src/lib/sample-artifacts.ts

/**
 * Sample artifact builders — emitted by the demo engine at the end of a plan
 * execution (plus on-demand for phylo/MSA/structure prompts).
 *
 * Two shapes:
 *   1) Synthetic JSON/CSV artifacts embedded as `data:` URLs so downloads work
 *      fully client-side without a backend round-trip. This mirrors the existing
 *      test fixture at `.data/tasks/sess_b2c176cf8e6b.json`.
 *   2) Public sample files (`/sample-data/*.{nwk,fasta,pdb}`) referenced by URL.
 *      Agent A owns those viewers; we just point at them with realistic sizes.
 */

import type { Artifact, PlanProposal, TraceEntry } from "./types";

// ---- utilities -------------------------------------------------------------

function encodeBase64(s: string): string {
  if (typeof window === "undefined") {
    return Buffer.from(s, "utf8").toString("base64");
  }
  return btoa(unescape(encodeURIComponent(s)));
}

function dataUrl(mime: string, content: string): string {
  return `data:${mime};base64,${encodeBase64(content)}`;
}

function newId(): string {
  return "art-" + Math.random().toString(36).slice(2, 10);
}

function utf8Bytes(s: string): number {
  if (typeof TextEncoder !== "undefined") {
    return new TextEncoder().encode(s).length;
  }
  return Buffer.byteLength(s, "utf8");
}

// ---- synthetic artifact builders ------------------------------------------

export interface DesignBriefInput {
  target: string;
  indication?: string;
  mechanism?: string;
  format?: string;
  epitope?: string;
  affinity?: string;
  species?: string;
  developability?: string;
  route?: string;
  competition?: string;
  seed?: string;
  scale?: string;
}

export function buildDesignBriefJson(brief: DesignBriefInput): string {
  const payload = {
    schema: "proteinea.design_brief/v1",
    generated_at: new Date().toISOString(),
    target: brief.target,
    indication: brief.indication || null,
    mechanism: brief.mechanism || null,
    format: brief.format || "VHH",
    epitope: brief.epitope || "no preference",
    affinity_goal: brief.affinity || null,
    species_cross_reactivity: brief.species || null,
    developability: brief.developability || null,
    route_of_administration: brief.route || null,
    competitive_landscape: brief.competition || null,
    starting_point: brief.seed || "de novo",
    design_count: parseInt(brief.scale || "50", 10) || 50,
  };
  return JSON.stringify(payload, null, 2);
}

export function buildPipelinePlanJson(plan: PlanProposal): string {
  const payload = {
    schema: "proteinea.pipeline_plan/v1",
    generated_at: new Date().toISOString(),
    title: plan.title,
    steps: plan.steps.map((s, i) => ({
      index: i + 1,
      label: s.label,
      tools: s.tools,
    })),
  };
  return JSON.stringify(payload, null, 2);
}

export function buildExecutionTraceJson(trace: TraceEntry[]): string {
  const payload = {
    schema: "proteinea.execution_trace/v1",
    generated_at: new Date().toISOString(),
    event_count: trace.length,
    events: trace,
  };
  return JSON.stringify(payload, null, 2);
}

export function buildTopDesignsCsv(n = 10): string {
  const aa = "ACDEFGHIKLMNPQRSTVWY";
  const rows: string[] = [
    "rank,design_id,cdr_h3,length,iPAE,pLDDT,affinity_nM,consensus_score",
  ];
  for (let i = 0; i < n; i++) {
    const id = `d${String(i + 1).padStart(3, "0")}`;
    const len = 14 + Math.floor(Math.random() * 8);
    let cdr = "";
    for (let j = 0; j < len; j++) cdr += aa[Math.floor(Math.random() * aa.length)];
    const iPAE = (5 + Math.random() * 5).toFixed(2);
    const pLDDT = (82 + Math.random() * 12).toFixed(1);
    const kd = (0.3 + Math.random() * 40).toFixed(2);
    const score = (0.55 + Math.random() * 0.4).toFixed(3);
    rows.push(`${i + 1},${id},${cdr},${len},${iPAE},${pLDDT},${kd},${score}`);
  }
  return rows.join("\n");
}

// ---- artifact constructors -------------------------------------------------

export function makeJsonArtifact(name: string, content: string): Artifact {
  return {
    id: newId(),
    name,
    bytes: utf8Bytes(content),
    mime: "application/json",
    url: dataUrl("application/json", content),
    createdAt: Date.now(),
  };
}

export function makeCsvArtifact(name: string, content: string): Artifact {
  return {
    id: newId(),
    name,
    bytes: utf8Bytes(content),
    mime: "text/csv",
    url: dataUrl("text/csv", content),
    createdAt: Date.now(),
  };
}

// ---- public sample-data artifacts (Agent A owns the files) -----------------

export function makePhyloTreeArtifact(): Artifact {
  return {
    id: newId(),
    name: "spike.nwk",
    bytes: 4096,
    mime: "text/x-nh",
    url: "/sample-data/spike.nwk",
    createdAt: Date.now(),
  };
}

export function makeMsaArtifact(): Artifact {
  return {
    id: newId(),
    name: "igg-fc.fasta",
    bytes: 8192,
    mime: "text/x-fasta",
    url: "/sample-data/igg-fc.fasta",
    createdAt: Date.now(),
  };
}

export function makeStructureArtifact(): Artifact {
  return {
    id: newId(),
    name: "her2-1n8z-mini.pdb",
    bytes: 24576,
    mime: "chemical/x-pdb",
    url: "/sample-data/her2-1n8z-mini.pdb",
    createdAt: Date.now(),
  };
}

/**
 * Classify a raw user prompt for which sample-data viewer(s) to surface.
 * Keyword-based; matches are non-exclusive.
 */
export function detectSampleDataHints(msg: string): {
  phylo: boolean;
  msa: boolean;
  structure: boolean;
} {
  const m = msg.toLowerCase();
  return {
    phylo:
      m.includes("phylogen") ||
      m.includes("tree") ||
      m.includes(".nwk") ||
      m.includes("newick") ||
      m.includes("clade"),
    msa:
      m.includes("msa") ||
      m.includes("alignment") ||
      m.includes("multiple sequence") ||
      m.includes("fasta") ||
      m.includes("conservation"),
    structure:
      m.includes("structure") ||
      m.includes("pdb") ||
      m.includes("crystal") ||
      m.includes("cryo-em") ||
      m.includes("cryoem") ||
      m.includes("3d structure"),
  };
}

// ---- KB topic catalog ------------------------------------------------------

export interface KbTopic {
  name: string;
  ms: number;
}

/**
 * Deterministic-ish KB topics we "load" before planning. Timings are realistic
 * (60-300ms) and slightly randomized so no two plan previews look identical.
 */
export function buildKbTopics(): KbTopic[] {
  const jitter = (base: number, spread: number) =>
    Math.round(base + (Math.random() - 0.5) * spread);
  return [
    { name: "antibody design best practices", ms: jitter(150, 60) },
    { name: "benchmark catalog", ms: jitter(80, 40) },
    { name: "target intelligence", ms: jitter(120, 60) },
    { name: "developability heuristics", ms: jitter(95, 40) },
  ];
}