drop orphan workload stubs from explorer

Author: atharvas

src/routes/explorer/+page.server.js

@@ -1,5 +1,3 @@
-import fs from "node:fs";
-import path from "node:path";
 // The explorer ships a curated slice pinned to commit 8907cb99, where the
 // CSV's agent_recording paths line up with the .cast files actually present
 // under src/data/recordings/. The site-wide CSV (website_data_lite.csv) was
@@ -8,99 +6,9 @@ import path from "node:path";
 import rawRows from "$data/website_data_lite.explorer.csv";
 import codesData from "$data/website_data_codes.explorer.json";
 import { buildWorkloads, summarizeFacets } from "$utils/explorer.js";
-import { AGENT_IDS } from "$utils/agents.js";
-
-const RECORDINGS_DIR = path.resolve("src/data/recordings");
-
-// Per the historical CSV at 8907cb99 (which was committed alongside these
-// .cast files), the on-disk agent-N suffixes map as follows:
-const AGENT_NUM_TO_ID = {
-	"2": AGENT_IDS.HUMAN,
-	"3": AGENT_IDS.CLAUDE,
-	"4": AGENT_IDS.GPT5
-};
-
-function buildRecordingManifest(root) {
-	const manifest = {};
-	if (!fs.existsSync(root)) return manifest;
-	for (const ts of fs.readdirSync(root)) {
-		const tsPath = path.join(root, ts);
-		if (!fs.statSync(tsPath).isDirectory()) continue;
-		for (const taskId of fs.readdirSync(tsPath)) {
-			const taskPath = path.join(tsPath, taskId);
-			if (!fs.statSync(taskPath).isDirectory()) continue;
-			for (const runDir of fs.readdirSync(taskPath)) {
-				const runPath = path.join(taskPath, runDir);
-				if (!fs.statSync(runPath).isDirectory()) continue;
-				if (!fs.existsSync(path.join(runPath, "sessions", "agent.cast")))
-					continue;
-				const m = runDir.match(/\.agent-(\d+)-/);
-				if (!m) continue;
-				const agentId = AGENT_NUM_TO_ID[m[1]];
-				if (!agentId) continue;
-				if (!manifest[taskId]) manifest[taskId] = {};
-				manifest[taskId][agentId] =
-					`/recordings/${ts}/${taskId}/${runDir}/sessions/agent.cast`;
-			}
-		}
-	}
-	return manifest;
-}
-
-// task_id like "pandas_dev-pandas_12" → repo_name "pandas_dev-pandas".
-// The last `_<number>` segment is the task index; everything before is repo.
-function repoFromTaskId(taskId) {
-	const m = taskId.match(/^(.*)_\d+$/);
-	return m ? m[1] : taskId;
-}
-
-// Build stub workloads for task_ids that have recordings on disk but no row
-// in the CSV. They land in the explorer as bare cards: task_id, repo, and
-// the per-agent recording links — no speedups or benchmark code to show.
-function synthesizeOrphans(manifest, csvTaskIds) {
-	const out = [];
-	let synthId = 1_000_000;
-	for (const [taskId, recordings] of Object.entries(manifest)) {
-		if (csvTaskIds.has(taskId)) continue;
-		out.push({
-			id: String(synthId++),
-			task_id: taskId,
-			level: "",
-			benchmark_name: taskId,
-			repo_name: repoFromTaskId(taskId),
-			agent_id: "",
-			"agent/nop": null,
-			"oracle/nop": null,
-			agent_recording: null,
-			__orphanRecordings: recordings
-		});
-	}
-	return out;
-}
 
 export async function load() {
-	const recordings = buildRecordingManifest(RECORDINGS_DIR);
-	const csvTaskIds = new Set(rawRows.map((r) => r.task_id));
-	const orphanWorkloads = synthesizeOrphans(recordings, csvTaskIds);
-	const workloads = [
-		...buildWorkloads(rawRows, codesData),
-		...orphanWorkloads.map((o) => ({
-			key: `${o.task_id}::::`,
-			task_id: o.task_id,
-			level: o.level,
-			benchmark_name: o.benchmark_name,
-			repo_name: o.repo_name,
-			id: o.id,
-			oracle: null,
-			agents: {},
-			recordings: o.__orphanRecordings,
-			codeText: null,
-			codeFqName: null,
-			bestAgentId: null,
-			bestAgentSpeedup: null,
-			beatsOracle: false
-		}))
-	];
+	const workloads = buildWorkloads(rawRows, codesData);
 	const facets = summarizeFacets(workloads);
 	return {
 		workloads,