How ReCreate decides
Last updated: August 3, 2026
ReCreate makes a handful of automatic decisions for you: what your Today feed shows first, which events match your search, how close a nearby person actually is, what gets held for review, and how often you’re notified. This page says, in plain terms, exactly how each one works — including the parts that are still simple rules, not AI.
Your Today feed
Today’s tile order can be handled by an assistant, reading a fresh, per-user summary of your own upcoming events, RSVPs and groups — never anyone else’s data. It reorders and can hide tiles, most-pressing-first, but it operates inside a hard floor it can’t cross: your hero card can never be hidden, and at least one other tile must stay visible. A reply that would leave the board all-hidden is thrown away before you ever see it — the app falls back to a plain rule instead (hero pinned first, everything else sorted by how soon it starts).
It only re-runs when something about your day actually changed since last time, it’s rate-limited per person, and it’s metered against a shared spending cap — hitting either limit is invisible to you, it just quietly falls back to the same deterministic rule. This reordering may still be rolling out for your account; the fallback is identical either way, just without the assistant’s judgment call about what’s pressing today.
Matching your filters to events
Filtering isn’t a relevance score — it’s a strict, ordered checklist. An event has to clear every step to show up at all: not already past → host isn’t someone who blocked or excluded you → inside your radius → inside your date window → inside your time-of-day window → matches your category → matches your text search → matches your guest/adult setting → clears the sensitive checks below → has room left. Fail any step and the event is gone, full stop — there’s no “close enough” tier.
Age, gender-preference and group-eligibility filters are re-checked on our servers, not just applied from data your phone already has. Your phone asks the server which of these events you’re actually eligible to see, and the server answers under your real account permissions — the same permissions that would apply if you tried to join directly. While that answer is loading, the map shows nothing for those events rather than a guess that might be wrong in either direction.
How close is that person, really?
Never your exact distance. When ReCreate tells you how far away someone or something is, it starts from the true distance and adds a small, consistent randomization — up to about a third of a mile — before rounding to the nearest half-mile (never below half a mile, even for something genuinely close by).
“Consistent” is the important part: the same viewer looking at the same target always gets the same number back, so the distance stays useful for planning — but repeatedly asking, or asking from a different account, can’t be used to triangulate someone’s precise location, because the randomization is tied to that specific pair of people, not generated fresh each time. Your exact coordinates are read only by features that inherently need them — a map pin you’ve chosen to publish, a live share you’ve turned on. Your home location and live position are never included in anything the app computes in aggregate.
The Explore heat layer — “is it alive here?”
- Never your location data — it never reads your home address or your live location, full stop. The only inputs are event venues that are already public and check-in timestamps, never a location table.
- Broad cells only — activity is bucketed into geographic cells no finer than roughly neighborhood-size, and never resolves to street level.
- A 5-person floor — a cell only appears once at least 5 distinct people (hosts plus people who actually checked in) are behind it. Below that, ReCreate shows nothing for that cell — never a smaller, more identifying number.
Reports and moderation
The report queue is triaged, but nothing automated hands down a real penalty. A classifier labels each report to help a human prioritize — that’s all it can do; it can’t approve, dismiss, or clear anyone’s name. The one thing it can act on directly is spam it’s extremely confident about, and even then the only action is a hold pending human review, never a deletion or an account action. Several different people reporting the same thing in a short window can also trigger an automatic, reversible hold. A real penalty — a warning, a pause, a ban — is always a human decision.
Notifications
Notifications are batched so a burst of activity reads as one useful message instead of a wall of pings: fast-moving activity on something you host collapses into a single “N new” push roughly every 30 minutes rather than one push per person. Every notification carries a stable identifier for what it’s about, so a duplicate write can never create a second push — and on your device, a repeat notification replaces the last one instead of stacking underneath it.
One more thing
The AI-assisted parts above fold back to a plain, deterministic rule the moment they’re unavailable for any reason — there’s always a working fallback, never a blank screen. There’s also one master AI switch in the app’s Settings that turns every AI-powered surface off at once. The full technical version of this page — file paths, table names, exact formulas — lives in the app’s own Help surface and in the open-source help corpus alongside it.