Creating the physics for exploration within Unity

Design analysis for Golgotha (a Zelda-like in Unity 6 / URP) on how to evolve its world from discrete portal-connected scenes toward a continuous, ride-across-the-land exploration feel (the BOTW-on-horseback fantasy). This is a direction/architecture decision, not code yet.

Where the world is today

• Each region / sky zone / dungeon is a separate Unity scene. RealmManager.CrossTo unloads the current content scene and additively loads the target, with a blend to hide it.
• Net effect: “rooms connected by doorways.” Great for building breadth fast (we have 4 ground regions
  • 5 sky zones + 5 dungeons chained) and correct for dungeons (which should feel enclosed). The thing that breaks the open-world feel is the teleport at a region boundary — you touch a portal and blink, you don’t ride across a plain.

The core reframing

Two separable problems, solved by different techniques:

1. Actual continuity — walk A→B with no load screen.
2. Perceived scale — does it feel vast regardless of real size. You can buy ~90% of the “feel” (perceived) cheaply and defer the expensive “truly seamless streaming” (actual). Raw dimensions matter far less than perception.

The honest constraints (cost tradeoffs, not hard caps)

• Memory: can’t hold the whole world resident at once → this is what forces streaming.
• Float precision: past ~5–10 km from origin things get jittery → solved by a floating origin (silently recenter the world on the player) or chunked coordinate spaces.
• Iteration/build cost: a streaming + LOD + occlusion pipeline is real ongoing engineering; the expensive option for a small team.

What actually sells “near-infinite” (perception levers)

• Atmospheric depth — fog/haze compressing how far you can see (we just shipped gradient skyboxes + per-realm fog; this is literally the tool, already in place).
• Visible distant landmarks pulling you forward (we already have the DressFar far-backdrop system).
• Density of things to find (owlet rescues, shrines, dungeons) — a world feels big by content, not size.
• Traversal speed — a mount makes a medium world feel expansive.

Staged path (cheapest → most expensive)

1. Visible next-region proxies (cheap, no streaming): drop the adjacent region’s low-detail silhouette on the horizon at each boundary — see the Maplewaste from the Drowned See’s shore. Direct extension of DressFar. Does most of the psychological “it’s continuous” work for near-zero cost.
2. Seam-stream the region boundaries (medium, the real payoff): evolve CrossTo so that nearing the A→B boundary co-loads B additively alongside A at the correct world offset; the player walks across; then A unloads behind them. Turns portals into “crest the ridge into the next land.” The hard parts already exist (additive load/unload in RealmManager, cross-scene state in WorldStateBus); new work is co-loading two scenes with a coordinate offset + a trigger volume instead of a teleport, and hiding the swap in a natural choke point (canyon/bridge/fog bank). This is the single biggest jump toward the horseback feel without a full streaming engine.
3. Mount + larger authored regions (medium): grow regions from ~±40 m toward ±150 m and add a mount/sprint. Sky-glide already gives the vertical version of the fantasy.
4. Full streaming / LOD / floating-origin (expensive): only for genuine BOTW-scale seamlessness across the whole map at once. Defer until 1–3 prove the feel is worth it.

Recommendation & gotchas

• Do #1 + #2 first: visible destinations + seam-streamed boundaries → converts every “doorway” into “ride over the hill and the next land opens up,” riding on infrastructure we already have.
• Gotcha to respect: seam-streaming touches load-bearing plumbing (RealmManager, coordinate offsets, save-state while two scenes are co-loaded), so unlike the fast additive scene-builds we’ve been doing, it warrants a careful prototype + tests (“measure twice”). #1 is a safe quick win to feel it immediately.