refactor: remove distant terrain LOD

.agents/skills/test-writer/SKILL.md

@@ -41,7 +41,6 @@ modules/
   world-core/       # chunk, block, block_registry, coordinates, lighting
   world-worldgen/   # terrain generation, biomes, caves, decorations
   world-meshing/    # chunk_storage, chunk_mesh, meshing/, GPU block buffers
-  world-lod/        # LOD chunks, meshes, scheduler, renderer, manager
   world-runtime/    # world facade, streaming, renderer, mutation, GPU mesher
   world-persistence/# level data, region files, save manager
 src/

.github/workflows/benchmark.yml

@@ -14,11 +14,6 @@ on:
         description: Benchmark duration per preset (seconds)
         required: false
         default: "60"
-      capture_gpu_culling:
-        description: Capture the bounded gpu-culling-scale extreme/traversal CPU-vs-GPU source pair
-        required: false
-        type: boolean
-        default: false
 
 concurrency:
   group: ${{ github.workflow }}-${{ github.ref }}
@@ -104,7 +99,7 @@ jobs:
           name: Benchmark
           timeout: ${{ github.event_name == 'schedule' && '55m' || github.event_name == 'workflow_dispatch' && '55m' || '20m' }}
           log-file: benchmark.log
-          command: mkdir -p "$MESA_SHADER_CACHE_DIR" && devenv shell --profile graphics -- bash scripts/run_benchmark.sh --duration "$BENCHMARK_DURATION" --presets low,medium,high --scenarios stationary,traversal,rapid-turn,teleport-eviction --compact-modes off,auto --benchmark-world overworld --output-dir benchmark-results --per-preset-timeout 600
+          command: mkdir -p "$MESA_SHADER_CACHE_DIR" && devenv shell --profile graphics -- bash scripts/run_benchmark.sh --duration "$BENCHMARK_DURATION" --presets low,medium,high --scenarios stationary,traversal,rapid-turn,teleport-eviction --benchmark-world overworld --output-dir benchmark-results --per-preset-timeout 600
         env:
           XDG_RUNTIME_DIR: /tmp/runtime-runner
           WAYLAND_DISPLAY: headless
@@ -114,58 +109,13 @@ jobs:
           # PR/push runs are bounded canaries. Scheduled and manually requested
           # captures are longer acceptance evidence, not synthetic fixtures.
           BENCHMARK_DURATION: ${{ github.event_name == 'schedule' && '60' || github.event_name == 'workflow_dispatch' && github.event.inputs.duration || '5' }}
-          ZIGCRAFT_BENCHMARK_GPU_ADAPTER: Lavapipe (Mesa software Vulkan)
-          ZIGCRAFT_BENCHMARK_GPU_DRIVER: Mesa Lavapipe supplied by the pinned devenv profile
-          ZIGCRAFT_BENCHMARK_RUNNER: blacksmith-2vcpu-ubuntu-2404 GitHub Actions runner
-          ZIGCRAFT_BENCHMARK_ZIG_TOOLCHAIN: Zig 0.16.0 supplied by the pinned devenv inputs
-
-      - name: Run Phase 5 long-session streaming stress gate
-        if: steps.gate.outputs.run == 'true'
-        uses: ./.github/actions/run-with-log
-        with:
-          name: Phase 5 Streaming Stress
-          timeout: 15m
-          log-file: phase5-streaming-stress.log
-          command: devenv shell --profile unit -- zig build phase5-stress-gate -Dphase5-stress-iterations="$PHASE5_STRESS_ITERATIONS"
-        env:
-          # The scheduled run is longer by operation count, never by a flaky
-          # elapsed-time threshold. PR feedback stays bounded.
-          PHASE5_STRESS_ITERATIONS: ${{ github.event_name == 'schedule' && '768' || '64' }}
 
-      - name: Validate compact benchmark evidence matrix
+      - name: Validate benchmark results
         if: steps.gate.outputs.run == 'true'
-        run: python3 scripts/benchmark_baseline.py validate-compact-matrix benchmark-results
-
-      - name: Capture dedicated CPU-vs-GPU culling sources
-        # Scheduled captures always retain a fresh pair. Manual runs opt in so
-        # ordinary canaries stay fast; each source is independently bounded.
-        if: steps.gate.outputs.run == 'true' && (github.event_name == 'schedule' || (github.event_name == 'workflow_dispatch' && github.event.inputs.capture_gpu_culling == 'true'))
-        uses: ./.github/actions/run-with-log
-        with:
-          name: GPU Culling Baseline Sources
-          timeout: 32m
-          log-file: gpu-culling-benchmark.log
-          command: mkdir -p "$MESA_SHADER_CACHE_DIR" && devenv shell --profile graphics -- bash -c 'scripts/run_benchmark.sh --duration 60 --presets extreme --scenarios traversal --compact-modes auto --gpu-culling off --gpu-culling-threshold 128 --benchmark-fixture gpu-culling-scale --benchmark-horizon-distance 4096 --benchmark-lod-memory-budget-mb 2048 --benchmark-require-gpu-candidates 1024 --benchmark-world flat --output-dir gpu-culling-results/cpu --per-preset-timeout 900 && scripts/run_benchmark.sh --duration 60 --presets extreme --scenarios traversal --compact-modes auto --gpu-culling on --gpu-culling-threshold 128 --benchmark-fixture gpu-culling-scale --benchmark-horizon-distance 4096 --benchmark-lod-memory-budget-mb 2048 --benchmark-require-gpu-candidates 1024 --benchmark-world flat --output-dir gpu-culling-results/gpu --per-preset-timeout 900'
-        env:
-          XDG_RUNTIME_DIR: /tmp/runtime-runner
-          WAYLAND_DISPLAY: headless
-          MESA_SHADER_CACHE_DIR: /tmp/mesa_shader_cache
-          SDL_AUDIODRIVER: dummy
-          ZIGCRAFT_SAFE_MODE: "1"
-          ZIGCRAFT_BENCHMARK_GPU_ADAPTER: Lavapipe (Mesa software Vulkan)
-          ZIGCRAFT_BENCHMARK_GPU_DRIVER: Mesa Lavapipe supplied by the pinned devenv profile
-          ZIGCRAFT_BENCHMARK_RUNNER: blacksmith-2vcpu-ubuntu-2404 GitHub Actions runner
-          ZIGCRAFT_BENCHMARK_ZIG_TOOLCHAIN: Zig 0.16.0 supplied by the pinned devenv inputs
-
-      - name: Validate fresh CPU-vs-GPU culling pair
-        if: steps.gate.outputs.run == 'true' && (github.event_name == 'schedule' || (github.event_name == 'workflow_dispatch' && github.event.inputs.capture_gpu_culling == 'true'))
         run: |
-          python3 scripts/benchmark_baseline.py assemble-gpu-culling --overwrite \
-            --output gpu-culling-results/gpu-culling-baseline.json \
-            gpu-culling-results/cpu/auto/extreme/traversal.json \
-            gpu-culling-results/gpu/auto/extreme/traversal.json
-          python3 scripts/benchmark_baseline.py validate-gpu-culling \
-            gpu-culling-results/gpu-culling-baseline.json
+          for result in benchmark-results/*/*.json; do
+            bash scripts/validate_benchmark_artifact.sh --result "$result"
+          done
 
       - name: Stop headless Wayland compositor
         if: always() && steps.gate.outputs.run == 'true'
@@ -179,20 +129,6 @@ jobs:
             exit 1
           fi
 
-      - name: Compare against baseline
-        # A 5-second canary must not be numerically compared with a 60-second
-        # acceptance capture. The compatibility command also enforces this.
-        if: steps.gate.outputs.run == 'true' && github.event_name != 'schedule' && github.event_name != 'workflow_dispatch'
-        id: compare
-        run: |
-          if ! python3 scripts/benchmark_baseline.py compatibility docs/benchmarks/baseline.json benchmark-results/auto/low/stationary.json --preset low --scenario stationary; then
-            printf 'Skipping numeric baseline comparison: checked reference hardware/provenance differs from this CI runner. Matrix, SLO, validation, and Bencher gates still apply.\n'
-            exit 0
-          fi
-          for preset in low medium high; do for scenario in stationary traversal rapid-turn teleport-eviction; do
-            bash scripts/compare_benchmarks.sh docs/benchmarks/baseline.json "benchmark-results/auto/${preset}/${scenario}.json" --preset "$preset" --scenario "$scenario"
-          done; done
-
       - name: Install Bencher CLI
         if: steps.gate.outputs.run == 'true' && env.BENCHER_API_TOKEN != '' && env.BENCHER_PROJECT != ''
         uses: bencherdev/bencher@30a740a2e4246560b1a5fd424057d84aa2b188d6
@@ -210,7 +146,7 @@ jobs:
               --token "$BENCHER_API_TOKEN" \
               --adapter json \
               --err \
-              "cat benchmark-results/auto/${preset}/${scenario}.json"
+              "cat benchmark-results/${preset}/${scenario}.json"
           done; done
 
       - name: Comment Bencher dashboard
@@ -232,10 +168,7 @@ jobs:
           name: benchmark-results
           path: |
             benchmark-results
-            gpu-culling-results
             benchmark.log
-            gpu-culling-benchmark.log
-            phase5-streaming-stress.log
             weston.log
           retention-days: 30
 
@@ -244,12 +177,10 @@ jobs:
         uses: actions/github-script@v9
         env:
           BENCHMARK_RUN_OUTCOME: ${{ steps.run_benchmark.outcome }}
-          BENCHMARK_COMPARE_OUTCOME: ${{ steps.compare.outcome }}
         with:
           script: |
             const runOutcome = process.env.BENCHMARK_RUN_OUTCOME;
-            const compareOutcome = process.env.BENCHMARK_COMPARE_OUTCOME;
-            const failed = runOutcome === 'failure' || compareOutcome === 'failure';
+            const failed = runOutcome === 'failure';
             const description = failed
               ? 'Benchmark regression or runtime failure'
               : 'Benchmark completed';

.github/workflows/build.yml

@@ -200,14 +200,6 @@ jobs:
           log-file: unit-test-${{ matrix.optimize }}.log
           command: devenv shell --profile unit -- zig build -Doptimize=${{ matrix.optimize }} test
 
-      - name: Run Phase 5 compact LOD gate
-        if: needs.changes.outputs.code_changes == 'true' && matrix.optimize == 'Debug'
-        run: devenv shell --profile unit -- zig build phase5-gate
-
-      - name: Run bounded Phase 5 streaming stress gate
-        if: needs.changes.outputs.code_changes == 'true' && matrix.optimize == 'Debug'
-        run: devenv shell --profile unit -- zig build phase5-stress-gate -Dphase5-stress-iterations=64
-
       - name: Upload unit test log
         if: failure()
         uses: actions/upload-artifact@v7
@@ -291,38 +283,15 @@ jobs:
           ZIGCRAFT_SMOKE_FRAMES: "3"
           ZIGCRAFT_SAFE_MODE: "1"
 
-      - name: Run Phase 5 production visual-motion and saved-world gate
-        if: needs.changes.outputs.code_changes == 'true'
-        uses: ./.github/actions/run-with-log
-        with:
-          name: Phase 5 Visual Motion and Saved-World Reload Gate
-          timeout: 35m
-          log-file: phase5-visual-gate.log
-          command: devenv shell --profile graphics -- zig build phase5-visual-gate
-        env:
-          XDG_RUNTIME_DIR: /tmp/runtime-runner
-          WAYLAND_DISPLAY: headless
-
-      - name: Upload Phase 5 visual evidence
-        if: always() && needs.changes.outputs.code_changes == 'true'
-        uses: actions/upload-artifact@v7
-        with:
-          name: phase5-visual-evidence
-          path: |
-            zig-out/phase5-visual-smoke/
-            phase5-visual-gate.log
-          if-no-files-found: ignore
-          retention-days: 14
-
       - name: Fail on Vulkan validation log errors
         if: needs.changes.outputs.code_changes == 'true'
         run: |
           set -euo pipefail
-          if rg -n -e 'Vulkan validation error|Validation Error:|VUID-' integration-test.log world-smoke-test.log phase5-visual-gate.log; then
-            echo "Vulkan validation errors were found in integration or Phase 5 visual logs." >&2
+          if rg -n -e 'Vulkan validation error|Validation Error:|VUID-' integration-test.log world-smoke-test.log; then
+            echo "Vulkan validation errors were found in integration logs." >&2
             exit 1
           fi
-          echo "No Vulkan validation errors found in integration or Phase 5 visual logs."
+          echo "No Vulkan validation errors found in integration logs."
 
       - name: Stop headless Wayland compositor
         if: always() && needs.changes.outputs.code_changes == 'true'
@@ -336,7 +305,6 @@ jobs:
           path: |
             integration-test.log
             world-smoke-test.log
-            phase5-visual-gate.log
             weston.log
           if-no-files-found: ignore
           retention-days: 7

.gitignore

@@ -30,8 +30,6 @@ zigcraft-minidumps/
 *.spv
 !assets/shaders/vulkan/lpv_inject.comp.spv
 !assets/shaders/vulkan/lpv_propagate.comp.spv
-!assets/shaders/vulkan/lod_compact_terrain.frag.spv
-!assets/shaders/vulkan/lod_compact_water.frag.spv
 !assets/shaders/vulkan/water.frag.spv
 !assets/shaders/vulkan/water.vert.spv
 wiki/

AGENTS.md

@@ -14,7 +14,7 @@
   devenv shell zig fmt src/ modules/
   devenv shell zig fmt --check src/ modules/
   ```
-- `zig build test` is the broad suite: aggregate/module tests, fuzz roots, shader compilation/validation, SPIR-V size checks, shadow ABI checks, and Phase 5 policy tests.
+- `zig build test` is the broad suite: aggregate/module tests, fuzz roots, shader compilation/validation, SPIR-V size checks, and shadow ABI checks.
   ```bash
   devenv shell zig build test
   devenv shell zig build test -Dtest-filter="name"
@@ -25,11 +25,9 @@
   ```bash
   devenv shell zig build test-integration -Dskip-present=true
   devenv shell zig build test-robustness
-  devenv shell zig build phase5-gate
-  devenv shell zig build phase5-visual-gate
   ```
 - `-Dskip-present=true` only suppresses presentation; SDL/Vulkan initialization still needs a display/compositor and driver. Use the repo skills `headless-crash-test`, `headless-screenshot`, `headless-benchmark`, or `headless-graphics-verification`, and always bound game/graphics commands with a timeout.
-- For deterministic startup checks, combine `-Dskip-present`, `-Dauto-world=<normal|overworld|overworld-v2|flat|test>`, and `-Dstartup-diagnostic-seconds=N`. `-Dchunk-debug-mode` disables LOD, water, caves, and decorations; selectively restore `lod,water,watergen,waterrender,caves,decorations` with `-Dchunk-debug-enable=`.
+- For deterministic startup checks, combine `-Dskip-present`, `-Dauto-world=<normal|overworld|overworld-v2|flat|test>`, and `-Dstartup-diagnostic-seconds=N`. `-Dchunk-debug-mode` disables water, caves, and decorations; selectively restore `water,watergen,waterrender,caves,decorations` with `-Dchunk-debug-enable=`.
 
 ## Benchmarks and generated artifacts
 
@@ -40,28 +38,27 @@
     -Dbenchmark-duration=60 -Dbenchmark-output=zig-out/benchmark-low.json
   ```
   Scenarios are `stationary`, `traversal`, `rapid-turn`, and `teleport-eviction`. Prefer the `headless-benchmark` skill for bounded runs.
-- Focused CPU-only tools: `devenv shell zig build worldgen-report` and `devenv shell zig build lod-bench`. Pass climate snapshot arguments after `--`, e.g. `devenv shell zig build worldgen-climate-snapshot -- --seed 42 ...`.
+- Focused CPU-only tools: `devenv shell zig build worldgen-report` and `devenv shell zig build worldgen-climate-snapshot`. Pass climate snapshot arguments after `--`, e.g. `devenv shell zig build worldgen-climate-snapshot -- --seed 42 ...`.
 - Building/tests compile GLSL and write tracked `*.spv` files beside sources in `assets/shaders/vulkan/`. After intentional shader-size changes, run `./scripts/update_spirv_baseline.sh`; `docs/shaders/spirv-sizes.json` and shadow runtime SPIR-V parity are test-enforced.
 - New/changed textures go through `./scripts/process_textures.sh <pack> 512`; preserve licensing/attribution for placeholder assets.
 
 ## Architecture boundaries
 
 - `src/main.zig` and `src/game/app.zig` wire the executable. Reusable session/player/settings/benchmark logic belongs in `modules/game-core`; screens belong in `modules/game-ui`.
 - `engine-rhi` owns rendering contracts and opaque handles; `engine-graphics` owns the Vulkan backend, render passes, resources, and pipelines. Vulkan is currently the only backend, so interface changes usually require both the RHI vtable/contracts and Vulkan implementation.
-- `world-core` owns shared block/chunk/coordinate types; world generation is split across `world-worldgen-*` packages; `world-meshing` owns chunk mesh/storage; `world-lod` owns distant terrain; `world-runtime` coordinates streaming, lighting, GPU meshing, mutation, and rendering; `world-persistence` owns saves/regions.
+- `world-core` owns shared block/chunk/coordinate types; world generation is split across `world-worldgen-*` packages; `world-meshing` owns chunk mesh/storage; `world-runtime` coordinates streaming, lighting, GPU meshing, mutation, and rendering; `world-persistence` owns saves/regions.
 - Never call RHI/SDL from worker jobs. Pin chunks while background jobs hold them and synchronize shared mesh/chunk state with the project mutex abstraction.
 - Convert global coordinates with `worldToChunk`/`worldToLocal` from `world-core`; chunk coordinates are floor-divided, so ad-hoc truncating division is wrong for negative positions.
 - GPU-shared structs and shader blocks are ABI contracts. Use explicit layouts (`extern`/packed as appropriate), assert sizes/offsets, and update CPU, GLSL, descriptors, and backend bindings together.
 
 ## Graphics hazards
 
-- `ZIGCRAFT_LOD_COMPACT=auto` is qualified only through immutable per-layer descriptors plus expanded fallback for unsupported shoreline topology; `off` remains the explicit supported fallback. `force` is diagnostic only. Do not weaken the saved-world RADV reload, validation, robustness, and Phase 5 qualification requirements based only on generated-world captures.
 - Dedicated transfer queues cannot upload exclusive graphics buffers safely with semaphore synchronization alone; queue-family release/acquire ownership transfers are required. Until implemented, keep these uploads on the graphics family.
 - A descriptor set update affects all previously recorded commands that bind that set when execution happens. Terrain/water or direct/indirect streams that need different buffers require immutable/per-layer descriptor sets, not sequential updates to one set.
 
 ## Verification and CI
 
-- Match verification to the change, but graphics/RHI/shader/runtime work normally requires: format check, `zig build test`, ReleaseFast build, integration/robustness as relevant, and the appropriate headless graphics skill. CI additionally runs Debug and ReleaseSafe tests, `phase5-gate`, Lavapipe/Weston integration smoke tests, and `phase5-visual-gate`.
+- Match verification to the change, but graphics/RHI/shader/runtime work normally requires: format check, `zig build test`, ReleaseFast build, integration/robustness as relevant, and the appropriate headless graphics skill. CI additionally runs Debug and ReleaseSafe tests and Lavapipe/Weston integration smoke tests.
 - Install the repo pre-push hook with `./scripts/setup-hooks.sh`, but do not treat it as CI parity: it formats only `src/` and omits ReleaseSafe and graphics jobs.
 
 ## Git workflow