Add /api/_diag probe; restore @duckdb/node-api 1.5.3-r.2

The 1.5.2 rollback did not help — the rebuilt Space still fails with
"don't know what type:" on a freshly purged + re-downloaded cache, on the
old binding. So neither the disk cache nor the binding version is the
cause. The Space's pinned snapshot (warehouse/2026-05-29T00-24-44Z) also
reads cleanly under linux/amd64 Rosetta emulation locally, which points
at a real-hardware (Sapphire Rapids / AVX-512) Parquet-decode path that
emulation doesn't exercise.

Adds a temporary read-only /api/_diag endpoint to isolate, on the Space
hardware itself, which nested column's to_json decode trips the Thrift
error, and whether thread count or the local-file vs httpfs read path
matters. Restores the binding to the preferred 1.5.3-r.2 (version proven
irrelevant to the failure).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

app/api/_diag/route.ts

@@ -0,0 +1,156 @@
+import "server-only"
+
+import { NextResponse } from "next/server"
+import { getConnection } from "@/lib/duckdb"
+
+export const dynamic = "force-dynamic"
+
+// Temporary read-only diagnostic for the linux-x64 Space "don't know what
+// type:" Parquet-decode failure. Probes the live DuckDB connection on the
+// Space hardware to isolate which column / read-path / thread setting
+// triggers it. Remove once the root cause is fixed.
+//
+// Usage: /api/_diag?id=deepseek/deepseek-v4-flash[&threads=1][&http=1]
+
+const NESTED_COLUMNS = [
+  "model_info",
+  "derived_tags",
+  "score_details",
+  "generation_config",
+  "source_metadata",
+  "source_data",
+  "eval_library",
+  "evalcards_annotations",
+]
+
+type ProbeResult = {
+  name: string
+  ok: boolean
+  rows?: number
+  ms: number
+  error?: string
+  note?: string
+}
+
+async function probe(
+  name: string,
+  fn: () => Promise<number>,
+  note?: string,
+): Promise<ProbeResult> {
+  const t0 = Date.now()
+  try {
+    const rows = await fn()
+    return { name, ok: true, rows, ms: Date.now() - t0, note }
+  } catch (err) {
+    return {
+      name,
+      ok: false,
+      ms: Date.now() - t0,
+      error: err instanceof Error ? `${err.name}: ${err.message}` : String(err),
+      note,
+    }
+  }
+}
+
+export async function GET(request: Request) {
+  const { searchParams } = new URL(request.url)
+  const modelKey = searchParams.get("id") ?? "deepseek/deepseek-v4-flash"
+  const threadsParam = searchParams.get("threads")
+  const useHttp = searchParams.get("http") === "1"
+
+  const connection = await getConnection()
+  const snapshotUrl = (process.env.SNAPSHOT_URL ?? "").replace(/\/+$/, "")
+
+  // When http=1, read straight from the remote parquet (range reads over
+  // HTTPS) instead of the disk-cached local file the views point at, so we
+  // can tell a local-file/mmap issue apart from a decode issue.
+  const source = useHttp
+    ? `read_parquet('${snapshotUrl}/eval_results_view.parquet')`
+    : `eval_results_view`
+
+  // Count rows through a reader; the JSON-row variant is what the app uses.
+  const run = async (sql: string, params?: unknown[]) => {
+    const reader = params?.length
+      ? await connection.runAndRead(sql, params as any[])
+      : await connection.runAndRead(sql)
+    await reader.readAll()
+    return reader.getRowObjectsJson().length
+  }
+
+  let appliedThreads: string | undefined
+  if (threadsParam && /^\d+$/.test(threadsParam)) {
+    try {
+      await connection.run(`SET threads=${Number(threadsParam)}`)
+      appliedThreads = threadsParam
+    } catch {
+      /* ignore */
+    }
+  }
+
+  const probes: ProbeResult[] = []
+
+  probes.push(
+    await probe("version_threads", async () =>
+      run("SELECT version() AS v, current_setting('threads') AS threads"),
+    ),
+  )
+
+  probes.push(
+    await probe("count_rows", async () =>
+      run(`SELECT count(*) AS n FROM ${source} WHERE model_key = ?`, [modelKey]),
+    ),
+  )
+
+  probes.push(
+    await probe("scan_primitives", async () =>
+      run(
+        `SELECT evaluation_id, score, model_key FROM ${source} WHERE model_key = ? AND score IS NOT NULL`,
+        [modelKey],
+      ),
+    ),
+  )
+
+  // The decisive probe: one to_json column at a time, so we learn exactly
+  // which nested column's decode trips the Thrift "don't know what type:".
+  for (const col of NESTED_COLUMNS) {
+    probes.push(
+      await probe(`tojson_${col}`, async () =>
+        run(
+          `SELECT CAST(to_json(${col}) AS VARCHAR) AS x FROM ${source} WHERE model_key = ? AND score IS NOT NULL`,
+          [modelKey],
+        ),
+      ),
+    )
+  }
+
+  // Same projection but reading every row (no model filter), to learn
+  // whether the bad page is confined to this model's row group.
+  probes.push(
+    await probe(
+      "tojson_evalcards_all_rows",
+      async () =>
+        run(
+          `SELECT CAST(to_json(evalcards_annotations) AS VARCHAR) AS x FROM ${source} WHERE score IS NOT NULL`,
+        ),
+      "all rows, not just this model",
+    ),
+  )
+
+  // Reset threads to engine default so the diagnostic doesn't leave the
+  // shared connection in a weird state for normal traffic.
+  if (appliedThreads) {
+    try {
+      await connection.run("RESET threads")
+    } catch {
+      /* ignore */
+    }
+  }
+
+  return NextResponse.json({
+    modelKey,
+    snapshotUrl,
+    source: useHttp ? "remote-httpfs" : "local-cache-view",
+    appliedThreads: appliedThreads ?? "(default)",
+    probes,
+  })
+}

package.json

@@ -17,7 +17,7 @@
     "audit-adapters": "tsx scripts/audit-adapters.mjs"
   },
   "dependencies": {
-    "@duckdb/node-api": "1.5.2-r.1",
+    "@duckdb/node-api": "1.5.3-r.2",
     "@hookform/resolvers": "^3.10.0",
     "@radix-ui/react-accordion": "1.2.2",
     "@radix-ui/react-alert-dialog": "1.1.4",

pnpm-lock.yaml

@@ -9,8 +9,8 @@ importers:
   .:
     dependencies:
       '@duckdb/node-api':
-        specifier: 1.5.2-r.1
-        version: 1.5.2-r.1
+        specifier: 1.5.3-r.2
+        version: 1.5.3-r.2
       '@hookform/resolvers':
         specifier: ^3.10.0
         version: 3.10.0(react-hook-form@7.62.0(react@19.1.1))
@@ -211,41 +211,51 @@ packages:
   '@date-fns/tz@1.2.0':
     resolution: {integrity: sha512-LBrd7MiJZ9McsOgxqWX7AaxrDjcFVjWH/tIKJd7pnR7McaslGYOP1QmmiBXdJH/H/yLCT+rcQ7FaPBUxRGUtrg==}
 
-  '@duckdb/node-api@1.5.2-r.1':
-    resolution: {integrity: sha512-OzBBnS0JGXMoS5mzKNY/Ylr7SshcRQiLFIoxQ4AlePwJ2fNeDL/fbHu/knjxUrXwW1fJBTUgwWftmxDdnZZb3A==}
+  '@duckdb/node-api@1.5.3-r.2':
+    resolution: {integrity: sha512-xjn6W9n8Sn9PaYmzrB6ayidNLQHpqNKdaRz4gx3yVoLUkrsIebQC1M1jV3BR8KqXj5P0RX4wLs8Tg/fK2ZQmNA==}
 
-  '@duckdb/node-bindings-darwin-arm64@1.5.2-r.1':
-    resolution: {integrity: sha512-v35FyKOb8EJCvaiPF7k0gvKiJTXR7PPQDNoWR0Gu+YSX5O9b+DIguzt1348Of3HebHy6ATSMzlUekaVA9YXu+g==}
+  '@duckdb/node-bindings-darwin-arm64@1.5.3-r.2':
+    resolution: {integrity: sha512-5RQ3+fUZn+OvIJEfa1i/KngJHo5d6ilpjunsHfHdQy5F7r5pewDkPxRQno4xInPqGukMibKDoMdpzCKnymFJtw==}
     cpu: [arm64]
     os: [darwin]
 
-  '@duckdb/node-bindings-darwin-x64@1.5.2-r.1':
-    resolution: {integrity: sha512-SU9dIJ1BluKkkGxi4UsP4keqkkstB2YDySF9KcYu3EZKIVM3FTv2zc7XO38dXnHOq6+F3WqhWWZvD+XU945p7A==}
+  '@duckdb/node-bindings-darwin-x64@1.5.3-r.2':
+    resolution: {integrity: sha512-E7Pc+do8kj82+hHXkp4mxfw9Kqfe6V5FUT2EMoRsacbvhuoM1AeZxOj9jknh23IoNxPtn3S2tKnD1krL0v0I7A==}
     cpu: [x64]
     os: [darwin]
 
-  '@duckdb/node-bindings-linux-arm64@1.5.2-r.1':
-    resolution: {integrity: sha512-3Tra9xM3aM3denaER4KhJ6//6PpmPbik9ECBQ+sh9PyKaEgHw/0kAcKnLm5EzWUnXF0qYmZlewvkCrse8KmOYw==}
+  '@duckdb/node-bindings-linux-arm64-musl@1.5.3-r.2':
+    resolution: {integrity: sha512-P9lTyML8M6pS300fJbnMNFSBhojC0xJcvGvuR44xd//NqaHk9EwcPIs4+r2ek4LP5uugm7konVZ3VMr0BOKujw==}
     cpu: [arm64]
     os: [linux]
 
-  '@duckdb/node-bindings-linux-x64@1.5.2-r.1':
-    resolution: {integrity: sha512-pcQvZRHiIfJ9cq8parkSQczQHEml/IeGfnDCMAbEgD6+jaV9Y9Y5Ph1kP9aR+bm6him1S5ZIEr3kZbihjKnWbA==}
+  '@duckdb/node-bindings-linux-arm64@1.5.3-r.2':
+    resolution: {integrity: sha512-R8RKkNgVL4kHodot1krfbB4XQ+bA90LPsnN1Lm5J03QJlFgNlezul0H2U1UuvEaKUBPWDQj8g9Yt4hhKKKX0zw==}
+    cpu: [arm64]
+    os: [linux]
+
+  '@duckdb/node-bindings-linux-x64-musl@1.5.3-r.2':
+    resolution: {integrity: sha512-V9kPvbMoJENvPo82LOzfhZ+KilFe1K8TKrksFmHe1Sy2pduNUVnxC+T4OQuOu4sqyCCVqVYRXe3pqTzKa/hCYg==}
     cpu: [x64]
     os: [linux]
 
-  '@duckdb/node-bindings-win32-arm64@1.5.2-r.1':
-    resolution: {integrity: sha512-Ji8tym+N3LkrhVt0Up3bsacD/kpg4/JXFJQqxswiYvBaNCQOk+D+aiVS0GN5pcqvmnG7V7TpsDRzkLEFaWp1vw==}
+  '@duckdb/node-bindings-linux-x64@1.5.3-r.2':
+    resolution: {integrity: sha512-6vYTXgIqnWEjlNZX/AveTU7EzicuGATOdy8ncUbBIDxoC9cIrjoIqQc37COKEyT5IHjLIt25MllJtags96vvxA==}
+    cpu: [x64]
+    os: [linux]
+
+  '@duckdb/node-bindings-win32-arm64@1.5.3-r.2':
+    resolution: {integrity: sha512-CjKqTI0WfRxwLgz8FV4fej/wkijPQnRgDk93aQlWQ+/+EwB9EnwQW3yfUy4Rs8cFjQkxkwlL5azoSR973l8U6w==}
     cpu: [arm64]
     os: [win32]
 
-  '@duckdb/node-bindings-win32-x64@1.5.2-r.1':
-    resolution: {integrity: sha512-5XqcqC+4R8ghBEEbnc2a0sqfz1zyPBRb9YcmIWfiuDoCYSYFbKhmHcEyNftZDHcwCoLOHXnUin45jraex4STqQ==}
+  '@duckdb/node-bindings-win32-x64@1.5.3-r.2':
+    resolution: {integrity: sha512-/oxN8ZFf2LxhForL+Uc+lEqnzHNvhSEsu3blnk3zu73rKTbD/WvPYIecrenb1ysApRK9ElGxitIYFx8PBuVKxg==}
     cpu: [x64]
     os: [win32]
 
-  '@duckdb/node-bindings@1.5.2-r.1':
-    resolution: {integrity: sha512-bUg3bLVj70YVku6fKyQJS8ASORl7kM7YFVFznsEB9pWbtazPj+ME2x2FUk0WiTzjJdutjzSSGXF066mB4bGGZA==}
+  '@duckdb/node-bindings@1.5.3-r.2':
+    resolution: {integrity: sha512-bveKx+LfqRazcH8/MuCkMsbLVQKqr/EyuYRzaKpBcEvRYenw/TqK5+1N4fkMyoTv8ZukA4UYobymNYHqJUpdOA==}
 
   '@emnapi/runtime@1.4.5':
     resolution: {integrity: sha512-++LApOtY0pEEz1zrd9vy1/zXVaVJJ/EbAF3u0fXIzPJEDtnITsBGbbK0EkM72amhl/R5b+5xx0Y/QhcVOpuulg==}
@@ -1946,6 +1956,10 @@ packages:
     resolution: {integrity: sha512-3UDv+G9CsCKO1WKMGw9fwq/SWJYbI0c5Y7LU1AXYoDdbhE2AHQ6N6Nb34sG8Fj7T5APy8qXDCKuuIHd1BR0tVA==}
     engines: {node: '>=8'}
 
+  detect-libc@2.1.2:
+    resolution: {integrity: sha512-Btj2BOOO83o3WyH59e8MgXsxEQVcarkUOpEYrubB0urwnN10yQ364rsiByU11nZlqWYZm05i/of7io4mzihBtQ==}
+    engines: {node: '>=8'}
+
   detect-node-es@1.1.0:
     resolution: {integrity: sha512-ypdmJU/TbBby2Dxibuv7ZLW3Bs1QEmM7nHjEANfohJLvE0XVujisn1qPJcZxg+qDucsr+bP6fLD1rPS3AhJ7EQ==}
 
@@ -2821,36 +2835,46 @@ snapshots:
 
   '@date-fns/tz@1.2.0': {}
 
-  '@duckdb/node-api@1.5.2-r.1':
+  '@duckdb/node-api@1.5.3-r.2':
     dependencies:
-      '@duckdb/node-bindings': 1.5.2-r.1
+      '@duckdb/node-bindings': 1.5.3-r.2
+
+  '@duckdb/node-bindings-darwin-arm64@1.5.3-r.2':
+    optional: true
 
-  '@duckdb/node-bindings-darwin-arm64@1.5.2-r.1':
+  '@duckdb/node-bindings-darwin-x64@1.5.3-r.2':
     optional: true
 
-  '@duckdb/node-bindings-darwin-x64@1.5.2-r.1':
+  '@duckdb/node-bindings-linux-arm64-musl@1.5.3-r.2':
     optional: true
 
-  '@duckdb/node-bindings-linux-arm64@1.5.2-r.1':
+  '@duckdb/node-bindings-linux-arm64@1.5.3-r.2':
     optional: true
 
-  '@duckdb/node-bindings-linux-x64@1.5.2-r.1':
+  '@duckdb/node-bindings-linux-x64-musl@1.5.3-r.2':
     optional: true
 
-  '@duckdb/node-bindings-win32-arm64@1.5.2-r.1':
+  '@duckdb/node-bindings-linux-x64@1.5.3-r.2':
     optional: true
 
-  '@duckdb/node-bindings-win32-x64@1.5.2-r.1':
+  '@duckdb/node-bindings-win32-arm64@1.5.3-r.2':
     optional: true
 
-  '@duckdb/node-bindings@1.5.2-r.1':
+  '@duckdb/node-bindings-win32-x64@1.5.3-r.2':
+    optional: true
+
+  '@duckdb/node-bindings@1.5.3-r.2':
+    dependencies:
+      detect-libc: 2.1.2
     optionalDependencies:
-      '@duckdb/node-bindings-darwin-arm64': 1.5.2-r.1
-      '@duckdb/node-bindings-darwin-x64': 1.5.2-r.1
-      '@duckdb/node-bindings-linux-arm64': 1.5.2-r.1
-      '@duckdb/node-bindings-linux-x64': 1.5.2-r.1
-      '@duckdb/node-bindings-win32-arm64': 1.5.2-r.1
-      '@duckdb/node-bindings-win32-x64': 1.5.2-r.1
+      '@duckdb/node-bindings-darwin-arm64': 1.5.3-r.2
+      '@duckdb/node-bindings-darwin-x64': 1.5.3-r.2
+      '@duckdb/node-bindings-linux-arm64': 1.5.3-r.2
+      '@duckdb/node-bindings-linux-arm64-musl': 1.5.3-r.2
+      '@duckdb/node-bindings-linux-x64': 1.5.3-r.2
+      '@duckdb/node-bindings-linux-x64-musl': 1.5.3-r.2
+      '@duckdb/node-bindings-win32-arm64': 1.5.3-r.2
+      '@duckdb/node-bindings-win32-x64': 1.5.3-r.2
 
   '@emnapi/runtime@1.4.5':
     dependencies:
@@ -4323,6 +4347,8 @@ snapshots:
 
   detect-libc@2.0.4: {}
 
+  detect-libc@2.1.2: {}
+
   detect-node-es@1.1.0: {}
 
   devtools-protocol@0.0.1595872: {}