npc_world · v1.0
NPC / World — Persistence & Reactions
How persistent are NPCs and world reactions under repeated observation?
Published protocol — results that use it must reference protocol_id + protocol_version. Old results stay bound to the version they were tested under.
Method steps
1. Fix location, time, crowd density setting. 2. Perform same player action (e.g., honk, bump) 10 times. 3. Observe NPC reaction category per window. 4. Record via video, categorize reactions. 5. Compute mode + count.
Equipment
Stock vehicle / controller vs KBM as stated; frame-counted video.
Controls
Location, time, weather, repetition interval, camera.
Variables
Player action.
Environment
Commercial district, noon, clear, medium density.
Sample size
10 observations per condition
Limitations
Single district, single time window; world simulation may vary by region/time.
Retest policy
Retest within 14 days of AI/world patch or map change.
No fabricated values
This page shows method only. Measured numbers live in Lab results and are bound to
TestRun + Measurement + GameVersion + Platform. When no shipping-build measurement exists, population is BLOCKED_EXTERNAL.Version history: protocol changes create a new row (new version), not an in-place edit. See test_protocols.version.