

The hash (industry mechanism, our framing) — the #section-name at the end of a URL. One id is four addresses at once: where a fragment is extracted from, the deep link a human lands on, the target a share link is minted against, and the @id an answer engine cites. The same address resolving identically for crawler, human, and agent is the cheapest verifiable claim on the web.

AEO (industry: answer-engine optimization) — being findable and citable by AI answer engines, not just ranked by search. In practice: structured data with stable addresses, so when an engine answers a question it can point at your exact section — and an agent can act on what it cited.

6. The wrap — security, privacy, and AI judgment on demand

Read back through the entries and they collapse into three sentences.

Security is not a product here; it is the shape of the arrows — every connection one-directional, every arrow authority-checked against a row, no standing credentials parked anywhere for an attacker to live in. Privacy is not a checkbox; it is the order of operations — consent evaluated before any signal fires, redaction at the source, the gate standing prior to and agnostic to any fee. And AI judgment on demand is the mechanism that makes both real: the six parameters of judgment, evaluated at the moment of action — load time for a script, call time for a tool, render time for a price — instead of configured in advance and trusted to stay true.

That is the whole dialog. When the meeting uses these words, this is what they are supposed to mean — and if a vendor uses them without parameters, now you know what to ask.

Companion: Fragments on Any Frontend — the architecture these terms come from, demonstrated live on AEM and Shopify.