@bergetai/ui 0.5.0 → 0.6.0

package/dist/neural-flow.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/components/neural-flow/NeuralFlow.tsx","../src/components/neural-flow/mnist-samples.ts","../src/components/neural-flow/sparse-weights.ts"],"sourcesContent":["/* cspell:disable */\nimport { Canvas, useFrame, useThree } from '@react-three/fiber';\nimport { useEffect, useMemo, useRef, type CSSProperties, type FC, type RefObject } from 'react';\nimport * as THREE from 'three';\nimport { MNIST_SAMPLES } from './mnist-samples';\nimport { SPARSE_WEIGHTS } from './sparse-weights';\n\n/**\n * NeuralFlow — AI inference, rendered live.\n *\n * An MLP in the shape of a classic MNIST classifier (784-256-128-64-10)\n * floats in the Berget green, layer plane after plane receding into the\n * depth. Tokens — signal packets — travel edge to edge through the net;\n * every hop a weight attenuates them (they fade) or amplifies them\n * (they swell bright), until a sample rolls the whole way through and\n * the predicted output node blooms white-green. A CPU sim drives the\n * state (1.2k nodes, ~1.8k live tokens is trivial arithmetic per\n * frame); three.js only paints points, lines and a fog plane — which\n * keeps this Safari-safe with zero float-FBO involvement.\n *\n * Props: every visual knob is a plain number prop read each frame —\n * exposure, focus/blurK depth-of-field, sim speed (a true timeline:\n * 0 pauses pixel-for-pixel, 0.2 is dream's pace), token intensity and\n * camera drift. Reduced motion paints one deterministic still.\n */\n\n// ---------------------------------------------------------------- layout ---\n\nconst LAYER_SIZES = [784, 256, 128, 64, 10] as const;\nconst LAYER_GRID: [rows: number, cols: number][] = [\n  [28, 28],\n  [16, 16],\n  [16, 8],\n  [8, 8],\n  [10, 1],\n];\nconst LAYER_X = 1.35; // camera-to-plane spacing along the stack axis\nconst CELL = 0.05; // world units between neighbours in a plane\nconst TOKEN_POOL = 1800;\nconst SAMPLE_INTERVAL = 6.5; // seconds between input samples\nconst SAMPLE_WINDOW = 2.6; // sample emission window — long enough to paint the numeral\nconst SAMPLE_RATE = 420; // tokens/second during a sample (x intensity)\nconst TRICKLE_RATE = 40; // idle tokens/second\nconst CLASSES = 10;\n\nexport type NeuralFlowOrientation = 'horizontal' | 'vertical';\n\n/** Classic 5x7 numerals — the actual \"images\" the net sees per class. */\nconst DIGITS: string[][] = [\n  ['.###.', '#...#', '#..##', '#.#.#', '##..#', '#...#', '.###.'],\n  ['..#..', '.##..', '..#..', '..#..', '..#..', '..#..', '#####'],\n  ['.###.', '#...#', '....#', '...#.', '..#..', '.#...', '#####'],\n  ['#####', '...#.', '..#..', '...#.', '....#', '#...#', '.###.'],\n  ['...#.', '..##.', '.#.#.', '#..#.', '#####', '...#.', '...#.'],\n  ['#####', '#....', '####.', '....#', '....#', '#...#', '.###.'],\n  ['..##.', '.#...', '#....', '####.', '#...#', '#...#', '.###.'],\n  ['#####', '....#', '...#.', '..#..', '.#...', '.#...', '.#...'],\n  ['.###.', '#...#', '#...#', '.###.', '#...#', '#...#', '.###.'],\n  ['.###.', '#...#', '#...#', '.####', '....#', '...#.', '.##..'],\n];\n\nconst PALETTE = {\n  deep: [0.027, 0.235, 0.2] as const,\n  mid: [0.051, 0.42, 0.322] as const,\n  crest: [0.137, 0.627, 0.36] as const,\n  slate: [0.12, 0.17, 0.19] as const,\n  white: [0.85, 1.0, 0.9] as const,\n};\n\n/** The dark well behind the fog plane, in brand tokens — the gradient\n    the raw canvas painted under itself. Requires the package's\n    styles (dist/styles/index.css), like every themed component here. */\nconst BACKDROP =\n  'radial-gradient(120% 95% at 28% 58%, var(--berget-brand-moss-900) 0%, var(--berget-brand-moss-950) 55%, var(--berget-brand-granite-950) 100%)';\n\n// ---------------------------------------------------------------- rng ------\n\nfunction mulberry32(seed: number) {\n  let a = seed >>> 0;\n  return () => {\n    a |= 0;\n    a = (a + 0x6d2b79f5) | 0;\n    let t = Math.imul(a ^ (a >>> 15), 1 | a);\n    t = (t + Math.imul(t ^ (t >>> 7), 61 | t)) ^ t;\n    return ((t ^ (t >>> 14)) >>> 0) / 4294967296;\n  };\n}\n\n// ------------------------------------------------------------- network -----\n\ntype Network = {\n  /** Flat node positions, xyz per node. */\n  positions: Float32Array;\n  nodeCount: number;\n  layerStart: number[]; // first flat index of each layer\n  /** For each node: outgoing targets [dstFlatIndex, weight] pairs. */\n  out: { dst: number[]; w: number[] }[];\n  /** For each node: incoming edges (needed for the forward pass). */\n  inEdges: { src: number; w: number }[][];\n  /** Static line segments between connected nodes, xyz per vertex. */\n  edgePositions: Float32Array;\n  /** Per-vertex edge weight, mirrored into the line shader. */\n  edgeWeights: Float32Array;\n  edgeCount: number;\n};\n\nexport type InputBlob = { i: number; w: number }[];\n/** One entry per class; each holds that class's candidate input cells. */\nexport type ClassBlobs = InputBlob[];\n\nfunction buildNetwork(seed: number, orientation: NeuralFlowOrientation): Network {\n  const rng = mulberry32(seed);\n  const nodeCount = LAYER_SIZES.reduce((a, b) => a + b, 0);\n  const positions = new Float32Array(nodeCount * 3);\n  const layerStart: number[] = [];\n  let flat = 0;\n\n  LAYER_SIZES.forEach((size, li) => {\n    layerStart.push(flat);\n    const [rows, cols] = LAYER_GRID[li]!;\n    for (let i = 0; i < size; i++) {\n      const r = Math.floor(i / cols);\n      const c = i % cols;\n      const o = flat * 3;\n      if (orientation === 'vertical') {\n        // Upright stack: layers rise in +Y, each plane a floor slab in X-Z.\n        positions[o] = (c - (cols - 1) / 2) * CELL;\n        positions[o + 1] = li * LAYER_X;\n        positions[o + 2] = (r - (rows - 1) / 2) * CELL;\n      } else {\n        positions[o] = li * LAYER_X;\n        positions[o + 1] = (r - (rows - 1) / 2) * CELL;\n        positions[o + 2] = (c - (cols - 1) / 2) * CELL;\n      }\n      flat++;\n    }\n  });\n\n  // Sparse random inter-layer connectivity — fan-in per node, mostly\n  // excitatory weights with a minority of inhibitory ones; weakly\n  // 1/sqrt(fanIn) normalised like a real init.\n  const fanIn = [0, 9, 10, 10, 7]; // sources per node, per layer\n  const inEdges: { src: number; w: number }[][] = [];\n  for (let i = 0; i < nodeCount; i++) inEdges.push([]);\n  for (let li = 1; li < LAYER_SIZES.length; li++) {\n    const srcBase = layerStart[li - 1]!;\n    const dstBase = layerStart[li]!;\n    for (let d = 0; d < LAYER_SIZES[li]!; d++) {\n      const k = fanIn[li]!;\n      for (let e = 0; e < k; e++) {\n        const src = srcBase + Math.floor(rng() * LAYER_SIZES[li - 1]!);\n        const excitatory = rng() < 0.85;\n        // Excitatory weights amplify (0.75-1.65) so a signal can really\n        // SURVIVE end-to-end; inhibitory ones bleed a token hard.\n        const w = excitatory ? 0.75 + 0.9 * rng() : -(0.35 + 0.35 * rng());\n        inEdges[dstBase + d]!.push({ src, w });\n      }\n    }\n  }\n\n  // Transpose to outgoing lists + build static line geometry.\n  const out: { dst: number[]; w: number[] }[] = [];\n  for (let i = 0; i < nodeCount; i++) out.push({ dst: [], w: [] });\n  let edgeCount = 0;\n  for (let d = 0; d < nodeCount; d++) edgeCount += inEdges[d]!.length;\n  const edgePositions = new Float32Array(edgeCount * 2 * 3);\n  const edgeWeights = new Float32Array(edgeCount * 2);\n  let eo = 0;\n  for (let d = 0; d < nodeCount; d++) {\n    for (const e of inEdges[d]!) {\n      out[e.src]!.dst.push(d);\n      out[e.src]!.w.push(e.w);\n      edgePositions[eo * 6] = positions[e.src * 3]!;\n      edgePositions[eo * 6 + 1] = positions[e.src * 3 + 1]!;\n      edgePositions[eo * 6 + 2] = positions[e.src * 3 + 2]!;\n      edgePositions[eo * 6 + 3] = positions[d * 3]!;\n      edgePositions[eo * 6 + 4] = positions[d * 3 + 1]!;\n      edgePositions[eo * 6 + 5] = positions[d * 3 + 2]!;\n      const a = Math.min(1, Math.abs(e.w) * 1.6);\n      edgeWeights[eo * 2] = a;\n      edgeWeights[eo * 2 + 1] = a;\n      eo++;\n    }\n  }\n\n  return {\n    positions,\n    nodeCount,\n    layerStart,\n    out,\n    inEdges,\n    edgePositions,\n    edgeWeights,\n    edgeCount,\n  };\n}\n\n// ------------------------------------------------------------------ sim ----\n\ntype Token = {\n  src: number;\n  dst: number;\n  w: number;\n  layer: number; // source layer index\n  t: number; // 0..1 along the edge\n  rate: number; // t units per second\n  gain: number;\n  pos: [number, number, number];\n};\n\nexport class FlowSim {\n  private net: Network;\n  private blobs: ClassBlobs[];\n  private images: Float32Array[][];\n  private trained: { w: Float32Array[]; b: Float32Array[] };\n  /** Per-node energy of the CURRENT sample's real forward pass; decays\n      after the sample window. */\n  private energy: Float32Array;\n  private predictedClass = 0;\n  private act: Float32Array;\n  private rng = mulberry32(0xc0ffee);\n  private tokens: (Token | null)[] = [];\n  private time = 0;\n  private nextSample = 0.8;\n  private sampleUntil = -10; // far past\n  private sampleClass = 0;\n  private sampleVariant = 0;\n  private emitAcc = 0;\n  readonly nodeAttr: Float32Array;\n  readonly tokenPos: Float32Array;\n  readonly tokenGain: Float32Array;\n\n  constructor(\n    seed = 1337,\n    orientation: NeuralFlowOrientation = 'horizontal',\n    blobs: ClassBlobs[] = glyphBlobs(),\n    images: Float32Array[][] = [],\n    trained: { w: Float32Array[]; b: Float32Array[] } = decodeWeights(),\n  ) {\n    this.net = buildNetwork(seed, orientation);\n    attachWeights(this.net, trained);\n    this.blobs = blobs;\n    this.images = images;\n    this.trained = trained;\n    this.energy = new Float32Array(this.net.nodeCount);\n    this.act = new Float32Array(this.net.nodeCount);\n    this.nodeAttr = new Float32Array(this.net.nodeCount);\n    this.tokenPos = new Float32Array(TOKEN_POOL * 3);\n    this.tokenGain = new Float32Array(TOKEN_POOL);\n    for (let i = 0; i < TOKEN_POOL; i++) this.tokens.push(null);\n  }\n\n  get network() {\n    return this.net;\n  }\n\n  /** Advance the whole inference world by dt seconds (clamped internally). */\n  step(dt: number, intensity: number) {\n    dt = Math.min(dt, 0.12);\n    this.time += dt;\n\n    // Activation half-lives: mid-net ~0.8 s (0.42^dt); input ~1.4 s\n    // (0.6^dt) so the numeral stays readable while it feeds; output\n    // ~0.6 s (0.3^dt) so the column flickers with live traffic while\n    // the continuously re-fed prediction still dominates.\n    const decay = Math.pow(0.42, dt);\n    const inDecay = Math.pow(0.6, dt);\n    const outDecay = Math.pow(0.3, dt);\n    const inEnd = this.net.layerStart[1]!;\n    const outStart = this.net.layerStart[LAYER_SIZES.length - 1]!;\n    for (let i = 0; i < this.act.length; i++) {\n      this.act[i] = this.act[i]! * (i >= outStart ? outDecay : i < inEnd ? inDecay : decay);\n    }\n\n    // Sample scheduling: enter the emission window, pick the class.\n    if (this.time >= this.nextSample) {\n      this.sampleClass = Math.floor(this.rng() * CLASSES);\n      const variants = this.blobs[this.sampleClass]!;\n      this.sampleVariant = Math.floor(this.rng() * variants.length);\n      this.sampleUntil = this.time + SAMPLE_WINDOW;\n      this.nextSample = this.time + SAMPLE_INTERVAL;\n      // The REAL forward pass: energy per node for this exact image,\n      // and which class the net actually predicts (honest misses and all).\n      const { predicted, energy } = forwardPass(\n        this.net,\n        this.trained,\n        this.imageOf(this.sampleClass, this.sampleVariant),\n      );\n      this.predictedClass = predicted;\n      this.energy = energy;\n    }\n\n    // The last sample's energy fades once its window ends.\n    const fade = Math.pow(0.55, dt);\n    if (this.time > this.sampleUntil) {\n      for (let i = 0; i < this.energy.length; i++) this.energy[i] = this.energy[i]! * fade;\n    }\n\n    // Token injection — sample burst from its class blob, plus an\n    // ever-present idle trickle so the net never looks dead.\n    const inSample = this.time < this.sampleUntil;\n    const rate = (inSample ? SAMPLE_RATE : TRICKLE_RATE) * intensity;\n\n    this.emitAcc += rate * dt;\n    while (this.emitAcc >= 1) {\n      this.emitAcc -= 1;\n      this.inject(inSample);\n    }\n\n    // Advance tokens, fire arrivals, hand on the baton or bury it.\n    for (let i = 0; i < TOKEN_POOL; i++) {\n      const tok = this.tokens[i];\n      if (!tok) continue;\n      tok.t += tok.rate * dt;\n      const p = this.net.positions;\n      const s3 = tok.src * 3;\n      const d3 = tok.dst * 3;\n      const t = Math.min(tok.t, 1);\n      tok.pos = [\n        p[s3]! + (p[d3]! - p[s3]!) * t,\n        p[s3 + 1]! + (p[d3 + 1]! - p[s3 + 1]!) * t,\n        p[s3 + 2]! + (p[d3 + 2]! - p[s3 + 2]!) * t,\n      ];\n      if (tok.t < 1) continue;\n\n      // Arrival: EVERY deposit lights its node, everywhere in the net.\n      // The output layer keeps a soft one-hot read — the net's OWN\n      // prediction blooms high above the rest, which shimmer with the\n      // live traffic like any other layer.\n      const isFinal = tok.layer === LAYER_SIZES.length - 2;\n      let pulse = tok.gain * 0.55;\n      if (isFinal) {\n        const c = tok.dst - this.net.layerStart[LAYER_SIZES.length - 1]!;\n        if (c === this.predictedClass) pulse *= 3.8;\n      }\n      this.act[tok.dst] = Math.min(2, Math.max(-2, this.act[tok.dst]! + pulse));\n\n      // Hop gating: during a sample the forward energy decides how much\n      // signal THIS node holds — hot routes carry on, dead routes fade.\n      // Between samples the idle trickle wanders freely, so the net\n      // never looks dead.\n      const energyHere = this.energy[tok.dst]!;\n      if (inSample && !isFinal && energyHere < 0.1 && this.rng() < 0.7) {\n        this.tokens[i] = null; // route is cold for this image\n        continue;\n      }\n      const gain = Math.min(0.25 + 1.6 * energyHere, 2.4);\n      if (isFinal) {\n        this.tokens[i] = null; // dissolved in the output\n        continue;\n      }\n      // Hand the signal on through a random outgoing edge.\n      const outs = this.net.out[tok.dst]!;\n      if (outs.dst.length === 0) {\n        this.tokens[i] = null;\n        continue;\n      }\n      const pick = Math.floor(this.rng() * outs.dst.length);\n      tok.src = tok.dst;\n      tok.dst = outs.dst[pick]!;\n      tok.w = outs.w[pick]!;\n      tok.layer += 1;\n      tok.t = 0;\n      const len = edgeLen(this.net.positions, tok.src, tok.dst);\n      tok.rate = (1.35 / Math.max(len, 0.05)) * (0.75 + this.rng() * 0.5);\n      tok.gain = gain;\n    }\n\n    // Push node activations into the point shader attribute.\n    for (let i = 0; i < this.act.length; i++) this.nodeAttr[i] = this.act[i]!;\n  }\n\n  /** The 784-vector for a class variant: real image if we have one,\n      otherwise intensities derived from the blob shape. */\n  private imageOf(cls: number, variant: number): Float32Array {\n    const list = this.images[cls];\n    if (list && list.length > 0) return list[Math.min(variant, list.length - 1)]!;\n    const img = new Float32Array(784);\n    const blob = this.blobs[cls]![Math.min(variant, this.blobs[cls]!.length - 1)]!;\n    for (const c of blob) img[c.i] = Math.min(1, c.w + 0.25);\n    return img;\n  }\n\n  private inject(fromBlob: boolean) {\n    const slot = this.tokens.indexOf(null);\n    if (slot < 0) return;\n    const rng = this.rng;\n    let src: number;\n    if (fromBlob) {\n      const variants = this.blobs[this.sampleClass]!;\n      const blob = variants[Math.min(this.sampleVariant, variants.length - 1)]!;\n      src = pickWeightedBlob(blob, rng());\n    } else {\n      src = Math.floor(rng() * LAYER_SIZES[0]);\n    }\n    const outs = this.net.out[src]!;\n    // Sparse fan-out: some nodes happen to have no outgoing edge\n    // (every input node is not reachable) — those tokens never spawn.\n    if (outs.dst.length === 0) return;\n    const pick = Math.floor(rng() * outs.dst.length);\n    const len = edgeLen(this.net.positions, src, outs.dst[pick]!);\n    this.tokens[slot] = {\n      src,\n      dst: outs.dst[pick]!,\n      w: outs.w[pick]!,\n      layer: 0,\n      t: 0,\n      rate: (1.35 / Math.max(len, 0.05)) * (0.75 + rng() * 0.5),\n      gain: fromBlob ? 1.0 : 0.6,\n      pos: [0, 0, 0],\n    };\n    // The emitting input node flashes — a sample draws its \"digit\" on\n    // the 28x28 plane, idle trickle pins it barely.\n    this.act[src] = Math.min(2, this.act[src]! + (fromBlob ? 0.9 : 0.12));\n  }\n\n  /** Write live token rows into the point buffers (dead rows stay 0). */\n  writeTokenBuffers() {\n    for (let i = 0; i < TOKEN_POOL; i++) {\n      const tok = this.tokens[i];\n      const o = i * 3;\n      if (!tok) {\n        this.tokenGain[i] = 0;\n        this.tokenPos[o] = 0;\n        this.tokenPos[o + 1] = 0;\n        this.tokenPos[o + 2] = 0;\n        continue;\n      }\n      this.tokenPos[o] = tok.pos[0];\n      this.tokenPos[o + 1] = tok.pos[1];\n      this.tokenPos[o + 2] = tok.pos[2];\n      this.tokenGain[i] = tok.gain;\n    }\n  }\n}\n\n/** Procedural 5x7-glyph digits — fallback when no samples are given. */\nfunction glyphBlobs(): ClassBlobs[] {\n  const rng = mulberry32(0xd1617);\n  const all: ClassBlobs[] = [];\n  for (let cls = 0; cls < CLASSES; cls++) {\n    const field = new Float32Array(28 * 28);\n    const glyph = DIGITS[cls]!;\n    const dx = 4 + Math.floor(rng() * 3);\n    const dy = Math.floor(rng() * 3) - 1;\n    for (let gr = 0; gr < 7; gr++) {\n      const row = glyph[gr]!;\n      for (let gc = 0; gc < 5; gc++) {\n        if (row[gc] !== '#') continue;\n        for (let k = 0; k < 4; k++) {\n          for (let l = 0; l < 4; l++) {\n            const r = gr * 4 + l + dy;\n            const c = gc * 4 + k + dx;\n            if (r < 28 && c < 28) field[r * 28 + c] = 1;\n          }\n        }\n      }\n    }\n    const blob: InputBlob = [];\n    for (let i = 0; i < 784; i++) {\n      if (field[i]! > 0) blob.push({ i, w: 1 });\n    }\n    for (let k = 0; k < 8; k++) blob.push({ i: Math.floor(rng() * 784), w: 0.15 });\n    all.push([blob]);\n  }\n  return all;\n}\n\n/** Decode base64url into raw bytes — shared by weights and samples. */\nfunction decodeB64(b64: string): Uint8Array {\n  const bin = atob(b64.replace(/-/g, '+').replace(/_/g, '/'));\n  const buf = new Uint8Array(bin.length);\n  for (let i = 0; i < bin.length; i++) buf[i] = bin.charCodeAt(i);\n  return buf;\n}\n\n/** Decode packed FP32 little-endian base64url into a Float32Array. */\nfunction decodeF32(b64: string): Float32Array {\n  return new Float32Array(decodeB64(b64).buffer);\n}\n\n/** The trained sparse weight set (see train-sparse.py). */\nexport function decodeWeights(): { w: Float32Array[]; b: Float32Array[] } {\n  return {\n    w: [\n      decodeF32(SPARSE_WEIGHTS.weights1),\n      decodeF32(SPARSE_WEIGHTS.weights2),\n      decodeF32(SPARSE_WEIGHTS.weights3),\n      decodeF32(SPARSE_WEIGHTS.weights4),\n    ],\n    b: [\n      decodeF32(SPARSE_WEIGHTS.biases1),\n      decodeF32(SPARSE_WEIGHTS.biases2),\n      decodeF32(SPARSE_WEIGHTS.biases3),\n      decodeF32(SPARSE_WEIGHTS.biases4),\n    ],\n  };\n}\n\n/**\n * Splice trained weights into the visible net: every rendered edge\n * gets its trained value, the edge-alpha map follows |w|, and `out`\n * is rebuilt with the trained magnitudes for the hop visuals.\n */\nfunction attachWeights(net: Network, trained: { w: Float32Array[]; b: Float32Array[] }) {\n  const layerStart = net.layerStart;\n  for (let li = 1; li < 5; li++) {\n    const base = layerStart[li]!;\n    const w = trained.w[li - 1]!;\n    let eo = 0;\n    const nextBase = li < 4 ? layerStart[li + 1]! : net.nodeCount;\n    for (let dlocal = 0; dlocal < nextBase - base; dlocal++) {\n      const dst = base + dlocal;\n      const list = net.inEdges[dst]!;\n      for (const e of list) {\n        e.w = w[eo]!;\n        eo++;\n      }\n    }\n    if (eo !== w.length) throw new Error(`layer ${li}: expected ${w.length} edges, got ${eo}`);\n  }\n  // rebuild out + edge alpha from trained weights\n  for (const o of net.out) {\n    o.dst.length = 0;\n    o.w.length = 0;\n  }\n  let eo = 0;\n  for (let d = 0; d < net.nodeCount; d++) {\n    for (const e of net.inEdges[d]!) {\n      net.out[e.src]!.dst.push(d);\n      net.out[e.src]!.w.push(e.w);\n      net.edgeWeights[eo * 2] = Math.min(1, Math.abs(e.w) * 1.7);\n      net.edgeWeights[eo * 2 + 1] = Math.min(1, Math.abs(e.w) * 1.7);\n      eo++;\n    }\n  }\n}\n\n/** Real forward pass through the visible, trained edges. */\nexport function forwardPass(\n  net: Network,\n  trained: { w: Float32Array[]; b: Float32Array[] },\n  img: Float32Array,\n): { predicted: number; energy: Float32Array } {\n  const act = new Float32Array(net.nodeCount);\n  act.set(img);\n  const layerStart = net.layerStart;\n  // NOTE: layerStart has one entry per layer (no sentinel) — the size\n  // must come from LAYER_SIZES, or the output layer gets size NaN and\n  // every prediction collapses to a constant class.\n  for (let li = 1; li < 5; li++) {\n    const base = layerStart[li]!;\n    const size = LAYER_SIZES[li]!;\n    const bias = trained.b[li - 1]!;\n    for (let dl = 0; dl < size; dl++) {\n      let z = bias[dl]!;\n      for (const e of net.inEdges[base + dl]!) z += act[e.src]! * e.w;\n      act[base + dl] = li < 4 ? Math.max(z, 0) : z;\n    }\n  }\n  // per-layer normalisation for the gating map\n  const energy = new Float32Array(act.length);\n  for (let li = 0; li < 5; li++) {\n    const base = layerStart[li]!;\n    const size = LAYER_SIZES[li]!;\n    let max = 1e-4;\n    for (let i = base; i < base + size; i++) max = Math.max(max, act[i]!);\n    for (let i = base; i < base + size; i++) energy[i] = Math.max(act[i]!, 0) / max;\n  }\n  const outBase = layerStart[4]!;\n  let predicted = 0;\n  let top = -1e9;\n  for (let c = 0; c < CLASSES; c++) {\n    if (act[outBase + c]! > top) {\n      top = act[outBase + c]!;\n      predicted = c;\n    }\n  }\n  return { predicted, energy };\n}\n\nexport function decodeAll(samples: Record<number, readonly string[]>): {\n  blobs: InputBlob[][];\n  images: Float32Array[][];\n} {\n  const blobs: InputBlob[][] = [];\n  const images: Float32Array[][] = [];\n  for (let c = 0; c < CLASSES; c++) {\n    const cellList: InputBlob[] = [];\n    const imgList: Float32Array[] = [];\n    for (const b64 of samples[c] ?? []) {\n      // One binary decode per sample drives both representations:\n      // the weighted emission blob and the 784-vector the forward\n      // pass consumes.\n      const buf = decodeB64(b64);\n      const cells: InputBlob = [];\n      const img = new Float32Array(784);\n      for (let n = 0; n < 784; n++) {\n        const byte = buf[n >> 1]!;\n        const v = n & 1 ? byte & 15 : byte >> 4;\n        if (v >= 5) cells.push({ i: n, w: (v - 4) / 11 });\n        img[n] = v / 15;\n      }\n      cellList.push(cells);\n      imgList.push(img);\n    }\n    blobs.push(cellList);\n    images.push(imgList);\n  }\n  return { blobs, images };\n}\n\nfunction pickWeightedBlob(blob: InputBlob, r: number): number {\n  let total = 0;\n  for (const c of blob) total += c.w;\n  let acc = r * total;\n  for (const c of blob) {\n    acc -= c.w;\n    if (acc <= 0) return c.i;\n  }\n  return blob[blob.length - 1]!.i;\n}\n\nfunction edgeLen(p: Float32Array, a: number, b: number): number {\n  const a3 = a * 3;\n  const b3 = b * 3;\n  const dx = p[b3]! - p[a3]!;\n  const dy = p[b3 + 1]! - p[a3 + 1]!;\n  const dz = p[b3 + 2]! - p[a3 + 2]!;\n  return Math.sqrt(dx * dx + dy * dy + dz * dz);\n}\n\n// -------------------------------------------------------------- shader -----\n\nconst POINT_VERT = /* glsl */ `\nattribute float aAct; // node activation, or token gain for tokens\nuniform float uHeightPx, uFovTan, uDpr, uFocusDist, uBlurK;\nuniform float uBaseSize;\nvarying float vAct;\nvarying float vFade;\nvoid main() {\n  vec4 mv = modelViewMatrix * vec4(position, 1.0);\n  float dist = max(-mv.z, 0.001);\n  vAct = aAct;\n  float a = max(aAct, 0.0);\n  float worldSize = uBaseSize * (1.0 + clamp(a, 0.0, 2.0) * 0.9);\n  // uHeightPx is the drawing-buffer height (already device px), so the\n  // world-size term needs no uDpr; the pixel-space defocus offset does.\n  float px = worldSize * (uHeightPx * 0.5) / (uFovTan * dist);\n  float dz = dist - uFocusDist;\n  px += (dz < 0.0 ? -dz * 0.55 : dz * 0.8) * uBlurK * uDpr;\n  gl_PointSize = min(px, 90.0 * uDpr);\n  vFade = clamp(1.0 - abs(dz) * 0.16, 0.25, 1.0);\n  gl_Position = projectionMatrix * mv;\n}\n`;\n\nconst NODE_FRAG = /* glsl */ `\nuniform float uExposure;\nvarying float vAct;\nvarying float vFade;\nconst vec3 DEEP = vec3(${PALETTE.deep.join(',')});\nconst vec3 MID = vec3(${PALETTE.mid.join(',')});\nconst vec3 CREST = vec3(${PALETTE.crest.join(',')});\nconst vec3 SLATE = vec3(${PALETTE.slate.join(',')});\nconst vec3 WHITE = vec3(${PALETTE.white.join(',')});\nvoid main() {\n  float d = length(gl_PointCoord - 0.5) * 2.0;\n  float disc = smoothstep(1.0, 0.35, d);\n  if (disc < 0.01) discard;\n  float a = vAct;\n  vec3 col;\n  float glow;\n  if (a >= 0.0) {\n    col = mix(MID, CREST, clamp(a, 0.0, 1.0));\n    // Overdriven nodes bloom toward the brand's crystal white-green.\n    float bloom = smoothstep(1.05, 1.9, a);\n    col = mix(col, WHITE, bloom);\n    glow = 0.36 + 0.64 * clamp(a, 0.0, 1.0) + bloom * 0.9;\n  } else {\n    col = mix(DEEP, SLATE, clamp(-a, 0.0, 1.0));\n    glow = 0.16 + 0.25 * clamp(-a, 0.0, 1.0);\n  }\n  gl_FragColor = vec4(col * uExposure, disc * 0.62 * glow * vFade);\n}\n`;\n\nconst TOKEN_FRAG = /* glsl */ `\nuniform float uExposure;\nvarying float vAct;\nvarying float vFade;\nconst vec3 CREST = vec3(${PALETTE.crest.join(',')});\nconst vec3 WHITE = vec3(${PALETTE.white.join(',')});\nvoid main() {\n  float g = vAct;\n  if (g < 0.004) discard;\n  float d = length(gl_PointCoord - 0.5) * 2.0;\n  float core = smoothstep(0.55, 0.05, d);\n  float halo = smoothstep(1.0, 0.3, d) * 0.45;\n  if (core + halo < 0.01) discard;\n  // Amplified tokens run hot toward white-green.\n  vec3 col = mix(CREST, WHITE, smoothstep(1.1, 2.1, g));\n  float b = (0.35 + 0.65 * clamp(g, 0.0, 1.6)) * (core + halo);\n  gl_FragColor = vec4(col * uExposure, b * 0.78 * vFade);\n}\n`;\n\nconst EDGE_VERT = /* glsl */ `\nattribute float aEdgeW;\nvarying float vW;\nvoid main() {\n  vW = aEdgeW;\n  gl_Position = projectionMatrix * modelViewMatrix * vec4(position, 1.0);\n}\n`;\n\nconst EDGE_FRAG = /* glsl */ `\nvarying float vW;\nconst vec3 DEEP = vec3(${PALETTE.crest.join(',')});\nvoid main() {\n  gl_FragColor = vec4(DEEP, 0.02 + 0.05 * vW);\n}\n`;\n\nconst FOG_VERT = /* glsl */ `\nvarying vec2 vUv;\nvoid main() {\n  vUv = uv;\n  gl_Position = projectionMatrix * modelViewMatrix * vec4(position, 1.0);\n}\n`;\n\nconst FOG_FRAG = /* glsl */ `\nuniform float uTime;\nvarying vec2 vUv;\n\nfloat hash(vec2 p) {\n  return fract(sin(dot(p, vec2(127.1, 311.7))) * 43758.5453123);\n}\nfloat noise(vec2 p) {\n  vec2 i = floor(p);\n  vec2 f = fract(p);\n  f = f * f * (3.0 - 2.0 * f);\n  return mix(\n    mix(hash(i), hash(i + vec2(1, 0)), f.x),\n    mix(hash(i + vec2(0, 1)), hash(i + vec2(1, 1)), f.x),\n    f.y\n  );\n}\n\nvoid main() {\n  vec2 c = vUv - 0.5;\n  float r = length(c) * 2.0;\n  // Soft brand-green well behind the stack, vignetted to darkness.\n  vec3 col = vec3(0.012, 0.05, 0.04) * max(1.0 - r, 0.0);\n  col += vec3(0.01, 0.06, 0.045) * max(1.5 - r * 1.25, 0.0);\n  // Slow drifting wisps crossing in front of the black.\n  float w = noise(vec2(vUv.x * 3.0 - uTime * 0.02, vUv.y * 6.0 + uTime * 0.013));\n  w = pow(w, 3.0) * 0.5;\n  col += vec3(0.05, 0.11, 0.09) * w * (1.0 - smoothstep(0.6, 1.2, r));\n  gl_FragColor = vec4(col, 1.0);\n}\n`;\n\n// ------------------------------------------------------------ component ----\n\nexport type NeuralFlowKnobs = {\n  /** Lens zoom divisor of the base fov. @default 1 */\n  zoom: number;\n  /** Exposure multiplier. @default 1.45 */\n  exposure: number;\n  /** Focal plane view-distance. @default 8.0 */\n  focus: number;\n  /** Depth-of-field strength in px. @default 4 */\n  blurK: number;\n  /** Timeline scale — 0 pauses time, 0.1-0.5 slow motion. @default 1 */\n  speed: number;\n  /** Token emission rate multiplier. @default 1 */\n  intensity: number;\n  /** Camera drift amplitude multiplier. @default 1 */\n  drift: number;\n  /** Turntable spin around the stack's centre, rad/s. 0 parks it.\n      @default 0.02 — a full turn in ~5 minutes */\n  spin: number;\n};\n\nconst DEFAULT_KNOBS: NeuralFlowKnobs = {\n  zoom: 1,\n  exposure: 1.45,\n  focus: 8.0,\n  blurK: 4,\n  speed: 1,\n  intensity: 1,\n  drift: 1,\n  spin: 0.02,\n};\n\nconst BASE_FOV = 36;\n// Horizontal: side-on vantage, the stack reads left (input 28x28) to\n// right (softmax column). Vertical: the stack rises bottom to top.\nfunction camFor(orientation: NeuralFlowOrientation) {\n  return orientation === 'vertical'\n    ? {\n        base: new THREE.Vector3(0.7, 3.5, 9.6),\n        target: new THREE.Vector3(0, 2.45, 0),\n        center: new THREE.Vector3(0, 2.7, 0),\n        fogSize: [13, 15] as const,\n        fogPos: new THREE.Vector3(1.5, 2.7, -6.5),\n        fogRot: 0,\n      }\n    : {\n        base: new THREE.Vector3(-0.9, 0.88, 6.1),\n        target: new THREE.Vector3(2.35, -0.08, -0.2),\n        center: new THREE.Vector3(2.7, 0, 0),\n        fogSize: [14, 8] as const,\n        fogPos: new THREE.Vector3(7.2, 0, 0),\n        fogRot: -Math.PI / 2,\n      };\n}\n\nfunction SceneContent({\n  knobs: knobsRef,\n  orientation,\n  blobs,\n  images,\n}: {\n  knobs: RefObject<NeuralFlowKnobs>;\n  orientation: NeuralFlowOrientation;\n  blobs: ClassBlobs[];\n  images: Float32Array[][];\n}) {\n  const { invalidate } = useThree();\n  const trained = useMemo(() => decodeWeights(), []);\n  const sim = useMemo(\n    () => new FlowSim(1337, orientation, blobs, images, trained),\n    [orientation, blobs, images, trained],\n  );\n  const camCfg = useMemo(() => camFor(orientation), [orientation]);\n\n  const fogPlane = useMemo(\n    () => new THREE.PlaneGeometry(camCfg.fogSize[0], camCfg.fogSize[1]),\n    [camCfg],\n  );\n\n  const geos = useMemo(() => {\n    const net = sim.network;\n    const nodes = new THREE.BufferGeometry();\n    nodes.setAttribute('position', new THREE.BufferAttribute(net.positions, 3));\n    const actAttr = new THREE.BufferAttribute(sim.nodeAttr, 1);\n    actAttr.setUsage(THREE.DynamicDrawUsage);\n    nodes.setAttribute('aAct', actAttr);\n\n    const tokens = new THREE.BufferGeometry();\n    const posAttr = new THREE.BufferAttribute(sim.tokenPos, 3);\n    posAttr.setUsage(THREE.DynamicDrawUsage);\n    tokens.setAttribute('position', posAttr);\n    const gainAttr = new THREE.BufferAttribute(sim.tokenGain, 1);\n    gainAttr.setUsage(THREE.DynamicDrawUsage);\n    tokens.setAttribute('aAct', gainAttr);\n    // Frustum culling must never hide the moving token cloud.\n    const b = 6;\n    nodes.boundingSphere = new THREE.Sphere(camCfg.center.clone(), b);\n    tokens.boundingSphere = new THREE.Sphere(camCfg.center.clone(), b);\n\n    const edges = new THREE.BufferGeometry();\n    edges.setAttribute('position', new THREE.BufferAttribute(net.edgePositions, 3));\n    edges.setAttribute('aEdgeW', new THREE.BufferAttribute(net.edgeWeights, 1));\n    edges.boundingSphere = new THREE.Sphere(camCfg.center.clone(), b);\n    return [nodes, tokens, edges];\n  }, [sim, camCfg]);\n\n  const nodeMat = useMemo(() => makePointMaterial(), []);\n  const tokenMat = useMemo(() => makeTokenMaterial(), []);\n  const edgeMat = useMemo(\n    () =>\n      new THREE.ShaderMaterial({\n        vertexShader: EDGE_VERT,\n        fragmentShader: EDGE_FRAG,\n        transparent: true,\n        depthWrite: false,\n        blending: THREE.AdditiveBlending,\n        uniforms: {},\n      }),\n    [],\n  );\n  const fogMat = useMemo(\n    () =>\n      new THREE.ShaderMaterial({\n        vertexShader: FOG_VERT,\n        fragmentShader: FOG_FRAG,\n        depthWrite: false,\n        blending: THREE.NoBlending,\n        uniforms: { uTime: { value: 0 } },\n      }),\n    [],\n  );\n\n  useEffect(\n    () => () => {\n      for (const g of geos) g.dispose();\n      fogPlane.dispose();\n      nodeMat.dispose();\n      tokenMat.dispose();\n      edgeMat.dispose();\n      fogMat.dispose();\n    },\n    [geos, fogPlane, nodeMat, tokenMat, edgeMat, fogMat],\n  );\n\n  const worldTimeRef = useRef(0);\n  const lastElapsedRef = useRef(0);\n  // Tracks WHICH sim got its warm-up burst — a rebuilt sim (orientation\n  // flip) must warm again or the net starts dead for seconds.\n  const warmedSimRef = useRef<FlowSim | null>(null);\n  const spinRef = useRef<THREE.Group>(null);\n\n  /* eslint-disable react-hooks/immutability -- R3F frame loop drives\n     GL uniform state by design; nothing here is React state. */\n  useFrame((state) => {\n    const speed = reduced() ? 0 : knobsRef.current.speed;\n    const dt = Math.min(state.clock.elapsedTime - lastElapsedRef.current, 0.25);\n    lastElapsedRef.current = state.clock.elapsedTime;\n    worldTimeRef.current += dt * Math.max(speed, 0);\n    const t = reduced() ? 8 : worldTimeRef.current;\n\n    // Warm streaks once per sim so the first painted frame already\n    // shows a working net (and the reduced-motion still isn't empty).\n    const justWarmed = warmedSimRef.current !== sim;\n    if (justWarmed) {\n      for (let i = 0; i < 120; i++) sim.step(1 / 30, 1);\n      warmedSimRef.current = sim;\n    } else if (!reduced()) {\n      sim.step(dt * Math.max(speed, 0), knobsRef.current.intensity);\n    }\n    sim.writeTokenBuffers();\n\n    for (const [geo, attr] of [\n      [geos[0]!, 'aAct'],\n      [geos[1]!, 'position'],\n      [geos[1]!, 'aAct'],\n    ] as const) {\n      const a = geo.getAttribute(attr) as THREE.BufferAttribute;\n      a.needsUpdate = true;\n    }\n\n    const cam = state.camera as THREE.PerspectiveCamera;\n    // Clamp: zoom=0 would yield fov=Infinity and a NaN projection.\n    const fov = BASE_FOV / Math.max(knobsRef.current.zoom, 0.05);\n    if (Math.abs(cam.fov - fov) > 0.01) {\n      cam.fov = fov;\n      cam.updateProjectionMatrix();\n    }\n    const drift = knobsRef.current.drift;\n    cam.position.set(\n      camCfg.base.x + Math.sin(t * 0.11) * 0.16 * drift,\n      camCfg.base.y + Math.sin(t * 0.073 + 2.0) * 0.09 * drift,\n      camCfg.base.z + Math.cos(t * 0.091) * 0.16 * drift,\n    );\n    cam.lookAt(camCfg.target);\n\n    const height = state.gl.domElement.height;\n    const fovTan = Math.tan(THREE.MathUtils.degToRad(cam.fov * 0.5));\n    for (const m of [nodeMat, tokenMat]) {\n      m.uniforms.uHeightPx!.value = height;\n      m.uniforms.uFovTan!.value = fovTan;\n      m.uniforms.uDpr!.value = state.gl.getPixelRatio();\n      m.uniforms.uFocusDist!.value = knobsRef.current.focus;\n      m.uniforms.uBlurK!.value = knobsRef.current.blurK;\n      m.uniforms.uExposure!.value = knobsRef.current.exposure;\n    }\n    if (spinRef.current) spinRef.current.rotation.y = t * knobsRef.current.spin;\n    fogMat.uniforms.uTime!.value = t;\n    // Under frameloop='demand' an unconditional invalidate keeps a\n    // vsync-rate loop alive forever — reduced motion paints its still\n    // on warm-up, then rendering truly stops.\n    if (!reduced() || justWarmed) invalidate();\n  });\n  /* eslint-enable react-hooks/immutability */\n\n  return (\n    <group>\n      <mesh\n        geometry={fogPlane}\n        material={fogMat}\n        position={camCfg.fogPos.toArray()}\n        rotation={[0, camCfg.fogRot, 0]}\n        renderOrder={0}\n        key={orientation}\n      />\n      {/* Turntable: rotate the whole net around the stack's centre\n          without touching the camera or the fog backdrop. */}\n      <group ref={spinRef} position={camCfg.center.toArray()}>\n        <group position={camCfg.center.clone().negate().toArray()}>\n          <lineSegments geometry={geos[2]} material={edgeMat} renderOrder={1} />\n          <points geometry={geos[0]} material={nodeMat} renderOrder={2} />\n          <points geometry={geos[1]} material={tokenMat} renderOrder={3} />\n        </group>\n      </group>\n    </group>\n  );\n}\n\nfunction makePointMaterial() {\n  return new THREE.ShaderMaterial({\n    vertexShader: POINT_VERT,\n    fragmentShader: NODE_FRAG,\n    transparent: true,\n    depthWrite: false,\n    blending: THREE.AdditiveBlending,\n    uniforms: {\n      uHeightPx: { value: 600 },\n      uFovTan: { value: 0.3 },\n      uDpr: { value: 1 },\n      uFocusDist: { value: 8.0 },\n      uBlurK: { value: 4 },\n      uBaseSize: { value: 0.025 },\n      uExposure: { value: 1.45 },\n    },\n  });\n}\n\nfunction makeTokenMaterial() {\n  return new THREE.ShaderMaterial({\n    vertexShader: POINT_VERT,\n    fragmentShader: TOKEN_FRAG,\n    transparent: true,\n    depthWrite: false,\n    blending: THREE.AdditiveBlending,\n    uniforms: {\n      uHeightPx: { value: 600 },\n      uFovTan: { value: 0.3 },\n      uDpr: { value: 1 },\n      uFocusDist: { value: 8.0 },\n      uBlurK: { value: 4 },\n      uBaseSize: { value: 0.011 },\n      uExposure: { value: 1.35 },\n    },\n  });\n}\n\nfunction reduced(): boolean {\n  return (\n    typeof window !== 'undefined' && window.matchMedia('(prefers-reduced-motion: reduce)').matches\n  );\n}\n\nexport interface NeuralFlowProps {\n  /** Additional CSS classes for the container element */\n  className?: string;\n  style?: CSSProperties;\n  /** Lay the net down left-to-right (default) or stand it upright,\n      input at the bottom, prediction at the top. */\n  orientation?: NeuralFlowOrientation;\n  /** Packed input samples: class index → base64url 4-bit-nibble strings\n      (see mnist-samples.ts). Omit to use the bundled real MNIST set;\n      pass a record missing any class (or {}) to fall back to\n      procedural 5x7 numerals. Keep a STABLE reference — decoding and\n      all geometry rebuild on prop identity. */\n  inputSamples?: Record<number, readonly string[]>;\n  zoom?: number;\n  exposure?: number;\n  focus?: number;\n  blurK?: number;\n  speed?: number;\n  intensity?: number;\n  drift?: number;\n  spin?: number;\n}\n\nexport const NeuralFlow: FC<NeuralFlowProps> = ({\n  className,\n  style,\n  orientation = 'horizontal',\n  inputSamples,\n  zoom,\n  exposure,\n  focus,\n  blurK,\n  speed,\n  intensity,\n  drift,\n  spin,\n}) => {\n  // atob-less environments (SSR, exotic runtimes): neither the bundled\n  // samples nor the trained weights can decode — render the static\n  // backdrop below instead of crashing mid-tree.\n  const noAtob = typeof atob === 'undefined';\n  const { blobs, images } = useMemo(() => {\n    const src = inputSamples === undefined ? MNIST_SAMPLES : inputSamples;\n    if (noAtob || Object.keys(src).length === 0) {\n      return { blobs: glyphBlobs(), images: [] as Float32Array[][] };\n    }\n    const decoded = decodeAll(src);\n    return decoded.blobs.every((v) => v.length > 0)\n      ? decoded\n      : { blobs: glyphBlobs(), images: [] as Float32Array[][] };\n  }, [inputSamples, noAtob]);\n  const knobs = useRef<NeuralFlowKnobs>({ ...DEFAULT_KNOBS });\n  useEffect(() => {\n    knobs.current = {\n      zoom: zoom ?? DEFAULT_KNOBS.zoom,\n      exposure: exposure ?? DEFAULT_KNOBS.exposure,\n      focus: focus ?? DEFAULT_KNOBS.focus,\n      blurK: blurK ?? DEFAULT_KNOBS.blurK,\n      speed: speed ?? DEFAULT_KNOBS.speed,\n      intensity: intensity ?? DEFAULT_KNOBS.intensity,\n      drift: drift ?? DEFAULT_KNOBS.drift,\n      spin: spin ?? DEFAULT_KNOBS.spin,\n    };\n  }, [zoom, exposure, focus, blurK, speed, intensity, drift, spin]);\n\n  if (noAtob) {\n    return (\n      <div\n        aria-hidden=\"true\"\n        className={className}\n        style={{ ...style, position: 'absolute', inset: 0, background: BACKDROP }}\n      />\n    );\n  }\n\n  return (\n    <div\n      className={className}\n      style={{\n        ...style,\n        position: 'absolute',\n        inset: 0,\n        background: BACKDROP,\n      }}\n    >\n      <Canvas\n        aria-hidden=\"true\"\n        dpr={[1, 2]}\n        frameloop={reduced() ? 'demand' : 'always'}\n        camera={{ position: camFor(orientation).base.toArray(), fov: BASE_FOV }}\n        gl={{\n          // Opaque canvas — sidesteps compositing quirks around\n          // premultiplied additive pixels (renders black-transparent on\n          // some stacks, stalls readback on others). The backdrop tone\n          // comes from the clear color + fog plane instead.\n          alpha: false,\n          antialias: true,\n          powerPreference: 'high-performance',\n          preserveDrawingBuffer: true,\n        }}\n        onCreated={({ gl }) => {\n          gl.setClearColor(0x010806, 1);\n        }}\n      >\n        <SceneContent knobs={knobs} orientation={orientation} blobs={blobs} images={images} />\n      </Canvas>\n    </div>\n  );\n};\n","/* cspell:disable */\n/**\n * Real MNIST test-set samples: four canonical images per class\n * (closest to the class centroid, via sklearn fetch_openml). 784 greyscale pixels\n * quantized to 4-bit nibbles, packed two per byte, base64url-encoded.\n * ~392 bytes per sample — 40 samples in ~16 KB text.\n */\nexport const MNIST_SAMPLES: Record<number, string[]> = {\n  0: [\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAF4tgAAAAAAAAAAAAAAff_-gAAAAAAAAAAAACv____8IAAAAAAAAAAD7_2v__-gAAAAAAAAAC3_-wAv__UAAAAAAAAAr_-yAGz_9QAAAAAAAAf_9wAAAu_1AAAAAAAALv-hAAAAb_kQAAAAAALf-wAAAABf_RAAAAAAA__iAAAAAF_-IAAAAAAG_4AAAAAAf_8wAAAAAAr_cAAAAAPv_AAAAAAACv8AAAAACP_VAAAAAAAK_wAAAACP_1AAAAAAAAr_AAAAB__6AAAAAAAACv9gAAKf_9EAAAAAAAAK_7iL___-YAAAAAAAAAb______7IAAAAAAAAAAY7___6kAAAAAAAAAAAABI35gQAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAACzKIAAAAAAAAAAAAAAE3_-wAAAAAAAAAAAAAE3__2AAAAAAAAAAAAAC3____HEAAAAAAAAAAEv__t__-gAAAAAAAAAAj9z6FL__YAAAAAAAAADfgXYAF__kAAAAAAAAA-0QAAABf_0AAAAAAAA9-AAAAAA__gAAAAAAAF_zAAAAAAn-AAAAAAAAn5AAAAAAG_4AAAAAAADvcAAAAABP-AAAAAAAAO9wAAAAA-_AAAAAAAAA76AAAAAJ_3AAAAAAAADvwAAAAZ_6AAAAAAAAAO_RAAAb_6AAAAAAAAAA7_1Cd9_6AAAAAAAAAADP_____6AAAAAAAAAAAE7____WAAAAAAAAAAAABIr_tQAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAHuAAAAAAAAAAAAAAAAU7_oAAAAAAAAAAAAAAG__9RAAAAAAAAAAAAAD7___ySAAAAAAAAAAAAn_____5SAAAAAAAAAAHP_XREXvsQAAAAAAAACf_zAAAE74AAAAAAAAAu__EAAAC_4QAAAAAAAG_9gQAAAL_0AAAAAAABz_cAAAAAv_gAAAAAAAn_kQAAAALf-AAAAAAAHP8wAAAACP_xAAAAAAB_-wAAAABP_6EAAAAAAL_7AAAAB9_8EAAAAAAAz_oAAAAf__UAAAAAAACf9AAABc__YQAAAAAAAJ_6AAWP_-UQAAAAAAAAb_52j__6IAAAAAAAAAAK____-UEAAAAAAAAAAAAjashAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA1VaYAAAAAAAAAAAAAed___lAAAAAAAAAAADj____bYAAAAAAAAAAAz_-Xz7EAAAAAAAAAAAj_-RB__BAAAAAAAAAAP_-QAAj_gAAAAAAAAAG_-hAAAZ_jAAAAAAAAK__lAAAAXvgAAAAAAAC__2AAAAAF_gAAAAAAAO_4AAAAAAL-MAAAAAAG_9MAAAAAAv-QAAAAAAn_kAAAAAAC_5AAAAAACf8gAAAAAAj_kAAAAAAJ_yAAAAAAf_4QAAAAAAn_IAAAAAfP-QAAAAAACf-yAAASj_2BAAAAAAAF__-Hd8__sQAAAAAAAADf______wgAAAAAAAAAAXv___9kQAAAAAAAAAAAEr7VVMAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n  ],\n  1: [\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAeyAAAAAAAAAAAAAAAAB_oAAAAAAAAAAAAAAAAH-wAAAAAAAAAAAAAAAAf8EAAAAAAAAAAAAAAACP8QAAAAAAAAAAAAAAAd_RAAAAAAAAAAAAAAAF_7AAAAAAAAAAAAAAAAj_sAAAAAAAAAAAAAAACv9wAAAAAAAAAAAAAAAO_1AAAAAAAAAAAAAAAD__EAAAAAAAAAAAAAAAj_sAAAAAAAAAAAAAAACv9gAAAAAAAAAAAAAAAe_iAAAAAAAAAAAAAAAI_5AAAAAAAAAAAAAAAAv_YAAAAAAAAAAAAAAAC_4gAAAAAAAAAAAAAAAf9wAAAAAAAAAAAAAAAB7SAAAAAAAAAAAAAAAAC3AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAlgAAAAAAAAAAAAAAAAf9AAAAAAAAAAAAAAAALf0AAAAAAAAAAAAAAABPlTAAAAAAAAAAAAAAAG_lAAAAAAAAAAAAAAAD_9EAAAAAAAAAAAAAAAr_oAAAAAAAAAAAAAAADP9QAAAAAAAAAAAAAAAM_1AAAAAAAAAAAAAAAD79EAAAAAAAAAAAAAAAb_wAAAAAAAAAAAAAAAPv9gAAAAAAAAAAAAAABv_RAAAAAAAAAAAAAAAd_7AAAAAAAAAAAAAAAF_-YAAAAAAAAAAAAAAAb4gAAAAAAAAAAAAAAADfogAAAAAAAAAAAAAAA-tgAAAAAAAAAAAAAAADrBAAAAAAAAAAAAAAAANJAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAABpAAAAAAAAAAAAAAAAAL9wAAAAAAAAAAAAAAAAv6AAAAAAAAAAAAAAAADPoAAAAAAAAAAAAAAABP-QAAAAAAAAAAAAAAAF_RAAAAAAAAAAAAAAAAv_MAAAAAAAAAAAAAAAC_wQAAAAAAAAAAAAAAAu-wAAAAAAAAAAAAAAAD_7AAAAAAAAAAAAAAAAP_UAAAAAAAAAAAAAAAA_wQAAAAAAAAAAAAAAAJ-QAAAAAAAAAAAAAAAF70AAAAAAAAAAAAAAAAr-IAAAAAAAAAAAAAAACvgAAAAAAAAAAAAAAAAK9AAAAAAAAAAAAAAAAArAAAAAAAAAAAAAAAAACWAAAAAAAAAAAAAAAAAHAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAMoAAAAAAAAAAAAAAAAC7QAAAAAAAAAAAAAAAAP-IAAAAAAAAAAAAAAACP9AAAAAAAAAAAAAAAAJ_QAAAAAAAAAAAAAAAAn9AAAAAAAAAAAAAAAADvoAAAAAAAAAAAAAAAAP8gAAAAAAAAAAAAAAAA_wAAAAAAAAAAAAAAAAr-AAAAAAAAAAAAAAAADfkAAAAAAAAAAAAAAAAN-QAAAAAAAAAAAAAAAB31AAAAAAAAAAAAAAAAT_MAAAAAAAAAAAAAAACPwAAAAAAAAAAAAAAAAK-wAAAAAAAAAAAAAAAAr2AAAAAAAAAAAAAAAAA-YAAAAAAAAAAAAAAAAAZwAAAAAAAAAAAAAAAAAcAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n  ],\n  2: [\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAABu3cQAAAAAAAAAAAAABv__-IAAAAAAAAAAAAAX3af-xAAAAAAAAAAAACHEAj_oAAAAAAAAAAAABEAAc-wAAAAAAAAAAAAAAAAb7AAAAAAAAAAAAAAAABf4wAAAAAAAAAAAAAAAA30AAAAAAAAAAAAAAAAPtIAAAAAAAAAAAAAAABvsAAAAAAAAAAAAAAAAJ-wAAAAAAAAAAAAAAAG_7AAAAAAAAAAAAAAAAn_QAAAAAAAAAAABEAAXvgQAAAAAAAAAANu7Mz_8gAAAAAAAAAAHf_u___EEAAAAAAAAAB_933___-0IQAAAAAAAI_73_1t3u7ZAAAAAAAADP__kAACUwAAAAAAAAAF_-YAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAASd-RAAAAAAAAAAAAAAff-_0gAAAAAAAAAAAAT__Bn8AAAAAAAAAAAAA_-hAM-AAAAAAAAAAAAAcwAAf3AAAAAAAAAAAAAAAABvoAAAAAAAAAAAAAAAAG-gAAAAAAAAAAAAAAAAb6AAAAAAAAAAAAAAAABvgAAAAAAAAAAAAAAAAG4gAAAAAAAAAAAAAAAC3AAAAAAAAAAAAAAAKb78AAAAAAAAAAARAnU___4gAAAAAAAAB-3P_9___4AAAAAAAAAHKfkiJcePUAAAAAAAAAicgQFN9G9wAAAAAAAACvoFnf9wTaAAAAAAAAAK-l_P_BACUAAAAAAAAAPP__-wAAAAAAAAAAAAAEz5lhAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAbHAAAAAAAAAAAAAAAAb_9wAAAAAAAAAAAAAAB-X8AAAAAAAAAAAAAAAMgM4AAAAAAAAAAAAAAC5w2QAAAAAAAAAAAAAACBP5AAAAAAAAAAAAAAABB_kAAAAAAAAAAAAAAAAH9gAAAAAAAAAAAAAAAAviAAAAAAAAAAAAAAAAHdAAAAAAAAAAAAAAAABvgAAAAAAAAAAAAAAAAK9gAAAAAAAAAAAAACEB3RAAAAAAAAAAAAAH_Xf4AAAAAAAAAAAAAF_u__QAAAAAAAAAAAAC2gLP-RAAAAAAAAAAAAn2BO__21AAAAAAAAAAB_Mc9ie5QAAAAAAAAAAC_N-QAAAAAAAAAAAAAAB-9wAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAACv9MAAAAAAAAAAAAAAD7__SAAAAAAAAAAAAAB7qN_0QAAAAAAAAAAAAj1AAn2AAAAAAAAAAAAAVAAAM0QAAAAAAAAAAAAAAAAbyAAAAAAAAAAAAAAAABPIAAAAAAAAAAAAAAAAH4gAAAAAAAAAAAAAAAArAAAAAAAAAAAAAAAAAP5AAAAAAAAAAAAAAAACtIAAAAAAAAAAAAAFHyvcAAAAAAAAAAAAEnf__-RAAAAAAAAAAAG__zP_foAAAAAAAAAAC38Mb-AeQAAAAAAAAAAv7EZ9wAAAAAAAAAAAADOIb-gAAAAAAAAAAAAAdo7-iAAAAAAAAAAAAAAz_-iAAAAAAAAAAAAAACOswAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n  ],\n  3: [\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAs8wAAAAAAAAAAAAAAF5_9cyAAAAAAAAAAAACf___7-gAAAAAAAAAAAH_93HjvkAAAAAAAAAAAbHEQAe-AAAAAAAAAAAAAAABO-yAAAAAAAAAAAAAACf_CAAAAAAAAAAAAE4q_9xAAAAAAAAAAAAB____SAAAAAAAAAAAAAH___4MAAAAAAAAAAAAAbNhb5QAAAAAAAAAAAAAAAAAosAAAAAAAAAAAAAAAAADWAAAAAAAAAAAAAAAAALkAAAAAAAAAAAAAAAAA6QAAAAAAAAAAAAAAAAr4AAAAAAAAFwAAAAAAH_MAAAAAAAAv54ZERK_ccAAAAAAAAAj______6EQAAAAAAAAAEaN_f-zAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAa___2BAAAAAAAAAAAACv____wQAAAAAAAAAAAG__7uv5AAAAAAAAAAAAFmUABvkAAAAAAAAAAAAAAAFO-QAAAAAAAAAAAAAAGf_4AAAAAAAAAAAAABOf_qEAAAAAAAAAAAA0v__1AAAAAAAAAAAAB_____kgAAAAAAAAAAAE7__O_NMAAAAAAAAAAABLthNBmQAAAAAAAAAAAAAAAAAJAAAAAAAAAAAAAAAAAA0QAAAAAAAAAAAAAAAADUAAAAAAAAAAAAAAAABeYAAAAAAAAAAAAAAAKP0QAAAAAAAAAXlmZmzv-QAAAAAAAAAI_______2AAAAAAAAAAX_____1RAAAAAAAAAAAYubpgAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAlPvpolhAAAAAAAAAAAE7P____1AAAAAAAAAAAK3jpV-_1AAAAAAAAAAABQAAAH_0AAAAAAAAAAAAAAASv9QAAAAAAAAAAAAAAXv_UAAAAAAAAAAAAAGj_-xAAAAAAAAAAAAAF__5AAAAAAAAAAAAAAATv_DAAAAAAAAAAAAAAAG__5QAAAAAAAAAAAAAABf_7EAAAAAAAAAAAAAAAO_-hAAAAAAAAAAAAAAAD7_QAAAAAAAAAAAAAAACv9AAAAAAAAAAAAAAAAb_0AAAAAAAAAAAAAAG8_8EAAAAAAAAAAAADjP__YAAAAAAAAAAAKq7__-UAAAAAAAAAAABv_9wzMAAAAAAAAAAAAI_3EAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAFYjugxAAAAAAAAAAAALP____wAAAAAAAAAAAA__Jn__0AAAAAAAAAAACcQABv-IAAAAAAAAAAAAAAAX_wAAAAAAAAAAAAAAAf_9QAAAAAAAAAAAAAAj_-QAAAAAAAAAAAAAj3_9gAAAAAAAAAAAABO__-1EAAAAAAAAAAAAK_____cUAAAAAAAAAAAPf___v_0AAAAAAAAAAAARVVDXvoAAAAAAAAAAAAAAAAv-gAAAAAAAAAAAAAAA9_zAAAAAAAAAAAAAAA9_3EAAAAAAAAAAAAAKO_5AAAAAAAAAAAABnfP_mAAAAAAAAAAAABP__-yAAAAAAAAAAAAAe__6AAAAAAAAAAAAAAAj8QQAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n  ],\n  4: [\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAH4QAAAAAAAAAAAAAAAAj3AAAAAAAAAAAAAAAADPcAAAAAAAAAAAAAAAAd8wAAAAAAAAAAAnMAAG_iAAAAAAAAAAAa9gAAj6AAAAAAAAAAAa_jAADPcAAAAAAAAAAa_3AABv9AAAAAAAAAAG_3AAAI_BAAAAAAAAAB78AAAB74cwAAAAAAAACv-3VXz__7AAAAAAAAABn______ZMAAAAAAAAAAEmcq_-hAAAAAAAAAAAAAAAG_0AAAAAAAAAAAAAAABz6AAAAAAAAAAAAAAAAX_UAAAAAAAAAAAAAAAHPoAAAAAAAAAAAAAAABP9AAAAAAAAAAAAAAAAJ-gAAAAAAAAAAAAAAAAbSAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAeQAAAAAAAAAAAAAAAAG-QAAAAAAAAAAAAAAAAb5AAAAAAAAAAALIAAACPYAAAAAAAAAAC5wAAAJ8wAAAAAAAAAABRAAAAqwAAAAAAAAAAJAAAAAFhAAAAAAAAAACVAAAABvgAAAAAAAAABPcAAABL9SAAAAAAAAANUAAAAq_8kAAAAAAAA6sRdXjO__-QAAAAAAAJ_93_7___tjAAAAAAAAK93dyHdMkQAAAAAAAAAAAAAAAW5QAAAAAAAAAAAAAAAHfhAAAAAAAAAAAAAAACnWAAAAAAAAAAAAAAAAiJAAAAAAAAAAAAAAAAOpYAAAAAAAAAAAAAAAAEoAAAAAAAAAAAAAAAAEswAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAtgAAAAAAAAAAAAAAAAHmAAAAAAAAAAABcAAABPYAAAAAAAAAAAfiAAAJ8gAAAAAAAAAACbAAAA7AAAAAAAAAAABPQAAAP3AAAAAAAAAAAJ0QAACPYAAAAAAAAAAE-AAAAL4QAAAAAAAAABz2IAAD-QAAAAAAAAAAb__tynv2AAAAAAAAAABv7-____YAAAAAAAAAABISEiJe4QAAAAAAAAAAAAAAAD9wAAAAAAAAAAAAAAAAb3AAAAAAAAAAAAAAAACPcAAAAAAAAAAAAAAAAL9wAAAAAAAAAAAAAAAA3yAAAAAAAAAAAAAAAADdEAAAAAAAAAAAAAAAAuwAAAAAAAAAAAAAAAAB5wAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAABtEAAAAAAAAAAAAAAAAG4gAAAAAAAAAAABoQAAbzAAAAAAAAAAAAX2AABvMAAAAAAAAAAACvYAAG0QAAAAAAAAAAAvwQAAbRAAAAAAAAAAAK-AAACsAAAAAAAAAAAB2hAAAMwAAAAAAAAAAAf2AAAE_AAAAAAAAAAACfgAAAn3AAAAAAAAAAAJ__-p7_IAAAAAAAAAAAf_____oAAAAAAAAAAAAFrbiv5QAAAAAAAAAAAAABAD-gAAAAAAAAAAAAAAAA32AAAAAAAAAAAAAAAAD_EAAAAAAAAAAAAAAAB_kAAAAAAAAAAAAAAAAd9wAAAAAAAAAAAAAAAD-wAAAAAAAAAAAAAAAAP0AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n  ],\n  5: [\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAsgAAAAAAAAAAAAAAAF-oQAAAAAAAAB5IAAkjtYAAAAAAAAAAC38zP-1AAAAAAAAAAAACcqqUgAAAAAAAAAAAAAtcAAAAAAAAAAAAAAAAH9AAAAAAAAAAAAAAAAB3AAAAAAAAAAAAAAAAAX-3VAAAAAAAAAAAAAABf_L-AAAAAAAAAAAAAADtQCfUAAAAAAAAAAAAAAAAB6QAAAAAAAAAAAAAAAACeEAAAAAAAAAAAAAAAAE9wAAAAAAAAAAAAAAAAT3AAAAAAAAAAAAAAAAB_QAAAAAAAAAAFAAAABewQAAAAAAAAAAexAAGv4gAAAAAAAAAAAa65v_kgAAAAAAAAAAAAF8_KQAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAALYAAAAAAAAAAAAAAABc5AAAAAAAAAAAAAAAKu0gAAAAAAAAAAAAABfvoQAAAAAAAAAAAAAFz8MAAAAAAAAAAAAAAr_nAAAAAAAAAAAAAAAr6yAAAAAAAAAAAAAAAG9gAAAAAAAAAAAAAAAAL_2WEAAAAAAAAAAAAAADjf_rUAAAAAAAAAAAAAAAA535AAAAAAAAAAAAAAAAAK9AAAAAAAAAAAAAAAAAHmAAAAAAAAAAAAAAAAANoAAAAAAAAAAAAAAAAB5gAAAAAAAAAAAAAAABr0AAAAAAAAAAAAAAABz3AAAAAAAAAAAAAAADz3AAAAAAAAAAAAAAJa7UAAAAAAAAAAAAAACfthAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAfxAAAAAAAAAAAAAAAAC_wQAAAAAAAAAAAAAAB-38EAAAAAAAAAAAAABewHoQAAAAAAAAAAAAAc4wAAAAAAAAAAAAAAAJ-AAAAAAAAAAAAAAAAA3zAAAAAAAAAAAAAAAAT6EgAAAAAAAAAAAAAABP_e1gAAAAAAAAAAAAAAXNzfgAAAAAAAAAAAAAAAEB_2AAAAAAAAAAAAAAAACPcAAAAAAAAAAAAAAAAC-QAAAAAAAAAAAAAAAAL3AAAAAAAAAAAAAAAACfcAAAAAAAAAAAAAAAAP5AAAAAAAAAAAAAAAAG9gAAAAAAAAAAAAAAAG7SAAAAAAAAAAAAAAAp6QAAAAAAAAAAAAAAAItQAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAABsEAAAAAAAAAAAAAAAAG8gAAAAAAAAAAAAAAAAmgB2AAAAAAAAAAAAAALYWfkAAAAAAAAAAAAAK__9kQAAAAAAAAAAAAB_6WEAAAAAAAAAAAAAAc0gAAAAAAAAAAAAAAAJ9AAAAAAAAAAAAAAAAq9gAAAAAAAAAAAAAABu_6mWMAAAAAAAAAAAAJ6e7u_5IAAAAAAAAAAAAAAAA4-QAAAAAAAAAAAAAAAAB_QAAAAAAAAAAAAAAAAJ9AAAAAAAAAAAAAAAAI-QAAAAAAAAAAAAAAAI-xAAAAAAAAAAAANQBt-QAAAAAAAAAAAAfiS_swAAAAAAAAAAAAX-z9YAAAAAAAAAAAAABf6nEAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n  ],\n  6: [\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAMoAAAAAAAAAAAAAAAAE_1AAAAAAAAAAAAAAABn-EAAAAAAAAAAAAAAAL_kAAAAAAAAAAAAAAAGv0wAAAAAAAAAAAAAACP9AAAAAAAAAAAAAAAAM-AAAAAAAAAAAAAAAAI_iAAAAAAAAAAAAAAAB73AAAAAAAAAAAAAAAAj8AAAgAAAAAAAAAAAAT_UABNxwAAAAAAAAAACPkABM_-UAAAAAAAAAAM8QXv___AAAAAAAAAAAzxn_7HT9IAAAAAAAAALpD_wQAvwAAAAAAAAABv9_0AAc_AAAAAAAAAAG___mR9_WAAAAAAAAAAPv_____EAAAAAAAAAAAE7____xAAAAAAAAAAAAAYjNhRAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA-kAAAAAAAAAAAAAAAAL9AAAAAAAAAAAAAAAAa-QAAAAAAAAAAAAAAAI_iAAAAAAAAAAAAAAAC72AAAAAAAAAAAAAAAAX_AAAAAAAAAAAAAAAACfsAAAAAAAAAAAAAAABf8wAAAAAAAAAAAAAAAI_QAAAAAAAAAAAAAAAD7zAAAAAAAAAAAAAAAAf9EAAAAAAAAAAAAAAACvoAAXyDAAAAAAAAAABO4wCO__9wAAAAAAAAAH-QGv_Lz_EAAAAAAAAAf3Pf-RBP0QAAAAAAAAA_3v1AAE-AAAAAAAAAAB7_9gIlzgAAAAAAAAAACP__vv_3AAAAAAAAAAAAr____nAAAAAAAAAAAAAp__-1AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA8cAAAAAAAAAAAAAAAAbwAAAAAAAAAAAAAAAAb9QAAAAAAAAAAAAAAAJ-AAAAAAAAAAAAAAAAF-yAAAAAAAAAAAAAAAF7TAAAAAAAAAAAAAAAD7kAAAAAAAAAAAAAAABv3AAAAAAAAAAAAAAAAj6AAAAAAAAAAAAAAAATdIAAAAAAAAAAAAAAAG_QAAFdzAAAAAAAAAABvgABc__9AAAAAAAAAAb8gGf_nPJAAAAAAAAAF9wK-cxALkAAAAAAAAAf2CuUAAD5wAAAAAAAABPZfUAAAnBAAAAAAAAAB6IsQAAbkAAAAAAAAAACv5wAAXnAAAAAAAAAAADz9mq33AAAAAAAAAAAAAY7_xzAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAC7QAAAAAAAAAAAAAAABziAAAAAAAAAAAAAAAAj2AAAAAAAAAAAAAAAAPrAAAAAAAAAAAAAAAACOMAAAAAAAAAAAAAAAA_kAAAAAAAAAAAAAAAAL9AAAAAAAAAAAAAAAAD-AAAAAAAAAAAAAAAAArSAAAAAAAAAAAAAAAAX4AAAAAAAAAAAAAAAADOAAAAAAAAAAAAAAAAAekAABIgAAAAAAAAAAAF9AAK3_sQAAAAAAAAAAzzA866n0AAAAAAAAAADuA-9wAOkAAAAAAAAAAO4J-AAA7QAAAAAAAAAAz0mQAAn0AAAAAAAAAABfmLIDv4AAAAAAAAAAAAz9_t7mAAAAAAAAAAAAAp__mSAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n  ],\n  7: [\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAnef-QAAAAAAAAAABq_____-MAAAAAAAAAAN____6v9gAAAAAAAAAAr_-pMW_9AAAAAAAAAAAUkwAAr_oAAAAAAAAAAAAAAAHP5AAAAAAAAAAAAAAAA__QAAAAAAAAAAAAAAAF_7AAAAAAAAAAAAAAABz_UAAAAAAAAAAAAAAAX_sQAAAAAAAAAAAAAAHP9QAAAAAAAAAAAAAACf-xAAAAAAAAAAAAAAA-_RAAAAAAAAAAAAAAAE_8AAAAAAAAAAAAAAAAT_cAAAAAAAAAAAAAAACf9AAAAAAAAAAAAAAAAM9wAAAAAAAAAAAAAAAD70AAAAAAAAAAAAAAAAX_sAAAAAAAAAAAAAAAAI9AAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAASHERif_EAAAAAAAAAAB_______cAAAAAAAAAAF____7P9wAAAAAAAAAAA5lDMh33AAAAAAAAAAAAAAAADPEAAAAAAAAAAAAAAABf8AAAAAAAAAAAAAAAAL-wAAAAAAAAAAAAAAAC74AAAAAAAAAAAAAAAAX_MAAAAAAAAAAAAAAAHPoAAAAAAAAAAAAAAABf8wAAAAAAAAAAAAAAAI_BAAAAAAAAAAAAAAAC32AAAAAAAAAAAAAAAAn9EAAAAAAAAAAAAAAADvwAAAAAAAAAAAAAAAAe4gAAAAAAAAAAAAAAAF-QAAAAAAAAAAAAAAAAr3AAAAAAAAAAAAAAAALfcAAAAAAAAAAAAAAABPwQAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAACOJn_EAAAAAAAAAAFfM7___9AAAAAAAAABpv__WZN-AAAAAAAAAAI__oxAE3iAAAAAAAAAALFAAAAX9EAAAAAAAAAAAAAAACvcAAAAAAAAAAAAAAABPsgAAAAAAAAAAAAAAAI-wAAAAAAAAAAAAAAABj6AAAAAAAAAAAAAAAAX6EAAAAAAAAAAAAAAAC_oAAAAAAAAAAAAAAAAO9QAAAAAAAAAAAAAAAK9wAAAAAAAAAAAAAAAB30AAAAAAAAAAAAAAAATqEAAAAAAAAAAAAAAACvQAAAAAAAAAAAAAAAAv0gAAAAAAAAAAAAAAAC-AAAAAAAAAAAAAAAAAj7AAAAAAAAAAAAAAAABtIAAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAEn7mdpAAAAAAAAAAAAF_____9MAAAAAAAAAAF3_____9QAAAAAAAAAAnu__3Mz5AAAAAAAAAAACJEQQAvsAAAAAAAAAAAAAAAAL-AAAAAAAAAAAAAAAAE_iAAAAAAAAAAAAAAAAr4AAAAAAAAAAAAAAAAr_QAAAAAAAAAAAAAAAX_gAAAAAAAAAAAAAAABf8wAAAAAAAAAAAAAAAW-wAAAAAAAAAAAAAAAL_gAAAAAAAAAAAAAAAJ_2AAAAAAAAAAAAAAAB76AAAAAAAAAAAAAAAAn-QAAAAAAAAAAAAAAAXvkAAAAAAAAAAAAAAADP4gAAAAAAAAAAAAAAAf9gAAAAAAAAAAAAAAAA3RAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n  ],\n  8: [\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAGP_9MAAAAAAAAAAAAAXv___BAAAAAAAAAAAAn__6rvMAAAAAAAAAAAf_tQAI9wAAAAAAAAAATvsQAAv3AAAAAAAAAADf9AAALvMAAAAAAAAAAG_2AALvsQAAAAAAAAAAC_sgTv4wAAAAAAAAAAAF_9bv5AAAAAAAAAAAAAB-__9AAAAAAAAAAAAAAB3_-wAAAAAAAAAAAAACv__9EAAAAAAAAAAAAAv-uO-QAAAAAAAAAAAAj_YAX-IAAAAAAAAAAADfYAAK8wAAAAAAAAAAANsAABz3AAAAAAAAAAAA34EF3_MAAAAAAAAAAACP____oQAAAAAAAAAAAAr__-MAAAAAAAAAAAAAAHz4IAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAGv_qxAAAAAAAAAAAAAK____swcgAAAAAAAAAE_8iI78eAAAAAAAAAAC34AAAc_0AAAAAAAAAAP_YAACv5AAAAAAAAAAAv9gABv8EAAAAAAAAAAAf6EBz9QAAAAAAAAAAAAe9gr8UAAAAAAAAAAAAAr-j_QAAAAAAAAAAAAAAL__UAAAAAAAAAAAAAAAn_8wAAAAAAAAAAAAAAf-73AAAAAAAAAAAAABr_Zv8gAAAAAAAAAAAAP_oR33AAAAAAAAAAAABv4wCvcAAAAAAAAAAAAJ9AAI_AAAAAAAAAAAAAn0AAb4AAAAAAAAAAAACP9Cj_cAAAAAAAAAAAAC7___0wAAAAAAAAAAAAAo_-kwAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAABI6_9MAAAAAAAAAAAAYz___-zAAAAAAAAAAAI_-pb_99QAAAAAAAAAC75EAf8HXAAAAAAAAAAj-EACvdfYAAAAAAAAACPsAANwaoAAAAAAAAAAG_5IAla8wAAAAAAAAAACv_VA7_RAAAAAAAAAAACnv2L-RAAAAAAAAAAAAAC3_9wAAAAAAAAAAAAAATf_wAAAAAAAAAAAAAAf_z_AAAAAAAAAAAAAAf-M_9wAAAAAAAAAAAAC-MD_3AAAAAAAAAAAAB-QALfcAAAAAAAAAAAAOwAA_9AAAAAAAAAAAAE-QAt_wAAAAAAAAAAAAf-id_5AAAAAAAAAAAAAd___5AAAAAAAAAAAAAAWN_nAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAACt9AAAAAAAAAAAAmu1vf_4AAAAAAAAAAAt_u_6dTYAAAAAAAAAAF-iXkAASAAAAAAAAAAArkAKAAPoAAAAAAAAAACaAAEAbfgAAAAAAAAAAE8QAAr_-AAAAAAAAAAACVAAf_5WAAAAAAAAAAAAiUj_wQMAAAAAAAAAAAAK7_wQAAAAAAAAAAAAAATP9QAAAAAAAAAAAAAAB_7YAAAAAAAAAAAAAAA95I4QAAAAAAAAAAAAAK-gb2AAAAAAAAAAAAADzABPkAAAAAAAAAAAAAT5AI9wAAAAAAAAAAAABPcE_lAAAAAAAAAAAAAE-QipAAAAAAAAAAAAAAL8OlIAAAAAAAAAAAAAAE78MAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n  ],\n  9: [\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAABJ74UgAAAAAAAAAAAASv___2AAAAAAAAAAAATulmd-kAAAAAAAAAAAHsIAAAewAAAAAAAAAABsEAAABbYQAAAAAAAAAHQAAAAJ_yAAAAAAAAAB5AAAAAz9EAAAAAAAAAHdQAAVv_UAAAAAAAAAAG_-ruzfYAAAAAAAAAAABHy3MH9QAAAAAAAAAAAAAAAAuQAAAAAAAAAAAAAAAAPxAAAAAAAAAAAAAAAACtAAAAAAAAAAAAAAAAAdQAAAAAAAAAAAAAAAAK4gAAAAAAAAAAAAAAAB1gAAAAAAAAAAAAAAAAXiAAAAAAAAAAAAAAAADIAAAAAAAAAAAAAAAAAuMAAAAAAAAAAAAAAAAJsQAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAny0AAAAAAAAAAAAAAGP_d-QAAAAAAAAAAAALfkwG_QAAAAAAAAAAALPcAAH-wAAAAAAAAAAGvYAAAv8AAAAAAAAAABvgAAALvYAAAAAAAAAAcsAAAG_0QAAAAAAAAAD-QAALP9gAAAAAAAAAAL7AATv_hAAAAAAAAAAAJ-5z9f4AAAAAAAAAAAAO-62CvIAAAAAAAAAAAAAEgA-oAAAAAAAAAAAAAAAAI9AAAAAAAAAAAAAAAAA7AAAAAAAAAAAAAAAAAH7AAAAAAAAAAAAAAAABvUAAAAAAAAAAAAAAAAM0AAAAAAAAAAAAAAAAA3AAAAAAAAAAAAAAAAAP3FwAAAAAAAAAAAAAAA_noAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAGO_pMAAAAAAAAAAAAAC____5ABIAAAAAAAAAB_tCJ9-QYwAAAAAAAABP4QAAL7BjAAAAAAAAAs4wAAC_sGMAAAAAAAAJ9wAABf9QUgAAAAAAAC7iAABO_QBQAAAAAAAAP_qZnf_4AAAAAAAAAAA-____7-EAAAAAAAAAAASIiGM_wAAAAAAAAAAAAAAAAI9wAAAAAAAAAAAAAAAA3wAAAAAAAAAAAAAAAAT9AAAAAAAAAAAAAAAACvYAAAAAAAAAAAAAAAAM8QAAAAAAAAAAAAAAAD_AAAAAAAAAAAAAAAAAr3AAAAAAAAAAAAAAAAHvIAAAAAAAAAAAAAAABf0AAAAAAAAAAAAAAAAD-AAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n    'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAEzz_xAAAAAAAAAAAAAHP_u_9QAAAAAAAAAAAHP0wA2_jAAAAAAAAAAB_1AAAb_QAAAAAAAAAAK-AAAHf4QAAAAAAAAAAriAACf4wAAAAAAAAAACuMABN-QAAAAAAAAAAAK_IjP_SAAAAAAAAAAAATv___4AAAAAAAAAAAAADbIbfcAAAAAAAAAAAAAAAA-wAAAAAAAAAAAAAAAAJ9gAAAAAAAAAAAAAAAA72AAAAAAAAAAAAAAAAL_IAAAAAAAAAAAAAAABvkAAAAAAAAAAAAAAAAd8wAAAAAAAAAAAAAAAG_RAAAAAAAAAAAAAAAAj0AAAAAAAAAAAAAAAAPvAAAAAAAAAAAAAAAAAaoAAAAAAAAAAAAAAAAAAAAAAAAAAAA=',\n  ],\n};\n","/* cspell:disable */\n/**\n * Trained sparse weights for NeuralFlow — MNIST MLP constrained to\n * exactly the rendered 4294-edge topology (train-sparse.py in this\n * directory). FP32 little-endian, base64url; edge order is dst-major,\n * identical to the component's inEdges construction (seed 1337).\n */\nexport const SPARSE_WEIGHTS = {\n  weights1:\n    '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',\n  weights2:\n    'UkZ7PkU_WT89J8I-yWs6KocFsz7U-rm-Okpfv7VFQ75JwDS_whZUP8HTxz9WoGU_bRLBvsNvG7_z6BI_UlcPP78wpr9grmo-SGldvXq-rD6k21Y_IWPrPlJlO75Oj3A-YQdNADfr7j41yIc-laBPv3Kbkj6GnW2_nUqHv1vZHz54w4M_g_2BP7RhvD0WHTG_jZUZP0cG7r6fpRg-xSPavih4rj4vPHC_tiRNP_ihBj-aioo-6FBdP2LQcD-2Gg2_QHGTPifpgL8KmRU6zNZ2P7G2m7-5_Ze-_XsfP8gZWYB_QUE-s5RtAAj9g7zgnty-gsS4vti1Lz__GqW_qdhPP0HVWz4PoM6-XudFP4FAH75_-Sc-TsORPwAGsr-efH0-fckcP9rZ676a6Oi9DQc4Pw3CDj_IelK_m0I1P6E8XwDCXa4_buoCPxGX5wBzHIC-ZFUJv7S1sD5d2xq_Bc5KvuJKBj99PGA_kHycPyhBET-YH6Q_5Dkiv_bgvz8oc8Y-v-xKAIo2kb8DFTa-04ROP7NwyD4xdy-_k-SMPwBzGz1PciE_mr06v3M5gL9qR2U_qqowv2aNxTs-uQ4_1xFmv5yNxL7eTDo_vKRGP3iRvD6eIWK_OFUaPzCHJLvcUsS-2TA6v5MtCD-BLGa__sMtP51VpL5amwo_V52IPtQGvL6OHQ6-y8VNP48WOD5bd8W-YuUXvq70ED9Tphq_HkFRvzwmNgEmK4c9-7hZP39nc79clV4_rCxRv7p6gr-ydHu-krLUvREogr9NLsa9XIXfvZBBFD-BFEI_kGRAvzj7MzsRTD0-rVM5v8dCjz6r1kQ-YX3NPmkJTj_mssA-nO3APLd4x745UJ8-uwkiv8XBjj_WDlU_Pf8qP9t5VT9NjcI-RgxSv_D0qb_COea_nxeoPc9eET5RPiY_NNhjvp515b9dtUU8pVUXP4JhDr41ZkQ_qy26PAB0AL7OqpC99F2BP4iVqr687CQ_gtcFP_uchz6ksv4-zTpXv01VjD-u0AS7_kl-PxR8Qz9eTto9AnmivaGKtb6KaJ4_i_YZP-Ysqz5Sees-3XqcvzwRCz-RCRO_PGuXP1pC-D6PV90-ja9hvk3tWz3zJVo_ANO-urrorrmvPLK6Ybmuu45qhD3c8T26LwUKgAlVhwCnBa-6X-bOtmf1sb1_V8o-6VuYP3sINj1bHYQ-j5abgAAlwr5g9Zy96xiRPu_dHr9--se_Ao9NP3SJFD-WeSQ-7HKjv-qN6b6Z2k8_Z5fYvra6rD5jLZq-iCggPsMI0z5HTQi_0EJsPmXItL77HD--byXZvvB7WT-RPZO-HLtPPntZIoBUMuIAcwokAWWb6YDKJjwBVsoaAMmwBYDz23sABvfsAOghHYAW0qy8Maquv9WzJz-Md-U-ofRHP8Lek75PsOY9kFqDvioJur9Uiwq_CTMYP1OQd7_t-Qo_RjqnPimyjb82CRC_nkSIP5flRj6KCcw-Zv2JPhNcW74Lijq_3WSKP9dzXD_WdVi_b3BvP4Kdpb1tZTm_Lg7OPggMmz1EmiA-qKmevjNK-L4ekqM-LggBP0Y8Gb-7SyU_43_qPYqfYD8Rwne_EZeOvQxooD8TLUi-W8olPjgD7z6ZTcY-StGVvkvDYT-G3EQ_U6aiv1HmUj--Kvo-c4IcP746FD8C7NE-soTOve0nmb6wQvq-YUdhvnVtkr_aeU0_j0VnP5tkAD8exgY_MTmWPrSyY7-NaPM-8Zcxv3K1XT9Gigk-mpyeP5K0GD-Bb1M9UEg1Pwbew76bm0e-aqh0P8w6FT_6CXw_jchrP7OapD25aZs91V0Cvy-DCz8ZbvW-1l4tP-nqiz8uHKU-b0d3PZ8QqT7Cs6U-FvN0v5gDVb-KUp2_rhJNv5USaT-I14u-i-QNvT9ooj-ciyq_6SB_gAn6Gz_HUS--rYsUPnrimL-8vek8uuPzPmNfYz9_Kei9uOxHPJaXgA681ySA1wscgci35oDy-OGHZ_YVgZjpgaw5VCqADY4FAX-xxYBEuvC-_gB1P2DhYj9zm20_nGfAv3lVjj9zpAq_v25MP0QLvr4Ws0e_rzEYvxd6uj4qFXs_TjCLP0m-jj2GM6w-qGx6PtcRdL-32wy_1Zqev2JAjT-iXeQ-MmF5P-2dlL2Jnsw9qzPpvn383D7O8oi_KAdgP9nftz3nTSo_lDXqPoNeXD87nhw_p8ZYPwXrhD8fOzK9tRgLv2-vyL6JX3i804gOP_ZAFr5ZWEQ--fsmPwV58jx6IIu-bUr7P7ZIHD-ffvs-dLzFvt4BGj-alO4-qQdFvybYjD-k7tU-kWhBP49AHL9F33E_nkPlPcfM1r7RCPg-69zgveiJIb_QueW_XcBcP18z4b4gsrw9p2TXPlylRz_xvLw9nUfuvqaHgr7iVI8-ctjmPupWG76fVjq--e3Kv4Xh8L5jna4_jV2sPmFm5D6KkL6-50OJPmShMj-w_Z68JQI2P2TTAz6P8q-_ACpJPxjwMj7wjEE_txuIPgpxGj8QVpk98s7bPt2FTb-msX8_auU1v5qyuwAiAuQ9en9Fv7ibFL0A5zu-ZZILvtbNvT8uelI_bissPL6PL79Hq2A_70Vbvlc9h79OvFg_L1rlPtvGnT-X0MQ-p5NBvzOyoj5nwXc_WfCOvuRGeb_cPwY_6D8Uv3UjaT6nu5c-1awFPubVlD_PjZI_z8X7PVNPu74RwNy-OBMovx7ERj8njUO_oqmCPwtNar675fE9hCO7v4IdCr8egaO-7UE2P2_CXIAmKwCA_tuqgGYxUKewZPAx76i6r9bvNx181D-BLB1pgFn5F4Am2-w-r3WWO4ZVjb_8VTk_mkMyP5hn8z5UVvs-D8KCvSCaTz4hKG6-7aebgIKjZYA5oi8BHbcWgO_7aABv3lIAYbWMgGCxNgAFP12As1T1ABQYIyVS-A--1QeOv7fXSL4Kij0_lz-Gv1_7XTs_554-J3ANP4VZzD8V5gG9f-eTPzkPSL74mRK_m8Q8PU_Zkz1j24Y9UxqIvcVlKD5AVHc_ADp-NV5UmS36fkKlA4zFoSpNGwCj08exf2ilstrTczuxa-Ktc8Pxsb5BkL_xEKE-yukgrZsk9T4Lad0-oZUfP-jiiL6Mpcc-lqCXP9dljT4bxNY_HYkUv53dGr8mfNI_-z0JP-FglL-tHpg--zRvvsfZmT8BmEw_sZkIvYOFFD-g_hI_9ughP2ipmj_H1D0_uDs3Pz3mhD-f8XW9_cgGP37YNK-6tXC_LFbCO7RrJzIThRs_qryaOl9x5by45YI-r9lFv6_wi77BkIg-gCCJPo9-fj7HiJm_GW15vwSAk7-Hr1O-pD_wvdwUDTg2x9g_-x4HPqfUAj8tXJw_eUfWvg3fMT8wR3I-oHlpPiAHbr-ajco-nW8nP6Px7b4nhoU_X6jYO9TbTr-g1V2-z5CXv6GcpL-YYSO_L2J5vRlrP7-zkpE-L7QDPr6A5j_OOp6__9fjPj2VVr_FR4m_nxlYv4WOaj_bdUE-QktnPnvRr4DZnW-9A1cDP07khDwP3gq8ZQnSvsUlm78csAq_Uk1JvifMVb_voi-_ySiMvZpwLT1Sghe_opfIvy4pPD-Vzac-g28vP61-Bj0yqw8AjHy6AKY0lQC27iYBvGEUARdBKIAC8CYBAqRhANstAABD4jIASc8NPn56jz87JgY_NVABK4JrR78w0yg_f0OWPnfZPj0lDJw-ln3rPgg4pD3X3rg-GFgBv5ga3z3mCIW-b-PQvZCvzr7RSYK_lIMEP1qmcj7d7li_s6jDPv_XEz1gPBk_v2ooPpKKMj9Czau-DQqBP6dxIDxYbZU-mTGOv0u0Oz9zTpU-Edd2vlGYgj2msyS_neoRv_XcKb8bxW--n7iys_j4b7-Chi4_VOn3vZKRZL54DRO_H1eOPtGLXD4yEA8_k-ZUvo2MDT7YfgU_dXwLPvP2Ez_Jny2_kurKPnNooz_69bu-8v22P3KZ3z4_7EI_O13ZvmnDyT0Z8_A-jFecv3lKFr62LXi_7CxtPz9pJT8xqxs_UINSP9JDgjr-UOK-pCmJPwQBjb--i4E_2FQEP1BDID9aMIa_yIYRvt2LlT7WHy--3wy4vaV8WT0K3DE_gNGWP5nGRD79Fdm-va17P4gUsD7tvi8_K_JsvxvbeT9Qcwi-XLQaP3VqK74f26o9bhAUvzRiq783BSE_r9ogP5d5A7_d8Ts_qbuJP7UCuz4Mh28_U_2Uv7yaXz5prF8_SZ6mPqhdIT91waQ_HlKHPwXsAD9R84GAnqeKP1ReDz8yD1M_ngzQP5tFYr8xMna9ZyvRv6GEBD5zxka_2MJyP8YLsT679QeBCtrXvnl01j50n9U-lsIqP7VNEb_rMsw-HtEHHuaIHr1DuL8-qkcIvxvaLz6ex_S9ULG-vPwQj72EVpi_SNp5PyhOfb97QUA_4euSvos-R79I40a__TZav0CcYj4IVME_7HOIIdibHgE1dGWAXn-7J8pRLgHPXYwAGGtAKKzbgowV8v0K3vAwASiD2j8io3g-OD8Pv9Av9D7m2xU_6rlRvxjCLr90xDe-gO98PfrZzz0RJxC-OvGIvfdxSL-rx0a_hQxiP1gQfj_I6Xw9XHTZP1K0nL4bhRS_ED0XP4KuSz9bkLQ-WBoDvz5ZA75W0Bo9sSdQv_I5m7ylwRs_KcEwv30M8b5Qxkg-K-4VPyutAz9tFeC-_78-Pmy95j7X1Zk_uQAaP1MrK7-xLj2_dVg7Pwq4WL-STEQ-Bnm6PYaLiL_VE4q_x0XEPg1FtL4V2Sg_CQDcvLKkEb937b894smLv_x3EL9NRYU_tFRmPwxMYL8skGc_IbYPAeNs5ACwXw-AmZwlAZlJJYFFeegAsLPZgH5XFQFi5ReAjez4AKkEr4C8-6S_Vb17v4ZL0T8rzhM-IrN6vmocXL9WoIw-uqtYPxuzAj9szZa-_vUeP94Zrj5NSi6_JsuWPgeVLT-YoJU9txCSv0V2Z79i3uA9ktR2P9_4xb2aazg_lyM6PlUPDT__vag_yCcZv5zXLD67dpG_Kcb3vrWqor7mjEYo7t3tPnYo3j48Dno_JgmoPYctc7-WhKg_9JvwPbc3cb9SHoY-pCROPyjXZz5iSNA9lLfDPnFsRD9zYkm-gvR6PmGKbD3WzKG-sOn1PgeTEj_60A4_erGhPwQqKb95s4A8gaclP_-DgL_KdIs9gwizPuVjMT_iJM-8Py1UvxwXSD9xUXy_rsO1v1XuNT5vLnY_bDsRPpyBlT5UrJy89hQlPr_09b9hj1g88jDjvoqawz9Kt0c_xg22vqGA1L1r0K-_Pi9Pvu8evT4rEi6-6jqUvgGuIz93TGQ_DiAJvhwSgr80gQM-raUIv6AqiC5E-6Y-FuVfP3I8Fr_oPYYA8bcovwTg9j7UcnO_FZaKP5oiGbwB_ts-uEAqvkyVUT8HjnO_eoCvP2jtcL7xeLE9rpTEP7fJ_j5z7dE-5sU5vMiLBr9p9_u9MV2Avd2vAr_45rG-lVIBv-xdQYBNf9Q-XScsPwtuQz9DvcW9Xp41PzyFF70jkzG_EsXfPgQFK79CI-O-LqnjPnRQSr6GllC_-0iavyKJ3j63Q7E-37Sovj0n0L7UUoc9s7LgP6j8ZT-tmC6-Kd_KPlLxmb5GVZs9PZClPm2LXb70zn-_ZznEABKAnT4JyAu---yAv16FaD9h0h4_5oSfIp60iD7ZE8S-4JyWvzc54z8mOQs_hKSpPJI38r0u2py8Yfh5PZP0hD5134A_XqFpPs369D1Yjpu_IjAIP3CnwL-_xV8_bH0Bv0DART-wRLY-90XkPoT4Xr8puqW_GCNdv_rc6z4hvVG_Urx3ADm3s72mBzW-nHa0Pox2Lz_h5o89bs5Rvr3tT794a8y_qfCXv5Hsij9MRbu_QRvqvgeRwr9zT-S-vP5hPRSccb_V-Gi_Ccz9Ps9rqr9hpv0-OAeVv_1ihYDEMcmALmKRAKaiGwBSwwEBcMZDADMmKwEW1z6AtcFjgMIyUYBAm50_87uIvjPeJD-LKZM_oWaVvq2RT7_RpP098ZjauxOycT7-Mj-_L9A5P9QdZj9NYIM_c_wqP3YuAz8XFLo-nHr7vmDqVb8JQnw-e0H7PhAhBj5qJuk9x3DxPt_zMz9Ev1Y_UYIuv_0SgL8v-04-hoFmADaPfD8_Rz0_F6SfPPbdOzzEEcYcBtsGP7PmhT7xI3g-QQCTP6-dRT9Yymc-M07jgJPA-YBFkniAFwECAKhDpwD4aXkAnZH0gBxduYC2mxSA5ADhAKyImz0ldjw-slKXP20goz4rijk_psYHPw5YEz-QXpy_obhAvoqwg75ehNK_DolVP9AWv79xqxoOL9Gov6FRhb3NKmW-45AQP8v4nD6CRR2_xtAlP2l2VL1n_li__apGv1XOr74X-Gg_m46svMlS5ICe_e8A3kAFvxo5OD3ubJQ_3Ti3vrRbmT7iBUI_0x0vPxdlND1yiEu_39Jvvpid4D38kOMA3hVMgF4bVIAdfRcBrFkKAEmYr4CIzQgAKLALAbbPWQAAqQ2BZCbovidyRr_k38E-ZrybPmwL1j6fJowhEUVzP_r1r76dEiO_ay3BPrsIiz9jD-0-NUOCPvCB0T_ggBm_UoEqP9TZQj44kwo8W_OuPsYYeT5vk42-ALerv9tbrD492o-_JQUDPh-SxzyzRFe_ohuXvZlyhj9YXwU9HfkdP7Pnr7679u2-8bxCP8XzVD94Oxo_CJpZP8GgGD_TpNc-BQmjvsdNFT_ZNgk_UcSXP0mFoL4oWkY_qqTZvsgQ5Lox3-y_47x1PkV3G78=',\n  weights3:\n    'w8mivx24Qb_1lfY-x7snPhfhJD0wJVm-n5WWv-cBjL9Bsb6-upd-P1ST9z7Q1Xg_NCngvrtkV78AbEO_oEUOP0YW4b6jn3E-o4vvvdrMRD4y7-Q-IF8UvZKplb4AEas9CwLAPjiOAT_JlK2_KumLP6HqhL3jo_i-dNUAACRjWYDSKzgB_hMcAabRtYDQF2iAOKbdgPki04DQqmIALIGGgJcIkj76_X6_g1iQP_Lgrz0kll-_etlhgBkgpT_3XRO_duGAPwlOiT90QGc_yNYzPqEnWL86wBY_ZVUvAUZJor_20mm-tL-DP1vw0754O4q9PlGAAOoW2QDILA4BCeMugUTjJoCSxLUBfEnCgJT1LYB4vX0AdJjsgEDmvgF0RicALOgrgV9J2AA8FwkBb3UEAM158IASeh0BraSqgL67NIDOLGcACoIyAEqcIAH0b0UAiQwcgIdBfYAqXA4BWlvngA2yRQCS95yAQ3UDPlAM1z5mfEw_GR2kvuksez8ywIk_J_5KP--FgL_8_A27eYZ0v26tzL6hKZc_B08PQJwmAL_WB0s_iC2jPlIhUj5VZN8_wAQNPTgd4j6AoBqBiF1EgJqqJ4C2yhsB_uA1AHfbn4DlkaYAKO2OACisWYAk5MuAI_OZPoviir9zO5a_mPTRv3R83j4xnbu-NnJ4P1Qviz9z4N08GeC2Pu5w-wDonBMA0E4sAfGeXAAIPoeACGMygYyETIA-nvCAoddmgMSiVAAkh8q-UhQ6PzWiEr-I5Rg_ggf-Puimhb-zpLW-8cmKPyq1-j4ZoX4-Pok2gS8uK747L4M_qkYyPooo-j6qVACAO_skvtqJkr35KI0-a7C3vUK2Nz9iKPk8BMC2PyF2K7_I-hG_IxJXP7Lkz7583Ra-L8w9PPUk8D4Qex2Ahg4PgCDChoCmvguBBsVNgDzEEoGS1xGAsNQDAV_8pgDA7qUAlp-_PqZerj6O4b2_yFtjPmr3tD_00hk9LHzoPqL8CL9e7hc_ufKqvqDR8j4WUcC85YhnPycvkT_FO3yA5dTbPZd7Sb8es2k_03OpP75-sL8i-FQ--CoCv2RVCT9xkj6_KeVqP04-TT3Xwbe_RKXZPh3umr6RpH6_81yVPnfA0r9q4-k-4UwTP3X9A0Bj2gw_jD2Iv-Vcv77UzzU_ectbP1uiBwCUBcSAbYx5ANvIKwBgpOCAhnMfgIynBAH1DR4B6kpuAK76BIBmgmg_EPZnPzKZpr8vY74_10XTPgc4gb_jHzg_SeJuv-xaaD6I1Om9505lgFyuFYHv-ZmAx51zgEam5gBwqR2BUGpHAKtxMgCOom4Ah4-XgAg5yL_mCG27__W7vhdCfz-Wzec-izphPAMhUr4Uf42_TrdGOMv2tL_lzPM-fqaBP_nhh7-Ewv4-Xj5iPLIC-z7Eyts6SEFEPxJDLL4Uris9-F_CAC4qU4DxSF0AFBsOgDZ0RAAy2BqBRLMiASyUNoAY-sWAqc0hAHCzj71IUYM_xJKmgOs4vT6dUxM_0UfRvE50wj6ZraQ_PNd4P5L-WD-GTgy_RA2nvzSMELsGWD6_p4FpP20OVD8WHAxAPp4BP4e9YT9IJZM_Q7KEgOKhMYBc3wKAtTmtACAFJoHHS6cA6MHngMLMdoDMRxEAuqMGAd_yLwDI0CYBNooLAXLDJQCAth0BGtkIgZiXjoCedbcAKehbgOyb9gDT1eU-QTCNOdAUQb_lc4k_iSeZPJUAlj-gcby8fjRQv5H9qb9Fibq_MCoZv_KVxb--6pc_eN99v3iL1b1BbYW_GSmTPsoCJj9VVWw_Y_eKv4f0BwGyxQKBq0m_gDzgGwE8ORqAqB0sgK5EMoGkXxIBQ2AAgOYnaoCUayCAOQu-AERVP4ADYikAebfhANYm1gAH_RkBcRoHAfW0FYHCwD4ACs9kvex5kT-B5-Y-mUcOvYzfaZlQrp4_5mvmPiz1Xz-fR_S-9Tq8vpMUO799Vrm_ExBRPQtbH7-tBf-9ZiO9PpyNnT7fy1c_oWmKP45yCD9aPjU_0cjAvr11ST8PP_u-G7DJv3YHiT2hgGkAUe0evpGQoT14Bzi-sSVev65O4j5-OhG-ukaVvinhjb58XK-_TKw0PhVJY79z_py_mip9P8zqGb9tfvQ-fn8kvtFbFj_-drQ-Mms0v1noLD5zWlI_0kjVvQKPrD4LB3c_w6yQP_9ok7-zCgw-c5q8v7-g2r7Og4-_HjBIPnAws71a6_O_8i4TgJTzd4DAnb6AlKK8ALF5XYCvlCeAYfKkgLoENgAdhSkA9vMiAVStuwANNc0AYNUGAToxFgANUbqALXOfgHqgvoCfSsIAU3QTgILH6YBlx4I_il1_v4VsB8Bo9Wa9I0tfP4TNXD8_5qM_4rvivoqglL94AJc_VEAHgQgMHwBN9HIAY23egNieLYDdmzaBWBc8gXmZNABujg2BtI0mATtyfYBxEBYAfyhlgOssqoCbFKyA-_upAEJC8YBe91cAzghogByy5ICqKgmBuy8UAa0UVAACIRcBTRW-gKhgJQCjEK2A9rnHgBjReQA2NO-AF8jsAEB0P4HkNSyB0UrQgF8_KADI7KgA6gZPAEUL2QAG1RuB14UcgRL5pwfk36k_nlk6PxK2uz_bzA2-_4HovQ7Ehr9k30I-8kvevkQuQDzrKKU_KxodO1Ge2r4PYhU_abKov2--Fz-feI4_m3HEPZxBOgFpXa4-GnhGgO9SGIGwiLKAPtMuAfDBBYEanMSAYl1cAFw7BgAOTegAiADagKmnb7961R-_8SrqPyugCL94qY4_Jd1OvuLeSr89J4e_ubrEPbftnr9Smlw-_ZOOP9Ueg74npbs_sAU1voQpLz9CI2s_2mFTv7KxZb92OpGAQhyJvgTBIz7aTCQ-EELiKPT9OL1eMrC-r6IsP0uPgL2I5Ym-F9z_vAR8JoCgAlyAww9lADh2FAEATneAkyW5AGobMgDRHmuAkxwbAFBIbACwZyu_lFFdP57Qob-6tj48bUBXP83MLL-fmw4_EIXBgI8trz87yZA_SE7ZPBiIST69LO4-CeMuv4c_fD-9lmA_ERanv0J5cT_yisG_LRfHvzF_aT_P51M_98QmPFQBYD8w0IY_tHMWPnQ_YT9Hc0e8f0Rkv4YfFb_Xe0C9vKmOP12QI71kdgo_egCBP5Bxs782y8k_Lp1DP6uFn79ajh0_4l3KPHM88j7ceaQ-fs5bP_g67b_r6Fu_gu0xv9PNsD6NE00-Lo5tgAHK6gAXPh2B49U-gPvZ24BKQtEAqho4gGhVL4FAK58AL0MTAbuzPoD9GGe_QIY-Pqtbaj9YGcs_kyTKPVNyYD_G61U-Vyrxv4Kj-j7DHgU-AhDvPncKwD781YE_OKs_v1yujz-6HaO_wfSFv5sI_j3P_xA9IJdIPw==',\n  weights4:\n    'gu1Wv-fgtb-3SkE_xEOnv-A4ij_-05u_cvM3P1sgzj9wh5a-rS20v_zLiL63nA29A4SBP8u42j8OYi-_UPH0vzHVor6TlNg_vLqUPwAE6T6Zvr6_XViNv5uOr79ybXQ_ckpHvwCXvL-zNZ6_PKQZP39AeL-zmEa-nToNvZkjgD6BrmE-mWUyPz-AxT_9hYG_cmGfPvaRvD_FXd4_qbnvv8fSK7-WoLC-gd2sP82M1j83SNe_-2AwP39FXb-siUy_iVYBgSn2OT8QGN6-_4KlvwiPNL_BzZ6_8wbEP6hC7L9Ccyo_816WP_gTOT-N5D2_NHSeP1tKjL7EfIy_iz8NPtAhXT9x5p2_5pPlvn2e1r-NJKC_L4cCPw==',\n  biases1:\n    'HaMhPSUacD4uCSc-waqzPlqCDT8oeeS9s8JDvMixgD9G0HU99_TOPmoqOT84tao9RdMQP5MqVTsb8UY-2EtmPnX_kT5d104_rO4BP0MByL7C_rs9ou2hPg6RqD61UzA_V7EKvpCxoz1787A-PqvcPY_jBD8cQK4-njzbPcUuzz3Y-as-Y6Y1O7-htj4EdBM-OqgtPisOBj9ZUis_WFRmPq23jj5ZKG4-z0FTPnM8Hb2pfqw9dTMTPjuQOj6zPoc_CfUDP4of6D72AXo-AXZlPjhNHz85-mI-YZ4MP0Ydbj8Egra-dB8mvvPa_D3H5zk_gpZaO_ckIT_HUPk-cHj9uzlXR7sYwTe-ds8QPyiXLT6p3R0_P-DxPmGGc75olzs8Mg9cvTF7lD7v01k8-GPWPtNmG7v3Ars-1C4hPxUtrz753i4_zFtWPRq2Oj77BYg-Chj7PrV6wjsYsRA_QANTPrLnsj4Nk0w7ebDDPq5J8D5bTds-5j5vPmJVOz8693c9v3rNPjZAJT__FtQ-N45XP5DoDT6aKw8901n7Pnr-o714t_e7bq-GP1qwzj5BkRM-WugkPr2vhj6PCPu9AMu7POzDgj4gWtk-07c_Pmb6Rz5aOpY8DTi5PntEtT4pMAI8PPuwPgrqED1Y3g4_iw3JPpeWrj7X3m6-bvjhPpD3k74T3wc_gBtoP4CGHD7JUoI-eAODO4Ko_7yYG6c-vQibPjt0Hz3nPwo_6lYPPRcYhL7UGPA-FZArPBI93D15MXg-iOYxPtoUeD2vp0M_hpGPPcM69j0dFfQ-ZXrtPUYKIL2Kml86dmvsOzwZ8z7fkWA-HASOPujnjj5_Wim8w0LevOWA_j22SNY-y1JrP5sbSj9-4wc_gCeqPqN3yz5EvUA-jnw_Pmaf0j5XGLO-XaJIvtiqqD4dbuQ-k71jvr1AuD7D_gm83xsXPqDoKj_Jr_I-qGiHPQKXwD49e7c9wpJiPiIk3D6-lY4-Oj0Evt7yEj-kEUA-wH0lPxv4bj4GXRU_V5nRPdTrkrmUuZ4-sm7JPagP_b2BV_I-f63MPnp8sL6rCPA-8lKjvC4ISr5j8L66i7W9PtxOSj0Ztfs-GHCdPkHXCj4bmgw_KMIyP8U1lz4wJLU-zwjSPovdX71is7M-f9L1vVAWrzywRM-7zugNPgj3GD-LPUQ-GzR3Ppz5Kz-ojwa9Ttm5ukuVqzzaCuG-Bc0lP1NCsD5PDr8-PQbsu0x4RD5eP3u8d41VPshsFT8WXe8-v4-sPuQyPTxAt8Y-hZ9ZP3JegLnbLLw-lBYFP6-IjL3Ff5Q9OdC2PfUMPD2dfSe8lPazPvr40D7KB8e9eNKKPh2ytL2LJh29m5QRvA==',\n  biases2:\n    '_Q9ZPZ8GQr68UR8_IbJ2Pgn0QD3ybP--wAe3Pr8LPDyG5vE-JygwPw57CD6dyKc-PvBtO3AsXz5wQkM-8ta2PVgSHj0EG8M9oeQHvYAlIj6P-K4-rgdMvv3T1r78Nwg_WJPNvmNOPrs7Z1M-YfB_vmU6hD31XyO-gwFPvuIWVT7RnG4-LEsBu6rKJL1BSMs9kpOqPvgulb3bshQ_GH-EPly16j1bVQ-9dRfdPb4vpj1SrMC-VjyfPi5kDT_e3bE-YF5Pvnm-DD4nKps-r736PiXlib3QgRq9CR5NvrL8Y72raiG-F2VEvtp2jj0rrZ0-eedGPif4LT1RsNM-yPdjPuEGyj6QCbU9cdj3PvQ_sz4KySa8dQtJPxBaHz_t12O-aAIuvrLxT73b_Is-ec-IPggFAj6QyTM-V3DdPh-MzT3kryY_cRhcvs7eCr5QbPE-R0ElvfyJYr3bJmI-Mug7P5pC1D2vgVw-OWerPpKaaDxWOMU-mbuzPsbRDj4k0oG-KY4bPkFV3L3lR1W95YmoPr6wm74Sw849kR5XPny7ZD3-VIc9YRnrvkyBKT687LK7zJ1cPn5RPT58iKM-j7JUP35YBbtj2xM-5iTiPpbGlD7Fg98-sfPwOVsRtj7kJw8_8ROJPvw9gr1riY-9352XvikwIT6I4Ys-PaVDP-hO-z4=',\n  biases3:\n    'wYidPj2HAT5ogRE_KJy5vPa4CT-Bw5o9fDn-Oqwr6bs28Pe7EmigvVmRFT9nnR-8PpSnPjyxt7oQA3Q9QS4dPbM4HD_csA67n_nVPdZlV734fBc-0ZiGPaXTm7xPqHM-WuOeOvb70T5tSAo9RrB3O_T07z6yJgU-NlsJvfWmYTxowio-sYzsPuR6aDuZaA46p4vsvpKSMz-GgnO9G2TCPpUkeL4rqak-TXy1O3wymbz6qIQ-Ik91Ow-Qyru3IQs7masouyAwmz4H_DQ-F-jruzb_6z6gV-E-IElNPfpB5Tu0ASk-0nQZPmGwbT1yjG4-sLBAPhSGRDzcpq4-dqjMPg==',\n  biases4: 'W7ejPKPttb42hDE9X2Z1PoV_Bb_msGG96ga8vgCSVz8b6KO-1rKUPg==',\n};\n"],"mappings":";AACA,SAAS,QAAQ,UAAU,gBAAgB;AAC3C,SAAS,WAAW,SAAS,cAA2D;AACxF,YAAY,WAAW;;;ACIhB,IAAM,gBAA0C;AAAA,EACrD,GAAG;AAAA,IACD;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF;AAAA,EACA,GAAG;AAAA,IACD;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF;AAAA,EACA,GAAG;AAAA,IACD;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF;AAAA,EACA,GAAG;AAAA,IACD;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF;AAAA,EACA,GAAG;AAAA,IACD;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF;AAAA,EACA,GAAG;AAAA,IACD;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF;AAAA,EACA,GAAG;AAAA,IACD;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF;AAAA,EACA,GAAG;AAAA,IACD;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF;AAAA,EACA,GAAG;AAAA,IACD;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF;AAAA,EACA,GAAG;AAAA,IACD;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF;AACF;;;AC7DO,IAAM,iBAAiB;AAAA,EAC5B,UACE;AAAA,EACF,UACE;AAAA,EACF,UACE;AAAA,EACF,UACE;AAAA,EACF,SACE;AAAA,EACF,SACE;AAAA,EACF,SACE;AAAA,EACF,SAAS;AACX;;;AF87BM,cAWE,YAXF;AAz7BN,IAAM,cAAc,CAAC,KAAK,KAAK,KAAK,IAAI,EAAE;AAC1C,IAAM,aAA6C;AAAA,EACjD,CAAC,IAAI,EAAE;AAAA,EACP,CAAC,IAAI,EAAE;AAAA,EACP,CAAC,IAAI,CAAC;AAAA,EACN,CAAC,GAAG,CAAC;AAAA,EACL,CAAC,IAAI,CAAC;AACR;AACA,IAAM,UAAU;AAChB,IAAM,OAAO;AACb,IAAM,aAAa;AACnB,IAAM,kBAAkB;AACxB,IAAM,gBAAgB;AACtB,IAAM,cAAc;AACpB,IAAM,eAAe;AACrB,IAAM,UAAU;AAKhB,IAAM,SAAqB;AAAA,EACzB,CAAC,SAAS,SAAS,SAAS,SAAS,SAAS,SAAS,OAAO;AAAA,EAC9D,CAAC,SAAS,SAAS,SAAS,SAAS,SAAS,SAAS,OAAO;AAAA,EAC9D,CAAC,SAAS,SAAS,SAAS,SAAS,SAAS,SAAS,OAAO;AAAA,EAC9D,CAAC,SAAS,SAAS,SAAS,SAAS,SAAS,SAAS,OAAO;AAAA,EAC9D,CAAC,SAAS,SAAS,SAAS,SAAS,SAAS,SAAS,OAAO;AAAA,EAC9D,CAAC,SAAS,SAAS,SAAS,SAAS,SAAS,SAAS,OAAO;AAAA,EAC9D,CAAC,SAAS,SAAS,SAAS,SAAS,SAAS,SAAS,OAAO;AAAA,EAC9D,CAAC,SAAS,SAAS,SAAS,SAAS,SAAS,SAAS,OAAO;AAAA,EAC9D,CAAC,SAAS,SAAS,SAAS,SAAS,SAAS,SAAS,OAAO;AAAA,EAC9D,CAAC,SAAS,SAAS,SAAS,SAAS,SAAS,SAAS,OAAO;AAChE;AAEA,IAAM,UAAU;AAAA,EACd,MAAM,CAAC,OAAO,OAAO,GAAG;AAAA,EACxB,KAAK,CAAC,OAAO,MAAM,KAAK;AAAA,EACxB,OAAO,CAAC,OAAO,OAAO,IAAI;AAAA,EAC1B,OAAO,CAAC,MAAM,MAAM,IAAI;AAAA,EACxB,OAAO,CAAC,MAAM,GAAK,GAAG;AACxB;AAKA,IAAM,WACJ;AAIF,SAAS,WAAW,MAAc;AAChC,MAAI,IAAI,SAAS;AACjB,SAAO,MAAM;AACX,SAAK;AACL,QAAK,IAAI,aAAc;AACvB,QAAI,IAAI,KAAK,KAAK,IAAK,MAAM,IAAK,IAAI,CAAC;AACvC,QAAK,IAAI,KAAK,KAAK,IAAK,MAAM,GAAI,KAAK,CAAC,IAAK;AAC7C,aAAS,IAAK,MAAM,QAAS,KAAK;AAAA,EACpC;AACF;AAwBA,SAAS,aAAa,MAAc,aAA6C;AAC/E,QAAM,MAAM,WAAW,IAAI;AAC3B,QAAM,YAAY,YAAY,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC;AACvD,QAAM,YAAY,IAAI,aAAa,YAAY,CAAC;AAChD,QAAM,aAAuB,CAAC;AAC9B,MAAI,OAAO;AAEX,cAAY,QAAQ,CAAC,MAAM,OAAO;AAChC,eAAW,KAAK,IAAI;AACpB,UAAM,CAAC,MAAM,IAAI,IAAI,WAAW,EAAE;AAClC,aAAS,IAAI,GAAG,IAAI,MAAM,KAAK;AAC7B,YAAM,IAAI,KAAK,MAAM,IAAI,IAAI;AAC7B,YAAM,IAAI,IAAI;AACd,YAAM,IAAI,OAAO;AACjB,UAAI,gBAAgB,YAAY;AAE9B,kBAAU,CAAC,KAAK,KAAK,OAAO,KAAK,KAAK;AACtC,kBAAU,IAAI,CAAC,IAAI,KAAK;AACxB,kBAAU,IAAI,CAAC,KAAK,KAAK,OAAO,KAAK,KAAK;AAAA,MAC5C,OAAO;AACL,kBAAU,CAAC,IAAI,KAAK;AACpB,kBAAU,IAAI,CAAC,KAAK,KAAK,OAAO,KAAK,KAAK;AAC1C,kBAAU,IAAI,CAAC,KAAK,KAAK,OAAO,KAAK,KAAK;AAAA,MAC5C;AACA;AAAA,IACF;AAAA,EACF,CAAC;AAKD,QAAM,QAAQ,CAAC,GAAG,GAAG,IAAI,IAAI,CAAC;AAC9B,QAAM,UAA0C,CAAC;AACjD,WAAS,IAAI,GAAG,IAAI,WAAW,IAAK,SAAQ,KAAK,CAAC,CAAC;AACnD,WAAS,KAAK,GAAG,KAAK,YAAY,QAAQ,MAAM;AAC9C,UAAM,UAAU,WAAW,KAAK,CAAC;AACjC,UAAM,UAAU,WAAW,EAAE;AAC7B,aAAS,IAAI,GAAG,IAAI,YAAY,EAAE,GAAI,KAAK;AACzC,YAAM,IAAI,MAAM,EAAE;AAClB,eAAS,IAAI,GAAG,IAAI,GAAG,KAAK;AAC1B,cAAM,MAAM,UAAU,KAAK,MAAM,IAAI,IAAI,YAAY,KAAK,CAAC,CAAE;AAC7D,cAAM,aAAa,IAAI,IAAI;AAG3B,cAAM,IAAI,aAAa,OAAO,MAAM,IAAI,IAAI,EAAE,OAAO,OAAO,IAAI;AAChE,gBAAQ,UAAU,CAAC,EAAG,KAAK,EAAE,KAAK,EAAE,CAAC;AAAA,MACvC;AAAA,IACF;AAAA,EACF;AAGA,QAAM,MAAwC,CAAC;AAC/C,WAAS,IAAI,GAAG,IAAI,WAAW,IAAK,KAAI,KAAK,EAAE,KAAK,CAAC,GAAG,GAAG,CAAC,EAAE,CAAC;AAC/D,MAAI,YAAY;AAChB,WAAS,IAAI,GAAG,IAAI,WAAW,IAAK,cAAa,QAAQ,CAAC,EAAG;AAC7D,QAAM,gBAAgB,IAAI,aAAa,YAAY,IAAI,CAAC;AACxD,QAAM,cAAc,IAAI,aAAa,YAAY,CAAC;AAClD,MAAI,KAAK;AACT,WAAS,IAAI,GAAG,IAAI,WAAW,KAAK;AAClC,eAAW,KAAK,QAAQ,CAAC,GAAI;AAC3B,UAAI,EAAE,GAAG,EAAG,IAAI,KAAK,CAAC;AACtB,UAAI,EAAE,GAAG,EAAG,EAAE,KAAK,EAAE,CAAC;AACtB,oBAAc,KAAK,CAAC,IAAI,UAAU,EAAE,MAAM,CAAC;AAC3C,oBAAc,KAAK,IAAI,CAAC,IAAI,UAAU,EAAE,MAAM,IAAI,CAAC;AACnD,oBAAc,KAAK,IAAI,CAAC,IAAI,UAAU,EAAE,MAAM,IAAI,CAAC;AACnD,oBAAc,KAAK,IAAI,CAAC,IAAI,UAAU,IAAI,CAAC;AAC3C,oBAAc,KAAK,IAAI,CAAC,IAAI,UAAU,IAAI,IAAI,CAAC;AAC/C,oBAAc,KAAK,IAAI,CAAC,IAAI,UAAU,IAAI,IAAI,CAAC;AAC/C,YAAM,IAAI,KAAK,IAAI,GAAG,KAAK,IAAI,EAAE,CAAC,IAAI,GAAG;AACzC,kBAAY,KAAK,CAAC,IAAI;AACtB,kBAAY,KAAK,IAAI,CAAC,IAAI;AAC1B;AAAA,IACF;AAAA,EACF;AAEA,SAAO;AAAA,IACL;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF;AACF;AAeO,IAAM,UAAN,MAAc;AAAA,EAsBnB,YACE,OAAO,MACP,cAAqC,cACrC,QAAsB,WAAW,GACjC,SAA2B,CAAC,GAC5B,UAAoD,cAAc,GAClE;AApBF,SAAQ,iBAAiB;AAEzB,SAAQ,MAAM,WAAW,QAAQ;AACjC,SAAQ,SAA2B,CAAC;AACpC,SAAQ,OAAO;AACf,SAAQ,aAAa;AACrB,SAAQ,cAAc;AACtB;AAAA,SAAQ,cAAc;AACtB,SAAQ,gBAAgB;AACxB,SAAQ,UAAU;AAYhB,SAAK,MAAM,aAAa,MAAM,WAAW;AACzC,kBAAc,KAAK,KAAK,OAAO;AAC/B,SAAK,QAAQ;AACb,SAAK,SAAS;AACd,SAAK,UAAU;AACf,SAAK,SAAS,IAAI,aAAa,KAAK,IAAI,SAAS;AACjD,SAAK,MAAM,IAAI,aAAa,KAAK,IAAI,SAAS;AAC9C,SAAK,WAAW,IAAI,aAAa,KAAK,IAAI,SAAS;AACnD,SAAK,WAAW,IAAI,aAAa,aAAa,CAAC;AAC/C,SAAK,YAAY,IAAI,aAAa,UAAU;AAC5C,aAAS,IAAI,GAAG,IAAI,YAAY,IAAK,MAAK,OAAO,KAAK,IAAI;AAAA,EAC5D;AAAA,EAEA,IAAI,UAAU;AACZ,WAAO,KAAK;AAAA,EACd;AAAA;AAAA,EAGA,KAAK,IAAY,WAAmB;AAClC,SAAK,KAAK,IAAI,IAAI,IAAI;AACtB,SAAK,QAAQ;AAMb,UAAM,QAAQ,KAAK,IAAI,MAAM,EAAE;AAC/B,UAAM,UAAU,KAAK,IAAI,KAAK,EAAE;AAChC,UAAM,WAAW,KAAK,IAAI,KAAK,EAAE;AACjC,UAAM,QAAQ,KAAK,IAAI,WAAW,CAAC;AACnC,UAAM,WAAW,KAAK,IAAI,WAAW,YAAY,SAAS,CAAC;AAC3D,aAAS,IAAI,GAAG,IAAI,KAAK,IAAI,QAAQ,KAAK;AACxC,WAAK,IAAI,CAAC,IAAI,KAAK,IAAI,CAAC,KAAM,KAAK,WAAW,WAAW,IAAI,QAAQ,UAAU;AAAA,IACjF;AAGA,QAAI,KAAK,QAAQ,KAAK,YAAY;AAChC,WAAK,cAAc,KAAK,MAAM,KAAK,IAAI,IAAI,OAAO;AAClD,YAAM,WAAW,KAAK,MAAM,KAAK,WAAW;AAC5C,WAAK,gBAAgB,KAAK,MAAM,KAAK,IAAI,IAAI,SAAS,MAAM;AAC5D,WAAK,cAAc,KAAK,OAAO;AAC/B,WAAK,aAAa,KAAK,OAAO;AAG9B,YAAM,EAAE,WAAW,OAAO,IAAI;AAAA,QAC5B,KAAK;AAAA,QACL,KAAK;AAAA,QACL,KAAK,QAAQ,KAAK,aAAa,KAAK,aAAa;AAAA,MACnD;AACA,WAAK,iBAAiB;AACtB,WAAK,SAAS;AAAA,IAChB;AAGA,UAAM,OAAO,KAAK,IAAI,MAAM,EAAE;AAC9B,QAAI,KAAK,OAAO,KAAK,aAAa;AAChC,eAAS,IAAI,GAAG,IAAI,KAAK,OAAO,QAAQ,IAAK,MAAK,OAAO,CAAC,IAAI,KAAK,OAAO,CAAC,IAAK;AAAA,IAClF;AAIA,UAAM,WAAW,KAAK,OAAO,KAAK;AAClC,UAAM,QAAQ,WAAW,cAAc,gBAAgB;AAEvD,SAAK,WAAW,OAAO;AACvB,WAAO,KAAK,WAAW,GAAG;AACxB,WAAK,WAAW;AAChB,WAAK,OAAO,QAAQ;AAAA,IACtB;AAGA,aAAS,IAAI,GAAG,IAAI,YAAY,KAAK;AACnC,YAAM,MAAM,KAAK,OAAO,CAAC;AACzB,UAAI,CAAC,IAAK;AACV,UAAI,KAAK,IAAI,OAAO;AACpB,YAAM,IAAI,KAAK,IAAI;AACnB,YAAM,KAAK,IAAI,MAAM;AACrB,YAAM,KAAK,IAAI,MAAM;AACrB,YAAM,IAAI,KAAK,IAAI,IAAI,GAAG,CAAC;AAC3B,UAAI,MAAM;AAAA,QACR,EAAE,EAAE,KAAM,EAAE,EAAE,IAAK,EAAE,EAAE,KAAM;AAAA,QAC7B,EAAE,KAAK,CAAC,KAAM,EAAE,KAAK,CAAC,IAAK,EAAE,KAAK,CAAC,KAAM;AAAA,QACzC,EAAE,KAAK,CAAC,KAAM,EAAE,KAAK,CAAC,IAAK,EAAE,KAAK,CAAC,KAAM;AAAA,MAC3C;AACA,UAAI,IAAI,IAAI,EAAG;AAMf,YAAM,UAAU,IAAI,UAAU,YAAY,SAAS;AACnD,UAAI,QAAQ,IAAI,OAAO;AACvB,UAAI,SAAS;AACX,cAAM,IAAI,IAAI,MAAM,KAAK,IAAI,WAAW,YAAY,SAAS,CAAC;AAC9D,YAAI,MAAM,KAAK,eAAgB,UAAS;AAAA,MAC1C;AACA,WAAK,IAAI,IAAI,GAAG,IAAI,KAAK,IAAI,GAAG,KAAK,IAAI,IAAI,KAAK,IAAI,IAAI,GAAG,IAAK,KAAK,CAAC;AAMxE,YAAM,aAAa,KAAK,OAAO,IAAI,GAAG;AACtC,UAAI,YAAY,CAAC,WAAW,aAAa,OAAO,KAAK,IAAI,IAAI,KAAK;AAChE,aAAK,OAAO,CAAC,IAAI;AACjB;AAAA,MACF;AACA,YAAM,OAAO,KAAK,IAAI,OAAO,MAAM,YAAY,GAAG;AAClD,UAAI,SAAS;AACX,aAAK,OAAO,CAAC,IAAI;AACjB;AAAA,MACF;AAEA,YAAM,OAAO,KAAK,IAAI,IAAI,IAAI,GAAG;AACjC,UAAI,KAAK,IAAI,WAAW,GAAG;AACzB,aAAK,OAAO,CAAC,IAAI;AACjB;AAAA,MACF;AACA,YAAM,OAAO,KAAK,MAAM,KAAK,IAAI,IAAI,KAAK,IAAI,MAAM;AACpD,UAAI,MAAM,IAAI;AACd,UAAI,MAAM,KAAK,IAAI,IAAI;AACvB,UAAI,IAAI,KAAK,EAAE,IAAI;AACnB,UAAI,SAAS;AACb,UAAI,IAAI;AACR,YAAM,MAAM,QAAQ,KAAK,IAAI,WAAW,IAAI,KAAK,IAAI,GAAG;AACxD,UAAI,OAAQ,OAAO,KAAK,IAAI,KAAK,IAAI,KAAM,OAAO,KAAK,IAAI,IAAI;AAC/D,UAAI,OAAO;AAAA,IACb;AAGA,aAAS,IAAI,GAAG,IAAI,KAAK,IAAI,QAAQ,IAAK,MAAK,SAAS,CAAC,IAAI,KAAK,IAAI,CAAC;AAAA,EACzE;AAAA;AAAA;AAAA,EAIQ,QAAQ,KAAa,SAA+B;AAC1D,UAAM,OAAO,KAAK,OAAO,GAAG;AAC5B,QAAI,QAAQ,KAAK,SAAS,EAAG,QAAO,KAAK,KAAK,IAAI,SAAS,KAAK,SAAS,CAAC,CAAC;AAC3E,UAAM,MAAM,IAAI,aAAa,GAAG;AAChC,UAAM,OAAO,KAAK,MAAM,GAAG,EAAG,KAAK,IAAI,SAAS,KAAK,MAAM,GAAG,EAAG,SAAS,CAAC,CAAC;AAC5E,eAAW,KAAK,KAAM,KAAI,EAAE,CAAC,IAAI,KAAK,IAAI,GAAG,EAAE,IAAI,IAAI;AACvD,WAAO;AAAA,EACT;AAAA,EAEQ,OAAO,UAAmB;AAChC,UAAM,OAAO,KAAK,OAAO,QAAQ,IAAI;AACrC,QAAI,OAAO,EAAG;AACd,UAAM,MAAM,KAAK;AACjB,QAAI;AACJ,QAAI,UAAU;AACZ,YAAM,WAAW,KAAK,MAAM,KAAK,WAAW;AAC5C,YAAM,OAAO,SAAS,KAAK,IAAI,KAAK,eAAe,SAAS,SAAS,CAAC,CAAC;AACvE,YAAM,iBAAiB,MAAM,IAAI,CAAC;AAAA,IACpC,OAAO;AACL,YAAM,KAAK,MAAM,IAAI,IAAI,YAAY,CAAC,CAAC;AAAA,IACzC;AACA,UAAM,OAAO,KAAK,IAAI,IAAI,GAAG;AAG7B,QAAI,KAAK,IAAI,WAAW,EAAG;AAC3B,UAAM,OAAO,KAAK,MAAM,IAAI,IAAI,KAAK,IAAI,MAAM;AAC/C,UAAM,MAAM,QAAQ,KAAK,IAAI,WAAW,KAAK,KAAK,IAAI,IAAI,CAAE;AAC5D,SAAK,OAAO,IAAI,IAAI;AAAA,MAClB;AAAA,MACA,KAAK,KAAK,IAAI,IAAI;AAAA,MAClB,GAAG,KAAK,EAAE,IAAI;AAAA,MACd,OAAO;AAAA,MACP,GAAG;AAAA,MACH,MAAO,OAAO,KAAK,IAAI,KAAK,IAAI,KAAM,OAAO,IAAI,IAAI;AAAA,MACrD,MAAM,WAAW,IAAM;AAAA,MACvB,KAAK,CAAC,GAAG,GAAG,CAAC;AAAA,IACf;AAGA,SAAK,IAAI,GAAG,IAAI,KAAK,IAAI,GAAG,KAAK,IAAI,GAAG,KAAM,WAAW,MAAM,KAAK;AAAA,EACtE;AAAA;AAAA,EAGA,oBAAoB;AAClB,aAAS,IAAI,GAAG,IAAI,YAAY,KAAK;AACnC,YAAM,MAAM,KAAK,OAAO,CAAC;AACzB,YAAM,IAAI,IAAI;AACd,UAAI,CAAC,KAAK;AACR,aAAK,UAAU,CAAC,IAAI;AACpB,aAAK,SAAS,CAAC,IAAI;AACnB,aAAK,SAAS,IAAI,CAAC,IAAI;AACvB,aAAK,SAAS,IAAI,CAAC,IAAI;AACvB;AAAA,MACF;AACA,WAAK,SAAS,CAAC,IAAI,IAAI,IAAI,CAAC;AAC5B,WAAK,SAAS,IAAI,CAAC,IAAI,IAAI,IAAI,CAAC;AAChC,WAAK,SAAS,IAAI,CAAC,IAAI,IAAI,IAAI,CAAC;AAChC,WAAK,UAAU,CAAC,IAAI,IAAI;AAAA,IAC1B;AAAA,EACF;AACF;AAGA,SAAS,aAA2B;AAClC,QAAM,MAAM,WAAW,MAAO;AAC9B,QAAM,MAAoB,CAAC;AAC3B,WAAS,MAAM,GAAG,MAAM,SAAS,OAAO;AACtC,UAAM,QAAQ,IAAI,aAAa,KAAK,EAAE;AACtC,UAAM,QAAQ,OAAO,GAAG;AACxB,UAAM,KAAK,IAAI,KAAK,MAAM,IAAI,IAAI,CAAC;AACnC,UAAM,KAAK,KAAK,MAAM,IAAI,IAAI,CAAC,IAAI;AACnC,aAAS,KAAK,GAAG,KAAK,GAAG,MAAM;AAC7B,YAAM,MAAM,MAAM,EAAE;AACpB,eAAS,KAAK,GAAG,KAAK,GAAG,MAAM;AAC7B,YAAI,IAAI,EAAE,MAAM,IAAK;AACrB,iBAAS,IAAI,GAAG,IAAI,GAAG,KAAK;AAC1B,mBAAS,IAAI,GAAG,IAAI,GAAG,KAAK;AAC1B,kBAAM,IAAI,KAAK,IAAI,IAAI;AACvB,kBAAM,IAAI,KAAK,IAAI,IAAI;AACvB,gBAAI,IAAI,MAAM,IAAI,GAAI,OAAM,IAAI,KAAK,CAAC,IAAI;AAAA,UAC5C;AAAA,QACF;AAAA,MACF;AAAA,IACF;AACA,UAAM,OAAkB,CAAC;AACzB,aAAS,IAAI,GAAG,IAAI,KAAK,KAAK;AAC5B,UAAI,MAAM,CAAC,IAAK,EAAG,MAAK,KAAK,EAAE,GAAG,GAAG,EAAE,CAAC;AAAA,IAC1C;AACA,aAAS,IAAI,GAAG,IAAI,GAAG,IAAK,MAAK,KAAK,EAAE,GAAG,KAAK,MAAM,IAAI,IAAI,GAAG,GAAG,GAAG,KAAK,CAAC;AAC7E,QAAI,KAAK,CAAC,IAAI,CAAC;AAAA,EACjB;AACA,SAAO;AACT;AAGA,SAAS,UAAU,KAAyB;AAC1C,QAAM,MAAM,KAAK,IAAI,QAAQ,MAAM,GAAG,EAAE,QAAQ,MAAM,GAAG,CAAC;AAC1D,QAAM,MAAM,IAAI,WAAW,IAAI,MAAM;AACrC,WAAS,IAAI,GAAG,IAAI,IAAI,QAAQ,IAAK,KAAI,CAAC,IAAI,IAAI,WAAW,CAAC;AAC9D,SAAO;AACT;AAGA,SAAS,UAAU,KAA2B;AAC5C,SAAO,IAAI,aAAa,UAAU,GAAG,EAAE,MAAM;AAC/C;AAGO,SAAS,gBAA0D;AACxE,SAAO;AAAA,IACL,GAAG;AAAA,MACD,UAAU,eAAe,QAAQ;AAAA,MACjC,UAAU,eAAe,QAAQ;AAAA,MACjC,UAAU,eAAe,QAAQ;AAAA,MACjC,UAAU,eAAe,QAAQ;AAAA,IACnC;AAAA,IACA,GAAG;AAAA,MACD,UAAU,eAAe,OAAO;AAAA,MAChC,UAAU,eAAe,OAAO;AAAA,MAChC,UAAU,eAAe,OAAO;AAAA,MAChC,UAAU,eAAe,OAAO;AAAA,IAClC;AAAA,EACF;AACF;AAOA,SAAS,cAAc,KAAc,SAAmD;AACtF,QAAM,aAAa,IAAI;AACvB,WAAS,KAAK,GAAG,KAAK,GAAG,MAAM;AAC7B,UAAM,OAAO,WAAW,EAAE;AAC1B,UAAM,IAAI,QAAQ,EAAE,KAAK,CAAC;AAC1B,QAAIA,MAAK;AACT,UAAM,WAAW,KAAK,IAAI,WAAW,KAAK,CAAC,IAAK,IAAI;AACpD,aAAS,SAAS,GAAG,SAAS,WAAW,MAAM,UAAU;AACvD,YAAM,MAAM,OAAO;AACnB,YAAM,OAAO,IAAI,QAAQ,GAAG;AAC5B,iBAAW,KAAK,MAAM;AACpB,UAAE,IAAI,EAAEA,GAAE;AACV,QAAAA;AAAA,MACF;AAAA,IACF;AACA,QAAIA,QAAO,EAAE,OAAQ,OAAM,IAAI,MAAM,SAAS,EAAE,cAAc,EAAE,MAAM,eAAeA,GAAE,EAAE;AAAA,EAC3F;AAEA,aAAW,KAAK,IAAI,KAAK;AACvB,MAAE,IAAI,SAAS;AACf,MAAE,EAAE,SAAS;AAAA,EACf;AACA,MAAI,KAAK;AACT,WAAS,IAAI,GAAG,IAAI,IAAI,WAAW,KAAK;AACtC,eAAW,KAAK,IAAI,QAAQ,CAAC,GAAI;AAC/B,UAAI,IAAI,EAAE,GAAG,EAAG,IAAI,KAAK,CAAC;AAC1B,UAAI,IAAI,EAAE,GAAG,EAAG,EAAE,KAAK,EAAE,CAAC;AAC1B,UAAI,YAAY,KAAK,CAAC,IAAI,KAAK,IAAI,GAAG,KAAK,IAAI,EAAE,CAAC,IAAI,GAAG;AACzD,UAAI,YAAY,KAAK,IAAI,CAAC,IAAI,KAAK,IAAI,GAAG,KAAK,IAAI,EAAE,CAAC,IAAI,GAAG;AAC7D;AAAA,IACF;AAAA,EACF;AACF;AAGO,SAAS,YACd,KACA,SACA,KAC6C;AAC7C,QAAM,MAAM,IAAI,aAAa,IAAI,SAAS;AAC1C,MAAI,IAAI,GAAG;AACX,QAAM,aAAa,IAAI;AAIvB,WAAS,KAAK,GAAG,KAAK,GAAG,MAAM;AAC7B,UAAM,OAAO,WAAW,EAAE;AAC1B,UAAM,OAAO,YAAY,EAAE;AAC3B,UAAM,OAAO,QAAQ,EAAE,KAAK,CAAC;AAC7B,aAAS,KAAK,GAAG,KAAK,MAAM,MAAM;AAChC,UAAI,IAAI,KAAK,EAAE;AACf,iBAAW,KAAK,IAAI,QAAQ,OAAO,EAAE,EAAI,MAAK,IAAI,EAAE,GAAG,IAAK,EAAE;AAC9D,UAAI,OAAO,EAAE,IAAI,KAAK,IAAI,KAAK,IAAI,GAAG,CAAC,IAAI;AAAA,IAC7C;AAAA,EACF;AAEA,QAAM,SAAS,IAAI,aAAa,IAAI,MAAM;AAC1C,WAAS,KAAK,GAAG,KAAK,GAAG,MAAM;AAC7B,UAAM,OAAO,WAAW,EAAE;AAC1B,UAAM,OAAO,YAAY,EAAE;AAC3B,QAAI,MAAM;AACV,aAAS,IAAI,MAAM,IAAI,OAAO,MAAM,IAAK,OAAM,KAAK,IAAI,KAAK,IAAI,CAAC,CAAE;AACpE,aAAS,IAAI,MAAM,IAAI,OAAO,MAAM,IAAK,QAAO,CAAC,IAAI,KAAK,IAAI,IAAI,CAAC,GAAI,CAAC,IAAI;AAAA,EAC9E;AACA,QAAM,UAAU,WAAW,CAAC;AAC5B,MAAI,YAAY;AAChB,MAAI,MAAM;AACV,WAAS,IAAI,GAAG,IAAI,SAAS,KAAK;AAChC,QAAI,IAAI,UAAU,CAAC,IAAK,KAAK;AAC3B,YAAM,IAAI,UAAU,CAAC;AACrB,kBAAY;AAAA,IACd;AAAA,EACF;AACA,SAAO,EAAE,WAAW,OAAO;AAC7B;AAEO,SAAS,UAAU,SAGxB;AACA,QAAM,QAAuB,CAAC;AAC9B,QAAM,SAA2B,CAAC;AAClC,WAAS,IAAI,GAAG,IAAI,SAAS,KAAK;AAChC,UAAM,WAAwB,CAAC;AAC/B,UAAM,UAA0B,CAAC;AACjC,eAAW,OAAO,QAAQ,CAAC,KAAK,CAAC,GAAG;AAIlC,YAAM,MAAM,UAAU,GAAG;AACzB,YAAM,QAAmB,CAAC;AAC1B,YAAM,MAAM,IAAI,aAAa,GAAG;AAChC,eAAS,IAAI,GAAG,IAAI,KAAK,KAAK;AAC5B,cAAM,OAAO,IAAI,KAAK,CAAC;AACvB,cAAM,IAAI,IAAI,IAAI,OAAO,KAAK,QAAQ;AACtC,YAAI,KAAK,EAAG,OAAM,KAAK,EAAE,GAAG,GAAG,IAAI,IAAI,KAAK,GAAG,CAAC;AAChD,YAAI,CAAC,IAAI,IAAI;AAAA,MACf;AACA,eAAS,KAAK,KAAK;AACnB,cAAQ,KAAK,GAAG;AAAA,IAClB;AACA,UAAM,KAAK,QAAQ;AACnB,WAAO,KAAK,OAAO;AAAA,EACrB;AACA,SAAO,EAAE,OAAO,OAAO;AACzB;AAEA,SAAS,iBAAiB,MAAiB,GAAmB;AAC5D,MAAI,QAAQ;AACZ,aAAW,KAAK,KAAM,UAAS,EAAE;AACjC,MAAI,MAAM,IAAI;AACd,aAAW,KAAK,MAAM;AACpB,WAAO,EAAE;AACT,QAAI,OAAO,EAAG,QAAO,EAAE;AAAA,EACzB;AACA,SAAO,KAAK,KAAK,SAAS,CAAC,EAAG;AAChC;AAEA,SAAS,QAAQ,GAAiB,GAAW,GAAmB;AAC9D,QAAM,KAAK,IAAI;AACf,QAAM,KAAK,IAAI;AACf,QAAM,KAAK,EAAE,EAAE,IAAK,EAAE,EAAE;AACxB,QAAM,KAAK,EAAE,KAAK,CAAC,IAAK,EAAE,KAAK,CAAC;AAChC,QAAM,KAAK,EAAE,KAAK,CAAC,IAAK,EAAE,KAAK,CAAC;AAChC,SAAO,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK,KAAK,EAAE;AAC9C;AAIA,IAAM;AAAA;AAAA,EAAwB;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAuB9B,IAAM;AAAA;AAAA,EAAuB;AAAA;AAAA;AAAA;AAAA,yBAIJ,QAAQ,KAAK,KAAK,GAAG,CAAC;AAAA,wBACvB,QAAQ,IAAI,KAAK,GAAG,CAAC;AAAA,0BACnB,QAAQ,MAAM,KAAK,GAAG,CAAC;AAAA,0BACvB,QAAQ,MAAM,KAAK,GAAG,CAAC;AAAA,0BACvB,QAAQ,MAAM,KAAK,GAAG,CAAC;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAsBjD,IAAM;AAAA;AAAA,EAAwB;AAAA;AAAA;AAAA;AAAA,0BAIJ,QAAQ,MAAM,KAAK,GAAG,CAAC;AAAA,0BACvB,QAAQ,MAAM,KAAK,GAAG,CAAC;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAejD,IAAM;AAAA;AAAA,EAAuB;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAS7B,IAAM;AAAA;AAAA,EAAuB;AAAA;AAAA,yBAEJ,QAAQ,MAAM,KAAK,GAAG,CAAC;AAAA;AAAA;AAAA;AAAA;AAAA;AAMhD,IAAM;AAAA;AAAA,EAAsB;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAQ5B,IAAM;AAAA;AAAA,EAAsB;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAsD5B,IAAM,gBAAiC;AAAA,EACrC,MAAM;AAAA,EACN,UAAU;AAAA,EACV,OAAO;AAAA,EACP,OAAO;AAAA,EACP,OAAO;AAAA,EACP,WAAW;AAAA,EACX,OAAO;AAAA,EACP,MAAM;AACR;AAEA,IAAM,WAAW;AAGjB,SAAS,OAAO,aAAoC;AAClD,SAAO,gBAAgB,aACnB;AAAA,IACE,MAAM,IAAU,cAAQ,KAAK,KAAK,GAAG;AAAA,IACrC,QAAQ,IAAU,cAAQ,GAAG,MAAM,CAAC;AAAA,IACpC,QAAQ,IAAU,cAAQ,GAAG,KAAK,CAAC;AAAA,IACnC,SAAS,CAAC,IAAI,EAAE;AAAA,IAChB,QAAQ,IAAU,cAAQ,KAAK,KAAK,IAAI;AAAA,IACxC,QAAQ;AAAA,EACV,IACA;AAAA,IACE,MAAM,IAAU,cAAQ,MAAM,MAAM,GAAG;AAAA,IACvC,QAAQ,IAAU,cAAQ,MAAM,OAAO,IAAI;AAAA,IAC3C,QAAQ,IAAU,cAAQ,KAAK,GAAG,CAAC;AAAA,IACnC,SAAS,CAAC,IAAI,CAAC;AAAA,IACf,QAAQ,IAAU,cAAQ,KAAK,GAAG,CAAC;AAAA,IACnC,QAAQ,CAAC,KAAK,KAAK;AAAA,EACrB;AACN;AAEA,SAAS,aAAa;AAAA,EACpB,OAAO;AAAA,EACP;AAAA,EACA;AAAA,EACA;AACF,GAKG;AACD,QAAM,EAAE,WAAW,IAAI,SAAS;AAChC,QAAM,UAAU,QAAQ,MAAM,cAAc,GAAG,CAAC,CAAC;AACjD,QAAM,MAAM;AAAA,IACV,MAAM,IAAI,QAAQ,MAAM,aAAa,OAAO,QAAQ,OAAO;AAAA,IAC3D,CAAC,aAAa,OAAO,QAAQ,OAAO;AAAA,EACtC;AACA,QAAM,SAAS,QAAQ,MAAM,OAAO,WAAW,GAAG,CAAC,WAAW,CAAC;AAE/D,QAAM,WAAW;AAAA,IACf,MAAM,IAAU,oBAAc,OAAO,QAAQ,CAAC,GAAG,OAAO,QAAQ,CAAC,CAAC;AAAA,IAClE,CAAC,MAAM;AAAA,EACT;AAEA,QAAM,OAAO,QAAQ,MAAM;AACzB,UAAM,MAAM,IAAI;AAChB,UAAM,QAAQ,IAAU,qBAAe;AACvC,UAAM,aAAa,YAAY,IAAU,sBAAgB,IAAI,WAAW,CAAC,CAAC;AAC1E,UAAM,UAAU,IAAU,sBAAgB,IAAI,UAAU,CAAC;AACzD,YAAQ,SAAe,sBAAgB;AACvC,UAAM,aAAa,QAAQ,OAAO;AAElC,UAAM,SAAS,IAAU,qBAAe;AACxC,UAAM,UAAU,IAAU,sBAAgB,IAAI,UAAU,CAAC;AACzD,YAAQ,SAAe,sBAAgB;AACvC,WAAO,aAAa,YAAY,OAAO;AACvC,UAAM,WAAW,IAAU,sBAAgB,IAAI,WAAW,CAAC;AAC3D,aAAS,SAAe,sBAAgB;AACxC,WAAO,aAAa,QAAQ,QAAQ;AAEpC,UAAM,IAAI;AACV,UAAM,iBAAiB,IAAU,aAAO,OAAO,OAAO,MAAM,GAAG,CAAC;AAChE,WAAO,iBAAiB,IAAU,aAAO,OAAO,OAAO,MAAM,GAAG,CAAC;AAEjE,UAAM,QAAQ,IAAU,qBAAe;AACvC,UAAM,aAAa,YAAY,IAAU,sBAAgB,IAAI,eAAe,CAAC,CAAC;AAC9E,UAAM,aAAa,UAAU,IAAU,sBAAgB,IAAI,aAAa,CAAC,CAAC;AAC1E,UAAM,iBAAiB,IAAU,aAAO,OAAO,OAAO,MAAM,GAAG,CAAC;AAChE,WAAO,CAAC,OAAO,QAAQ,KAAK;AAAA,EAC9B,GAAG,CAAC,KAAK,MAAM,CAAC;AAEhB,QAAM,UAAU,QAAQ,MAAM,kBAAkB,GAAG,CAAC,CAAC;AACrD,QAAM,WAAW,QAAQ,MAAM,kBAAkB,GAAG,CAAC,CAAC;AACtD,QAAM,UAAU;AAAA,IACd,MACE,IAAU,qBAAe;AAAA,MACvB,cAAc;AAAA,MACd,gBAAgB;AAAA,MAChB,aAAa;AAAA,MACb,YAAY;AAAA,MACZ,UAAgB;AAAA,MAChB,UAAU,CAAC;AAAA,IACb,CAAC;AAAA,IACH,CAAC;AAAA,EACH;AACA,QAAM,SAAS;AAAA,IACb,MACE,IAAU,qBAAe;AAAA,MACvB,cAAc;AAAA,MACd,gBAAgB;AAAA,MAChB,YAAY;AAAA,MACZ,UAAgB;AAAA,MAChB,UAAU,EAAE,OAAO,EAAE,OAAO,EAAE,EAAE;AAAA,IAClC,CAAC;AAAA,IACH,CAAC;AAAA,EACH;AAEA;AAAA,IACE,MAAM,MAAM;AACV,iBAAW,KAAK,KAAM,GAAE,QAAQ;AAChC,eAAS,QAAQ;AACjB,cAAQ,QAAQ;AAChB,eAAS,QAAQ;AACjB,cAAQ,QAAQ;AAChB,aAAO,QAAQ;AAAA,IACjB;AAAA,IACA,CAAC,MAAM,UAAU,SAAS,UAAU,SAAS,MAAM;AAAA,EACrD;AAEA,QAAM,eAAe,OAAO,CAAC;AAC7B,QAAM,iBAAiB,OAAO,CAAC;AAG/B,QAAM,eAAe,OAAuB,IAAI;AAChD,QAAM,UAAU,OAAoB,IAAI;AAIxC,WAAS,CAAC,UAAU;AAClB,UAAM,QAAQ,QAAQ,IAAI,IAAI,SAAS,QAAQ;AAC/C,UAAM,KAAK,KAAK,IAAI,MAAM,MAAM,cAAc,eAAe,SAAS,IAAI;AAC1E,mBAAe,UAAU,MAAM,MAAM;AACrC,iBAAa,WAAW,KAAK,KAAK,IAAI,OAAO,CAAC;AAC9C,UAAM,IAAI,QAAQ,IAAI,IAAI,aAAa;AAIvC,UAAM,aAAa,aAAa,YAAY;AAC5C,QAAI,YAAY;AACd,eAAS,IAAI,GAAG,IAAI,KAAK,IAAK,KAAI,KAAK,IAAI,IAAI,CAAC;AAChD,mBAAa,UAAU;AAAA,IACzB,WAAW,CAAC,QAAQ,GAAG;AACrB,UAAI,KAAK,KAAK,KAAK,IAAI,OAAO,CAAC,GAAG,SAAS,QAAQ,SAAS;AAAA,IAC9D;AACA,QAAI,kBAAkB;AAEtB,eAAW,CAAC,KAAK,IAAI,KAAK;AAAA,MACxB,CAAC,KAAK,CAAC,GAAI,MAAM;AAAA,MACjB,CAAC,KAAK,CAAC,GAAI,UAAU;AAAA,MACrB,CAAC,KAAK,CAAC,GAAI,MAAM;AAAA,IACnB,GAAY;AACV,YAAM,IAAI,IAAI,aAAa,IAAI;AAC/B,QAAE,cAAc;AAAA,IAClB;AAEA,UAAM,MAAM,MAAM;AAElB,UAAM,MAAM,WAAW,KAAK,IAAI,SAAS,QAAQ,MAAM,IAAI;AAC3D,QAAI,KAAK,IAAI,IAAI,MAAM,GAAG,IAAI,MAAM;AAClC,UAAI,MAAM;AACV,UAAI,uBAAuB;AAAA,IAC7B;AACA,UAAM,QAAQ,SAAS,QAAQ;AAC/B,QAAI,SAAS;AAAA,MACX,OAAO,KAAK,IAAI,KAAK,IAAI,IAAI,IAAI,IAAI,OAAO;AAAA,MAC5C,OAAO,KAAK,IAAI,KAAK,IAAI,IAAI,QAAQ,CAAG,IAAI,OAAO;AAAA,MACnD,OAAO,KAAK,IAAI,KAAK,IAAI,IAAI,KAAK,IAAI,OAAO;AAAA,IAC/C;AACA,QAAI,OAAO,OAAO,MAAM;AAExB,UAAM,SAAS,MAAM,GAAG,WAAW;AACnC,UAAM,SAAS,KAAK,IAAU,gBAAU,SAAS,IAAI,MAAM,GAAG,CAAC;AAC/D,eAAW,KAAK,CAAC,SAAS,QAAQ,GAAG;AACnC,QAAE,SAAS,UAAW,QAAQ;AAC9B,QAAE,SAAS,QAAS,QAAQ;AAC5B,QAAE,SAAS,KAAM,QAAQ,MAAM,GAAG,cAAc;AAChD,QAAE,SAAS,WAAY,QAAQ,SAAS,QAAQ;AAChD,QAAE,SAAS,OAAQ,QAAQ,SAAS,QAAQ;AAC5C,QAAE,SAAS,UAAW,QAAQ,SAAS,QAAQ;AAAA,IACjD;AACA,QAAI,QAAQ,QAAS,SAAQ,QAAQ,SAAS,IAAI,IAAI,SAAS,QAAQ;AACvE,WAAO,SAAS,MAAO,QAAQ;AAI/B,QAAI,CAAC,QAAQ,KAAK,WAAY,YAAW;AAAA,EAC3C,CAAC;AAGD,SACE,qBAAC,WACC;AAAA;AAAA,MAAC;AAAA;AAAA,QACC,UAAU;AAAA,QACV,UAAU;AAAA,QACV,UAAU,OAAO,OAAO,QAAQ;AAAA,QAChC,UAAU,CAAC,GAAG,OAAO,QAAQ,CAAC;AAAA,QAC9B,aAAa;AAAA;AAAA,MACR;AAAA,IACP;AAAA,IAGA,oBAAC,WAAM,KAAK,SAAS,UAAU,OAAO,OAAO,QAAQ,GACnD,+BAAC,WAAM,UAAU,OAAO,OAAO,MAAM,EAAE,OAAO,EAAE,QAAQ,GACtD;AAAA,0BAAC,kBAAa,UAAU,KAAK,CAAC,GAAG,UAAU,SAAS,aAAa,GAAG;AAAA,MACpE,oBAAC,YAAO,UAAU,KAAK,CAAC,GAAG,UAAU,SAAS,aAAa,GAAG;AAAA,MAC9D,oBAAC,YAAO,UAAU,KAAK,CAAC,GAAG,UAAU,UAAU,aAAa,GAAG;AAAA,OACjE,GACF;AAAA,KACF;AAEJ;AAEA,SAAS,oBAAoB;AAC3B,SAAO,IAAU,qBAAe;AAAA,IAC9B,cAAc;AAAA,IACd,gBAAgB;AAAA,IAChB,aAAa;AAAA,IACb,YAAY;AAAA,IACZ,UAAgB;AAAA,IAChB,UAAU;AAAA,MACR,WAAW,EAAE,OAAO,IAAI;AAAA,MACxB,SAAS,EAAE,OAAO,IAAI;AAAA,MACtB,MAAM,EAAE,OAAO,EAAE;AAAA,MACjB,YAAY,EAAE,OAAO,EAAI;AAAA,MACzB,QAAQ,EAAE,OAAO,EAAE;AAAA,MACnB,WAAW,EAAE,OAAO,MAAM;AAAA,MAC1B,WAAW,EAAE,OAAO,KAAK;AAAA,IAC3B;AAAA,EACF,CAAC;AACH;AAEA,SAAS,oBAAoB;AAC3B,SAAO,IAAU,qBAAe;AAAA,IAC9B,cAAc;AAAA,IACd,gBAAgB;AAAA,IAChB,aAAa;AAAA,IACb,YAAY;AAAA,IACZ,UAAgB;AAAA,IAChB,UAAU;AAAA,MACR,WAAW,EAAE,OAAO,IAAI;AAAA,MACxB,SAAS,EAAE,OAAO,IAAI;AAAA,MACtB,MAAM,EAAE,OAAO,EAAE;AAAA,MACjB,YAAY,EAAE,OAAO,EAAI;AAAA,MACzB,QAAQ,EAAE,OAAO,EAAE;AAAA,MACnB,WAAW,EAAE,OAAO,MAAM;AAAA,MAC1B,WAAW,EAAE,OAAO,KAAK;AAAA,IAC3B;AAAA,EACF,CAAC;AACH;AAEA,SAAS,UAAmB;AAC1B,SACE,OAAO,WAAW,eAAe,OAAO,WAAW,kCAAkC,EAAE;AAE3F;AAyBO,IAAM,aAAkC,CAAC;AAAA,EAC9C;AAAA,EACA;AAAA,EACA,cAAc;AAAA,EACd;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AACF,MAAM;AAIJ,QAAM,SAAS,OAAO,SAAS;AAC/B,QAAM,EAAE,OAAO,OAAO,IAAI,QAAQ,MAAM;AACtC,UAAM,MAAM,iBAAiB,SAAY,gBAAgB;AACzD,QAAI,UAAU,OAAO,KAAK,GAAG,EAAE,WAAW,GAAG;AAC3C,aAAO,EAAE,OAAO,WAAW,GAAG,QAAQ,CAAC,EAAsB;AAAA,IAC/D;AACA,UAAM,UAAU,UAAU,GAAG;AAC7B,WAAO,QAAQ,MAAM,MAAM,CAAC,MAAM,EAAE,SAAS,CAAC,IAC1C,UACA,EAAE,OAAO,WAAW,GAAG,QAAQ,CAAC,EAAsB;AAAA,EAC5D,GAAG,CAAC,cAAc,MAAM,CAAC;AACzB,QAAM,QAAQ,OAAwB,EAAE,GAAG,cAAc,CAAC;AAC1D,YAAU,MAAM;AACd,UAAM,UAAU;AAAA,MACd,MAAM,QAAQ,cAAc;AAAA,MAC5B,UAAU,YAAY,cAAc;AAAA,MACpC,OAAO,SAAS,cAAc;AAAA,MAC9B,OAAO,SAAS,cAAc;AAAA,MAC9B,OAAO,SAAS,cAAc;AAAA,MAC9B,WAAW,aAAa,cAAc;AAAA,MACtC,OAAO,SAAS,cAAc;AAAA,MAC9B,MAAM,QAAQ,cAAc;AAAA,IAC9B;AAAA,EACF,GAAG,CAAC,MAAM,UAAU,OAAO,OAAO,OAAO,WAAW,OAAO,IAAI,CAAC;AAEhE,MAAI,QAAQ;AACV,WACE;AAAA,MAAC;AAAA;AAAA,QACC,eAAY;AAAA,QACZ;AAAA,QACA,OAAO,EAAE,GAAG,OAAO,UAAU,YAAY,OAAO,GAAG,YAAY,SAAS;AAAA;AAAA,IAC1E;AAAA,EAEJ;AAEA,SACE;AAAA,IAAC;AAAA;AAAA,MACC;AAAA,MACA,OAAO;AAAA,QACL,GAAG;AAAA,QACH,UAAU;AAAA,QACV,OAAO;AAAA,QACP,YAAY;AAAA,MACd;AAAA,MAEA;AAAA,QAAC;AAAA;AAAA,UACC,eAAY;AAAA,UACZ,KAAK,CAAC,GAAG,CAAC;AAAA,UACV,WAAW,QAAQ,IAAI,WAAW;AAAA,UAClC,QAAQ,EAAE,UAAU,OAAO,WAAW,EAAE,KAAK,QAAQ,GAAG,KAAK,SAAS;AAAA,UACtE,IAAI;AAAA;AAAA;AAAA;AAAA;AAAA,YAKF,OAAO;AAAA,YACP,WAAW;AAAA,YACX,iBAAiB;AAAA,YACjB,uBAAuB;AAAA,UACzB;AAAA,UACA,WAAW,CAAC,EAAE,GAAG,MAAM;AACrB,eAAG,cAAc,OAAU,CAAC;AAAA,UAC9B;AAAA,UAEA,8BAAC,gBAAa,OAAc,aAA0B,OAAc,QAAgB;AAAA;AAAA,MACtF;AAAA;AAAA,EACF;AAEJ;","names":["eo"]}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@bergetai/ui",
-  "version": "0.5.0",
+  "version": "0.6.0",
   "description": "Berget Design System - Scandinavian-inspired design system with console aesthetic",
   "type": "module",
   "license": "MIT",
@@ -19,6 +19,11 @@
       "types": "./dist/shiki.d.ts",
       "import": "./dist/shiki.js",
       "require": "./dist/shiki.cjs"
+    },
+    "./neural-flow": {
+      "types": "./dist/neural-flow.d.ts",
+      "import": "./dist/neural-flow.js",
+      "require": "./dist/neural-flow.cjs"
     }
   },
   "main": "./dist/index.cjs",
@@ -45,16 +50,29 @@
     "tailwind-merge": "^3.5.0"
   },
   "peerDependencies": {
+    "@react-three/fiber": "^9.6.1",
     "lucide-react": "^1.23.0",
     "motion": "^12.38.0",
     "react": "^19.0.0",
     "react-dom": "^19.0.0",
-    "shiki": "^4.3.1"
+    "shiki": "^4.3.1",
+    "three": "^0.184.0"
+  },
+  "peerDependenciesMeta": {
+    "three": {
+      "optional": true
+    },
+    "@react-three/fiber": {
+      "optional": true
+    }
   },
   "devDependencies": {
+    "@react-three/fiber": "^9.6.1",
     "@storybook/react": "^10.4.6",
     "@types/react": "^19.0.0",
     "@types/react-dom": "^19.0.0",
+    "@types/three": "^0.184.1",
+    "three": "^0.184.0",
     "tsup": "^8.3.0",
     "typescript": "^5.7.0"
   }