// $vectorSearch from the official Node.js driver. Reads the shared query vectors and compares the top-5 ids with pymongo.
import { MongoClient } from "mongodb";
import { readFileSync } from "node:fs";

const uri = process.env.MONGODB_URI ?? "mongodb://mongo:27017/?directConnection=true";
const qvecs = JSON.parse(readFileSync("/work/qvecs.json", "utf8"));
const py = JSON.parse(readFileSync("/work/top5_pymongo.json", "utf8"));

const client = new MongoClient(uri);
await client.connect();
const col = client.db("kb").collection("chunks");
console.log("driver mongodb 7.8.0, node", process.version);

let same = 0;
for (const [q, vec] of Object.entries(qvecs)) {
  const docs = await col
    .aggregate([
      { $vectorSearch: { index: "kb_vec", path: "embedding", queryVector: vec, numCandidates: 100, limit: 5 } },
      { $project: { url: 1, score: { $meta: "vectorSearchScore" } } },
    ])
    .toArray();
  const ids = docs.map((d) => d._id);
  const key = q.replace(/^search_query: /, "");
  const ok = JSON.stringify(ids) === JSON.stringify(py[key]);
  same += ok;
  console.log(`${ok ? "same" : "DIFF"}  ${key.padEnd(72)} top hit ${docs[0].url.replace("https://webcrawlerapi.com", "")} ${docs[0].score.toFixed(4)}`);
}
console.log(`top-5 ids identical to pymongo (same order) on ${same} of ${Object.keys(qvecs).length} queries`);
await client.close();