Декоративный
и вычислительный
шум

W,H = canvas size in px A_canvas_px = W * H A_allowed_px = area(usableCanvas - hardProtectedZones) A_allowed_N = A_allowed_px / A_canvas_px // dimensionless, range 0..1 rhoN = requested point count per normalized full-canvas area // unit: points / normalized-area-unit N_target = round(rhoN * A_allowed_N * detailScale) where: detailScale = STYLE_PRESET or output-context multiplier default detailScale = 1.0 EXAMPLE: W=1080, H=1350 A_allowed_px = 1,166,400 px² A_canvas_px = 1,458,000 px² A_allowed_N = 0.80 rhoN = 75 detailScale = 1.0 N_target = round(75 * 0.80) = 60

All field values are dimensionless in [0,1]. F_raw(x,y) = M_H(x,y) * M_S(x,y) * Z(x,y) * G(x,y) F_norm(x,y) = F_raw(x,y) / integralAllowed(F_raw) Expected local count in region R: E[N_R] = N_target * integral_R(F_norm) IMPORTANT: F_norm defines WHERE points should appear. rhoN / N_target defines HOW MANY points should appear. Do not use rhoN directly as an acceptance probability.

DEFAULT PIPELINE 1. Build hard masks. 2. Build soft masks. 3. Compute A_allowed_N. 4. Compute N_target from rhoN. 5. Build normalized field F_norm(x,y). 6. Initialize seeded PRNG. 7. Generate weighted candidate pool from F_norm. 8. Hard reject protected candidates. 9. Apply minimum-neighbor-distance filter. 10. Stop when N_target reached OR attempt budget exhausted. 11. Assign radius / opacity / visual role constraints. 12. Compute LOCAL metrics only. 13. Correct underfilled / overfilled zones. 14. Send proposal to global solver. 15. Freeze only after global acceptance.

Hard mask: M_H(p) ∈ {0,1} if M_H(p)=0 → candidate REJECT Soft exclusion around protected geometry: d = signedDistance(p, expandedProtectedGeometry) M_S(p) = smoothstep(0, softMarginShortN, dShortN) Suggested INITIAL_HEURISTIC margins: text / microcopy: 0.012..0.035 of short side dominant geometry: 0.010..0.030 of short side identity-critical image zones: DEFERRED to image-semantics owner

referenceWeight = minEffectiveWeight(semanticPointsNearby) noiseWeight = Point.visualWeightContextual @fresh else Point.visualWeightBase roleWeightRatio = noiseWeight / max(referenceWeight, EPS) INITIAL_HEURISTIC: preferred roleWeightRatio ≤ 0.45 soft violation 0.45..0.70 above 0.70 → reduce radius/opacity OR reclassify visualRoleCandidate This is a LOCAL role-consistency constraint. It is NOT global.hierarchy_score.

noise.local_score = w_count * noise.count_fit + w_field * noise.field_match + w_spacing * noise.spacing_distribution - w_soft * noise.soft_intrusion_penalty - w_press * noise.local_neighbor_pressure Weights = ENGINE_DEFAULT or STYLE_PRESET. ANTI-DOUBLE-COUNTING: Global solver may consume noise.local_score OR selected component metrics. It must not sum local_score + all of its components again.

OVERRIDE PRIORITY 1. protected semantic HARD constraints 2. global composition 3. global hierarchy / negative-space intent 4. local noise constraints 5. style preset 6. seeded random variation Randomness never overrides 1..5.

{ "document_id": "Compo 1.08", "version": "2.0", "module": "decorative_and_computational_point_noise", "scope": "micro-point field generation only", "finality": "LOCAL_PROPOSAL_ONLY", "consumes": [ "Compo0.Canvas", "Compo0.ProtectedZones[]", "Scene.occupiedGeometry[]", "Point.functionalRole", "Point.visualRoleCandidate", "Point.roleRigidity", "Point.visualWeightBase", "Point.visualWeightContextual?@fresh", "Compo1.06.Grid?", "Compo1.06.GridCandidate[]?", "StylePreset.noise", "Seed" ], "provides": [ "NoiseField", "NoisePoint[]", "NoiseField.targetCount", "NoiseField.actualCount", "NoiseField.allowedAreaN", "NoiseField.localScore", "NoiseField.diagnostics" ], "owns_metrics": [ "noise.count_fit", "noise.spacing_distribution", "noise.field_match", "noise.soft_intrusion_penalty", "noise.local_neighbor_pressure", "noise.local_score" ], "metric_groups": { "noise.local_score": { "components": [ "noise.count_fit", "noise.spacing_distribution", "noise.field_match", "noise.soft_intrusion_penalty", "noise.local_neighbor_pressure" ], "aggregation_rule": "consume aggregate OR components, never both" } }, "density_model": { "allowedAreaN": "A_allowed_px / (W*H)", "rhoN_unit": "points_per_normalized_canvas_area", "targetCount": "round(rhoN * allowedAreaN * detailScale)", "detailScale": {"default":1.0,"source":"STYLE_PRESET"} }, "style_density_presets_rhoN": { "quiet": [8,24], "editorial": [20,60], "technical": [45,120], "dense_field": [100,260] }, "initial_heuristics": { "radiusShortN": [0.0012,0.0045], "softMarginShortN": [0.010,0.035], "preferredNoiseToSemanticWeightRatioMax": 0.45, "softNoiseToSemanticWeightRatioMax": 0.70, "attemptBudgetMultiplier": [20,80] }, "hard_constraints": [ "no surviving noise point inside hard protected zone", "seeded repeatability for identical inputs", "finite generation attempt budget", "no local ownership of global metrics" ], "soft_constraints": [ "minimum_neighbor_distance", "target_count_fit", "target_field_match", "soft_zone_avoidance", "noise_role_weight_consistency" ], "deferred_inputs": [ "global.balance_score", "global.negative_space_score", "global.hierarchy_score", "global.eye_flow_score", "typography.protectedZones", "image_semantics.protectedZones" ], "forbidden_metric_ownership": [ "global.balance_score", "global.negative_space_score", "global.hierarchy_score", "global.eye_flow_score", "grid.fit_score", "point.visualWeightBase", "point.visualWeightContextual" ], "pipeline": [ "build_masks", "compute_allowed_area", "compute_target_count", "build_normalized_field", "generate_seeded_candidates", "hard_filter", "neighbor_distance_filter", "assign_visual_parameters", "compute_owned_metrics", "corrective_pass", "request_deferred_global_context", "apply_global_override", "freeze_after_acceptance" ], "recompute_when": [ "protected zones change", "rhoN or detailScale changes", "style noise field changes", "grid-jitter source changes", "semantic point effective weight changes materially", "global solver changes active zones" ] }