Highlights
- New world: a tiny world inhabited by a single character, the Void Realm — one mind pulled out on its own, so we can actually watch whether its memory, emotion, and self-image are working.
- New · Deep Replay: you can now look inside a character's head — what she's thinking right now, how she feels, what she remembers, what she's forgotten, what she's aiming for — step by step along the timeline, with a pure-text trace to read it all through.
- Polish: after the first real simulation was viewed in a browser, nine experience issues were fixed (character placement, drifting during conversation, camera follow, cooking position, and more).
Why build a world with only one person in it
Uva Village ran 19 characters at once. Lively — but when you want to confirm whether the memory, self-image (ego), and forgetting machinery is actually working, the noise is overwhelming: you can't tell which character or which step went wrong.
So this release goes the other way: a minimal stage, the Void Realm — black void all around, one house and a patch of grass in the middle, home to a single genuinely-thinking character, Lina. She keeps all of her original personality, skills, and memories; she's just moved house. Now every move and every thought of hers is legible, so the whole cognitive system can be checked on its own.
Who interacts with her: visitors from the void
A lone resident isn't enough — someone has to talk to her and jog her memory. So we set up a few "probe" visitors: they don't think for themselves (they're scripted characters), they simply step in from the edge of the void on schedule and ask Lina pre-set questions — "what did your mother leave you?", "have you used that magic?".
The key: only the questions are scripted; Lina's answers are entirely her real mind at work — she actually retrieves memories, weighs her relationship with the visitor, composes a reply, and genuinely forms new memories from the conversation. That lets us watch directly which memories stay sharp, which blur, and which are gone for good.
Deep Replay: seeing a character's inner world
To surface all of that "looking inside the head," the viewer gains Deep Replay. Click Lina in the world replay and her inner panel opens on the right:
- her current emotion, inner monologue, present goal, and the action she's doing;
- the memories formed this step, and the older ones slowly fading over the days;
- her standing settings — identity, skills, perception, daily schedule — folded away and expandable on demand;
- and a pure-text step-by-step trace to read her whole day's inner journey end to end.
Ordinary characters (with no inner data) simply show a text replay and are unaffected.
Nine polishes after the playtest
Once the first real simulation (a full in-game day on a real AI) had been watched in the browser, we cleaned up a batch of experience issues: characters no longer float in the cracks between tiles ("wall-walking"), both parties now hold still during a conversation instead of drifting, different characters can move at different speeds (a knight moves faster — but still walks around obstacles, never through walls), cooking walks up to the stove as a real "use this object," camera follow is smoother, and more. One principle we hold to: a character's behavior is always decided by its own "mind" — never a hardcoded "if in the kitchen then cook" rule. Hardcoding is reserved for world and physics rules only.
Technical Notes
Of the five version axes, this release advances world, system, replay runtime, and replay bundle — the cognitive logic is untouched (cognition stays v1.0.2):
| Axis | v1.0.3 | v1.0.4 |
|---|---|---|
| world (world_id) | uva_village/v1.0.2 | void_realm/v1.0 |
| system | v1.0.3 | v1.0.4 |
| cognition | v1.0.2 | v1.0.2 |
| replay runtime | v1.0.2 | v1.0.4 |
| replay bundle | v1.0.2 | v1.0.4 |
The Void Realm is entirely additive: all of Uva Village's old replays play unchanged; Deep Replay's data is an optional extra file, and old replays play fine without it. In the completed 144-step real run, all five core cognition goals were met — she sleeps in the bed, cooks at the kitchen stove, reflects in Chinese, and forms over a hundred new memories across the day.
What's Next
The next release (v1.0.5) keeps polishing what the Void Realm surfaced: smoothing out the residual jitter in the follow camera, and introducing a concept of object occupancy — so "lying down on the bed" actually puts the character on the bed, and sitting actually sits on the chair (instead of standing beside it), while still never clipping through furniture.