Progress / Phase 7

Phase 7: Crowds and a Larger World

A town square holds hundreds of players who can all see each other, with each player’s bandwidth held to a budget, on a much larger map where spatial queries drive chat, combat and AI.

Phase 70 / 20 · 0%

This phase is still ahead.

Ships: v0.7.0
Lesson: Spatial queries, bandwidth budgets, priority and update-rate tiers, and graceful degradation under overload
Depends on: Phase 6

7.1 Spatial queries everywhere

0 of 3 done

Say chat range and combat events use the same spatial queries.
Mob aggro, leashing and AI perception use the zone’s spatial grid instead of scanning every entity.
Benchmarks of the grid’s queries with 1,000 players and 1,000 mobs in one zone.

7.2 Bandwidth budgets for crowds

0 of 4 done

Per-player byte budget with a priority accumulator: nearby, moving and targeted entities first, everyone else as the budget allows (scaling.md Case 3).
Update-rate tiers by distance (every tick, every second tick, every fourth tick), and a cap on full-rate entities per player.
Graceful degradation under overload: as a zone’s tick nears its budget, the server lowers update rates for distant entities first and logs it; game rules never degrade.
Tests: the budget is never exceeded, every visible entity is updated at least at its tier’s minimum rate, and nothing flickers at area edges.

7.3 A larger map

0 of 3 done

Build a large map (about 200×150 tiles) with a town, a forest edge and ruins, as the first real zone layout.
Decision Measure client rendering and memory on the large map, and decide whether chunked loading is needed.
Minimap showing nearby terrain and other players.

7.4 Art Batch 07: zone tiles

0 of 2 done

Draw Batch 07 by hand: forest variants, town building exteriors, lanterns, ruins, water and bridges.
Intake and integrate.

7.5 Measurement

0 of 3 done

make loadtest SCENARIO=crowd at Case 3 Tier 1 and Tier 2 (100 and 300 players in one area) on dev-m5, before and after the budgets, in docs/perf/phase-07.md.
Case 3 Tier 3 (1,000 in one area) run as a degradation test: the server stays up, holds its budgets and recovers when the crowd disperses.
Client frame rate with 300 players on screen measured, and fixed if it can’t hold 60 FPS on dev-m5.

7.6 Walkthrough

0 of 2 done

docs/walkthroughs/phase-07-crowds.md: why fan-out is O(n²), spatial partitioning, hysteresis, priority and bandwidth budgets, and degrading gracefully.
Devlog draft, release notes, tag v0.7.0 and publish builds.

Acceptance criteria

0 of 3 done

300 players in one area meet Case 3 Tier 2 in scaling.md on dev-m5: tick p99 under 25 ms, at most 32 KB/s per player, every visible entity updated at least 5 times a second.
Entities don’t visibly pop in or out at area edges during normal movement.
Server tick p99 stays under 12.5 ms with 100 bots spread across the large map.