diff --git a/docs/research/procedural-doctrine.md b/docs/research/procedural-doctrine.md index 9c67d8f..ae49052 100644 --- a/docs/research/procedural-doctrine.md +++ b/docs/research/procedural-doctrine.md @@ -86,7 +86,7 @@ Code shape from the Bloom pass: ```python def add_hubward_endcap_terraced_slums(prefix, x, mats): # Bloom lore frame: on an endcap, "up" is inward toward the axial Spire. - rings = [0.95, 1.28, 1.62, 1.98, 2.36, 2.76, 3.18, 3.62, 4.08, 4.55, 4.95] + rings = [0.72, 0.98, 1.24, 1.52, 1.82, 2.14, 2.48, 2.84, 3.22, 3.62, 4.04, 4.48, 4.9, 5.18] for tier, radius in enumerate(rings[:-1]): inner_r = radius outer_r = rings[tier + 1] @@ -100,6 +100,11 @@ Verification cue: name representative objects after the frame they prove, then render from a view that can reveal the attachment surface. `hubward_endcap_*` objects prove something different from `surface_*` objects. +Practical rule: before implementing a lore-bound scene, write the coordinate +contract in code comments and object names. If the verifier cannot name an +object that proves the contract, the script is probably trusting vibes where it +needs geometry. + ## Use Noise For Organic Occupancy Reach for this when repetition needs to feel grown, crowded, eroded, informal, @@ -132,6 +137,10 @@ Verification cue: do not only count objects. Check density-bearing meshes for face counts, add a camera that looks at the actual domain surface, and inspect whether the pattern still reads as organic after batching. +Batching rule: when noise controls many small details, prefer one mesh per +material/class band and one multi-spline curve per line family. Thousands of +standalone Blender objects are not detail. They are a performance invoice. + ## Start With A Pipeline Reach for this when the form feels like it has stages: a seed shape, a sampling diff --git a/notes/fresh-workspace-handoff.md b/notes/fresh-workspace-handoff.md index f3cb089..c994342 100644 --- a/notes/fresh-workspace-handoff.md +++ b/notes/fresh-workspace-handoff.md @@ -47,17 +47,14 @@ zones, gizmo nodes, and rejected empty-output `WindowBeams` harness. IRCSS maps live in `docs/research/translations/ircss-french-houses-*.md`. First from-scratch build test: `examples/geometry_script/aetheria_bloom_habitat.py` -generates an Aetheria Bloom whole-habitat study using Bloom lore plus Eric -Bruneton's Rama making-of as a cylindrical-world reference. It now emphasizes -whole-cylinder mapped spaces and frame interfaces: full shell layers, despun -hub cap and docking core, spun spire sheath, shared x/angle/radius coordinate -helpers, golden-angle spoke/transfer loops, layered endcaps, mapped civic -bands, roads, rivers, cloud patches, and service seams. The hubward terrace -slums are on the endcap: noise-driven annular shelf fields around the docking -hub office complex step inward toward the Spire as "up" and grow outward into a -classier high-gravity urban crown. Plazas and towns have nested buildings, -kiosks, feeder walks, and lane networks. Kappa is only a small service marker. -Build notes live at +generates an Aetheria Bloom whole-habitat study. It now carries the key spatial +lessons: define the lore coordinate frame before ornament, use cylindrical and +endcap helpers for attachment, use golden-angle curves for transfer networks, +use deterministic noise for organic settlement pressure, batch dense detail by +material/class, and render inspection cameras that face the domain being judged. +The hubward terrace slums are noise-driven endcap shelves around the docking hub +office complex, stepping inward toward the Spire as "up" and outward into a +classier high-gravity urban crown. Build notes live at `docs/research/builds/aetheria-bloom-habitat.md`; verification lives at `tools/verify_aetheria_bloom_habitat.py`. @@ -80,10 +77,10 @@ The GameCult fork is intentionally pruned to `main` plus VibeGeometry-owned tooling branches. Do not copy upstream branch soup forward unless a future patch actually needs it. -The current next action is to continue the Aetheria Bloom build by moving the -shared cylindrical coordinate helpers plus spiral spoke/transfer-artery network -into reusable Geometry Script groups, then improving atmospheric/light -readability from render evidence. +The current next action is to extract the Bloom coordinate/noise/detail lessons +into a small reusable builder seam, starting with either endcap terrace slums or +spiral transfer arteries, then bind that seam to scene-brief, manifest, +inspection-render, and verifier contract drafts. ## Important Invariants diff --git a/notes/vibegeometry-implementation-plan.md b/notes/vibegeometry-implementation-plan.md index a1f8e41..ebeb05e 100644 --- a/notes/vibegeometry-implementation-plan.md +++ b/notes/vibegeometry-implementation-plan.md @@ -29,21 +29,34 @@ not carry live next-action clutter. script execution. - City-scale graphs expose the next toolchain pressure: repeat zones, foreach zones, gizmo nodes, repeat `Top` outputs, and converter edge cases. +- Source-bound scenes need a written coordinate contract before ornament. + Define basis axes, up/down, attachment surfaces, and frame transitions in the + script and verifier before adding detail. +- Organic distribution needs a field, not tidy placement. Use deterministic + noise to drive density, offsets, sizes, heights, material/class bands, and + gaps when the visual target is accretion, poverty, erosion, cloud, vegetation, + or informal growth. +- Dense procedural detail must be batched by material/class and line family. + Thousands of standalone Blender objects are the dumb kind of expensive. +- Render evidence must include cameras that face the domain being judged. A + beauty camera can hide the exact spatial mistake that matters. ## Near-Term Sequence -1. Choose the next IRCSS French Houses move. - - Either fix one repeat-zone/foreach/gizmo blocker in the Geometry Script - fork or select a non-zone architectural helper with a non-empty behavioral - harness. - - Keep the chosen target small enough to verify against evaluated geometry. - - Record rejected paths only when they change future decisions. +1. Promote the Bloom coordinate/noise lessons into reusable builder seams. + - Extract shared cylindrical/endcap coordinate helpers into a clear module or + Geometry Script-oriented builder layer. + - Preserve the split between coordinate contract, density field, batched + realization, domain camera, and verifier checks. + - Keep the first extraction small: endcap terrace slums or spiral transfer + arteries, not the whole Bloom at once. 2. Define the scene brief contract. - Capture references, text instructions, target visual traits, constraints, camera/composition requirements, acceptance criteria, and known unknowns. - Preserve source paths or URLs for every reference image. - - Separate visual target description from implementation strategy. + - Require a source-coordinate section for lore/reference-bound builds: + basis axes, local up/down, attachment surfaces, and known frame changes. 3. Define the node graph manifest contract. - Describe node groups, sockets, exposed controls, modifiers, attribute @@ -61,6 +74,8 @@ not carry live next-action clutter. - Store render path, scene brief id, manifest id, script path, Blender version, review status, visual misses, graph issues, and next hypothesis. - Treat rendered output as the acceptance surface for visual claims. + - Require at least one domain-facing inspection render for every major + spatial claim, not just a composed overview. 6. Promote doctrine only after it survives use. - Keep examples small enough to inspect. @@ -75,6 +90,13 @@ not carry live next-action clutter. - Add visual comparison or reviewer prompts. - Add richer examples only after the contracts stop sliding. +8. Return to IRCSS graph blockers after the scene loop has these contracts. + - Either fix one repeat-zone/foreach/gizmo blocker in the Geometry Script + fork or select a non-zone architectural helper with a non-empty behavioral + harness. + - Keep the chosen target small enough to verify against evaluated geometry. + - Record rejected paths only when they change future decisions. + ## Deferred - Large procedural asset library diff --git a/state/evidence.jsonl b/state/evidence.jsonl index ffaef57..be40e11 100644 --- a/state/evidence.jsonl +++ b/state/evidence.jsonl @@ -40,3 +40,4 @@ {"ts": "2026-05-05T00:21:08+00:00", "type": "implementation", "status": "accepted", "note": "Bloom detail pass added batched fractal favela wall detail, elaborated spoke plazas, and nested town fields; initial per-object version timed out, batching preserved visual detail and verifier passed with 1102 geometry/text objects."} {"ts": "2026-05-05T00:33:45+00:00", "type": "correction", "status": "accepted", "note": "Corrected Bloom terrace-slum orientation against lore: on hubward endcaps, up is inward toward the Spire, so slums are annular shelves around the docking hub office complex growing outward to the surface; removed the misleading cylinder-wall favela framing and verifier now checks hubward_endcap_rice_paddy_slums objects."} {"ts": "2026-05-05T00:55:36+00:00", "type": "correction", "status": "accepted", "note": "Bloom endcap terrace slums now use fractal value-noise occupancy rather than neat concentric box rings; verified dedicated hubward-endcap render and mesh-density checks for poor shacks, midrise housing, and outer urban crown."} +{"ts": "2026-05-05T01:12:46+00:00", "type": "maintenance", "status": "accepted", "note": "Polished persistent state after Bloom corrections: map, handoff, plan, and doctrine now prioritize source coordinate contracts, deterministic noise for organic occupancy, batched dense detail, domain-facing inspection renders, and verifier checks as reusable machine behavior rather than one-off scene scars."} diff --git a/state/map.yaml b/state/map.yaml index 6c2cb34..a82242d 100644 --- a/state/map.yaml +++ b/state/map.yaml @@ -33,10 +33,11 @@ live_architecture: pipeline: - Ingest reference images and instructions into a scene brief. - Extract target visual traits: silhouette, repetition, distribution, material response, surface breakup, camera, lighting, palette, and scale cues. + - Define the source coordinate contract: basis axes, local up/down, attachment surfaces, and frame transitions. - Plan a geometry strategy in plain language before scripting. - Emit a node graph manifest describing groups, sockets, attributes, fields, modifiers, materials, and exposed controls. - Generate Blender Python that builds the scene and node graphs from the manifest. - - Run Blender headlessly or interactively to produce `.blend` files and render evidence. + - Run Blender headlessly or interactively to produce `.blend` files, overview renders, and domain-facing inspection renders. - Review renders against the visual target and record accepted, rejected, or revised hypotheses. authoring_stack: - Python structures intent, repetition, data tables, variants, cleanup, validation, documentation, and scene/render orchestration. @@ -90,14 +91,14 @@ current_artifacts: data_flow: - Rehydrate from map, handoff, system map, implementation plan, and `npm run state:status`. - Work one bounded pipeline organ. - - Verify the seam that matters. + - Verify the seam that matters, including coordinate-frame objects, density checks, and inspection cameras when visual claims depend on them. - Add distilled evidence only when future belief changes. - Commit completed work. - If state grows faster than understanding, cut it back. current_status: - summary: VibeGeometry has an initial persistence scaffold plus verified Geometry Script translations for local math/utility samples, a first official Blender demo tranche, IRCSS French Houses city-scale graph doctrine, and a from-scratch Aetheria Bloom whole-habitat build. The IRCSS pass verified MakeSpiral and PointyGothicCone, then mapped the 108-group town kit through a dependency map, all-node atlas, and node mechanism study. The Bloom pass used Aetheria lore plus Bruneton's Rama making-of to generate a full cylindrical world study with shell anatomy, despun hub cap and docking core, spun spire sheath, shared x/angle/radius coordinate helpers, golden-angle spiral spoke and transfer-loop distribution, layered endcaps, noise-driven endcap terrace slums around the docking hub office complex that expand outward into a classier urban crown, social-gradient inner geography, elaborated spoke plazas/town fields, roads/rivers, cloud-flow patches, surface service seams, and service markers. The remaining large-system blockers are repeat-zone feedback, repeat Top outputs, foreach zones, and gizmo nodes. - next_action: Continue the Aetheria Bloom build by moving the shared cylindrical/endcap coordinate helpers plus endcap terrace-slum, town/plaza, and spiral spoke/transfer-artery networks into reusable Geometry Script groups, then improve atmospheric/light readability from render evidence. + summary: VibeGeometry has verified Geometry Script translations for local samples, official Blender demos, and IRCSS French Houses footholds, plus a from-scratch Aetheria Bloom whole-habitat build. Durable lessons now driving the pipeline: Python orchestrates intent and batching; Geometry Script emits inspectable node groups; source-bound scenes need explicit coordinate contracts before ornament; organic occupancy needs deterministic noise fields; dense detail must be batched; and visual acceptance needs domain-facing inspection renders plus verifier checks, not only overview renders. Bloom currently demonstrates shared cylindrical/endcap helpers, golden-angle transfer loops, layered shell/endcaps, noise-driven endcap terrace slums, social-gradient geography, roads/rivers/clouds, and service markers. Known graph-tooling blockers remain repeat-zone feedback, repeat Top outputs, foreach zones, and gizmo nodes. + next_action: Extract the Bloom coordinate/noise/detail lessons into a small reusable builder seam, starting with either endcap terrace slums or spiral transfer arteries, then bind that seam to scene-brief, manifest, inspection-render, and verifier contract drafts. remotes: vibegeometry: https://github.com/GameCult/VibeGeometry geometry_script_fork: https://github.com/GameCult/geometry-script