Skip to content

Map: 64u grid — pixel-exact lanes from the full-res pathing map - #13

Merged
goetchstone merged 1 commit into
mainfrom
map/full-res-lanes-64u
Jul 9, 2026
Merged

Map: 64u grid — pixel-exact lanes from the full-res pathing map#13
goetchstone merged 1 commit into
mainfrom
map/full-res-lanes-64u

Conversation

@goetchstone

Copy link
Copy Markdown
Owner

Summary

Owner-directed: deduce the lanes from the full-res map. The original is authored on WC3's 64u grid — at 64u cells the pathing map quantizes with zero fractional cells, so the new 162x226 grid is a pixel-exact representation. New hard gate G6 asserts lane-topology equivalence (22 anchors, pairwise water-connectivity) against the raw 32u pathing map on every regeneration. Steering constants are now world-unit-derived.

Test plan

  • All extractor gates + NEW G6 lane-topology equivalence (0 mismatches)
  • Full suite 1,245 green; probes verified (first tower leak ~tick 5000, steady chip after)
  • Owner's verified fixes hold at 64u (lane separation, Library bottom-lane dock)
  • CI
  • Live playtest

🤖 Generated with Claude Code

Owner-directed: "the full res map is accurate if you can deduce the lanes
correctly." Measured, the original is authored on WC3's 64u grid (at 64u
cells the wpm quantizes with ZERO fractional cells), while real channels
bottom out at 128u and real walls at 64u — widths a 128u grid provably
cannot represent when unaligned (channels lost) or sub-cell (walls gone,
lanes merged). The emitted grid is now 64u (162x226, EMIT_SUBDIV=2): a
pixel-exact copy of the original's enforced ship-pathing.

- terrain.py: geometry subdivision; land-biased tie sampler (dormant here —
  the map aligns — but correct for any future non-aligned source) + a
  provably-merge-safe straddled-channel restore pass; island/access
  constants scaled (moat cores 9x9, cycleLen 40, single entrances hold);
  NEW HARD GATE G6: pairwise water-connectivity of all 22 gameplay anchors
  on the emitted grid must equal the raw 32u wpm's — any lost channel or
  false merge fails the build.
- Sim: cell-denominated steering constants converted to world units
  (dock-approach hand-off 128u, nav local-goal basin 768u, derived from
  mask cellSize) so grid resolution is a data property, not behavior.
- Probes: bot-vs-bot long-run window 6000->9000 ticks — on the true tighter
  lanes the first tower-chipping leaker lands ~tick 5000 (verified steady:
  3.5k tower hits by tick 24000, 10 captains); all other windows unchanged.
- Owner-verified fixes hold at 64u: NE/east lanes separate; the SE Library
  docks bottom-lane SW; corner-reach probe still reaches every quadrant.

Full suite 1,245 green (trader haul, creep clash/grind, AI siege, fog/
cliffs, determinism replays); lint clean.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
@goetchstone
goetchstone merged commit 2c9da8a into main Jul 9, 2026
1 check passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant