fix(embed): extend #1717 model labeling to the embed-backfill stale path

src/core/embed-stale.ts (used by the embed-backfill minion handler) built its merged ChunkInput without a model field, so every chunk on a touched page — re-embedded AND preserved — was relabeled to the engine default on each backfill pass. Mirror the embed.ts semantics: stamp the resolved gateway label on re-embedded chunks, carry the existing label on untouched ones. Pinned by a PGLite test that fails without the fix. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
fix(embed): label content_chunks.model with the model that produced the vector (#1717 )
2026-08-16 01:42:23 +00:00 · 2026-07-22 11:06:16 -07:00 · 2026-07-21 14:42:17 -07:00
18 changed files with 205 additions and 328 deletions
@@ -7,7 +7,7 @@
  <link rel="preconnect" href="https://fonts.googleapis.com" />
  <link rel="preconnect" href="https://fonts.gstatic.com" crossorigin />
  <link href="https://fonts.googleapis.com/css2?family=Inter:wght@400;500;600&family=JetBrains+Mono:wght@400;500&display=swap" rel="stylesheet" />
-  <script type="module" crossorigin src="/admin/assets/index-CzLjRij_.js"></script>
+  <script type="module" crossorigin src="/admin/assets/index-CoGEje3-.js"></script>
  <link rel="stylesheet" crossorigin href="/admin/assets/index-GxkWX7v3.css">
 </head>
 <body>
@@ -39,7 +39,6 @@ export const api = {
  stats: () => apiFetch('/admin/api/stats'),
  health: () => apiFetch('/admin/api/health-indicators'),
  agents: () => apiFetch('/admin/api/agents'),
-  sources: () => apiFetch('/admin/api/sources'),
  requests: (page = 1, qs = '') => apiFetch(`/admin/api/requests?page=${page}${qs}`),
  apiKeys: () => apiFetch('/admin/api/api-keys'),
  createApiKey: (name: string) => apiFetch('/admin/api/api-keys', { method: 'POST', body: JSON.stringify({ name }) }),
@@ -23,16 +23,9 @@ interface Agent {
  total_requests: number;
  requests_today: number;
  token_ttl: number | null;
-  source_id: string | null;
-  federated_read: string[] | null;
  status: 'active' | 'revoked';
 }

-interface SourceOption {
-  id: string;
-  name: string | null;
-}
-
 interface ApiKey {
  id: string;
  name: string;
@@ -95,7 +88,6 @@ export function AgentsPage() {
                <th>Name</th>
                <th>Type</th>
                <th>Scopes</th>
-                <th>Source</th>
                <th>Status</th>
                <th>Requests</th>
                <th>Last Used</th>
@@ -116,9 +108,6 @@ export function AgentsPage() {
                      <span key={s} className={`badge badge-${s}`} style={{ marginRight: 4 }}>{s}</span>
                    ))}
                  </td>
-                  <td className="mono" style={{ fontSize: 12, color: 'var(--text-secondary)' }}>
-                    {a.auth_type === 'oauth' ? (a.source_id || 'default') : 'all'}
-                  </td>
                  <td>
                    <span className={`badge ${a.status === 'active' ? 'badge-success' : 'badge-danger'}`}>{a.status}</span>
                  </td>
@@ -268,29 +257,9 @@ function RegisterModal({ onClose, onRegistered }: {
    Object.fromEntries(ALLOWED_SCOPES_LIST.map(s => [s, s === 'read'])) as Record<Scope, boolean>,
  );
  const [ttl, setTtl] = useState('86400'); // 24h default
-  const [redirectUris, setRedirectUris] = useState('');
-  const [sources, setSources] = useState<SourceOption[]>([{ id: 'default', name: 'default' }]);
-  const [sourceId, setSourceId] = useState('default');
-  const [readSources, setReadSources] = useState<Record<string, boolean>>({ default: true });
  const [loading, setLoading] = useState(false);
  const [error, setError] = useState('');

-  useEffect(() => {
-    api.sources()
-      .then((rows: SourceOption[]) => {
-        const nextSources = rows.length > 0 ? rows : [{ id: 'default', name: 'default' }];
-        const nextSourceId = nextSources.some(s => s.id === sourceId) ? sourceId : nextSources[0].id;
-        setSources(nextSources);
-        setSourceId(nextSourceId);
-        setReadSources(prev => {
-          const next = Object.fromEntries(nextSources.map(s => [s.id, Boolean(prev[s.id])])) as Record<string, boolean>;
-          next[nextSourceId] = true;
-          return next;
-        });
-      })
-      .catch(() => {});
-  }, []);
-
  const ttlOptions = [
    { label: '1 hour', value: '3600' },
    { label: '24 hours', value: '86400' },
@@ -308,24 +277,11 @@ function RegisterModal({ onClose, onRegistered }: {
    try {
      // Use the CLI registration endpoint (POST to admin API)
      const selectedScopes = Object.entries(scopes).filter(([, v]) => v).map(([k]) => k).join(' ');
-      const federatedRead = sources.map(s => s.id).filter(id => readSources[id]);
-      if (federatedRead.length === 0) { setError('Select at least one read source'); setLoading(false); return; }
-      // #1036: redirect URIs, one per line. Mirrors the CLI convention
-      // (src/commands/auth.ts): presence of redirect URIs implies
-      // authorization_code + refresh_token grants unless the user had none.
-      const uris = redirectUris.split('\n').map(u => u.trim()).filter(Boolean);
      const res = await fetch('/admin/api/register-client', {
        method: 'POST',
        credentials: 'same-origin',
        headers: { 'Content-Type': 'application/json' },
-        body: JSON.stringify({
-          name: name.trim(),
-          scopes: selectedScopes,
-          tokenTtl: ttl === '0' ? 315360000 : Number(ttl),
-          source_id: sourceId,
-          federated_read: federatedRead,
-          ...(uris.length > 0 ? { redirectUris: uris, grantTypes: ['authorization_code', 'refresh_token'] } : {}),
-        }),
+        body: JSON.stringify({ name: name.trim(), scopes: selectedScopes, tokenTtl: ttl === '0' ? 315360000 : Number(ttl) }),
      });
      if (!res.ok) throw new Error('Registration failed');
      const data = await res.json();
@@ -356,42 +312,6 @@ function RegisterModal({ onClose, onRegistered }: {
            ))}
          </div>
        </div>
-        <div style={{ marginBottom: 16 }}>
-          <label>Write Source</label>
-          <select value={sourceId} onChange={e => {
-            const id = e.target.value;
-            setSourceId(id);
-            setReadSources(p => ({ ...p, [id]: true }));
-          }}
-            style={{ width: '100%', background: 'var(--bg-secondary)', color: 'var(--text-primary)', border: '1px solid var(--border)', borderRadius: 6, padding: '6px 10px', fontSize: 14 }}>
-            {sources.map(s => <option key={s.id} value={s.id}>{s.name && s.name !== s.id ? `${s.id} (${s.name})` : s.id}</option>)}
-          </select>
-        </div>
-        <div style={{ marginBottom: 16 }}>
-          <label>Read Sources</label>
-          <div className="checkbox-group">
-            {sources.map(s => (
-              <label key={s.id} className="checkbox-label">
-                <input
-                  type="checkbox"
-                  checked={Boolean(readSources[s.id])}
-                  onChange={e => setReadSources(p => ({ ...p, [s.id]: e.target.checked }))}
-                />
-                {s.name && s.name !== s.id ? `${s.id} (${s.name})` : s.id}
-              </label>
-            ))}
-          </div>
-        </div>
-        <div style={{ marginBottom: 16 }}>
-          <label>Redirect URIs <span style={{ color: 'var(--text-secondary)', fontWeight: 400 }}>(optional, one per line — enables authorization_code flow)</span></label>
-          <textarea
-            placeholder={'https://example.com/callback'}
-            value={redirectUris}
-            onChange={e => setRedirectUris(e.target.value)}
-            rows={2}
-            style={{ width: '100%', background: 'var(--bg-secondary)', color: 'var(--text-primary)', border: '1px solid var(--border)', borderRadius: 6, padding: '6px 10px', fontSize: 13, fontFamily: 'inherit', resize: 'vertical' }}
-          />
-        </div>
        <div style={{ marginBottom: 20 }}>
          <label>Token Lifetime</label>
          <select value={ttl} onChange={e => setTtl(e.target.value)}
@@ -608,8 +528,6 @@ function AgentDrawer({ agent, onClose, onRevoked }: { agent: Agent; onClose: ()
      client_name: agentName,
      auth_type: agent.auth_type,
      scope: agent.scope,
-      source_id: agent.source_id,
-      federated_read: agent.federated_read,
    }, null, 2),
  };

@@ -629,10 +547,6 @@ function AgentDrawer({ agent, onClose, onRevoked }: { agent: Agent; onClose: ()
          <span>{(agent.scope || '').split(' ').filter(Boolean).map(s => (
            <span key={s} className={`badge badge-${s}`} style={{ marginRight: 4 }}>{s}</span>
          ))}</span>
-          <span style={{ color: 'var(--text-secondary)' }}>Write Source</span>
-          <span className="mono">{agent.source_id || (isOAuth ? 'default' : 'all')}</span>
-          <span style={{ color: 'var(--text-secondary)' }}>Read Sources</span>
-          <span className="mono">{isOAuth ? (agent.federated_read && agent.federated_read.length > 0 ? agent.federated_read.join(', ') : (agent.source_id || 'default')) : 'all'}</span>
          <span style={{ color: 'var(--text-secondary)' }}>Registered</span>
          <span>{new Date(agent.created_at).toLocaleDateString()}</span>
          <span style={{ color: 'var(--text-secondary)' }}>Token TTL</span>
@@ -1,13 +1,13 @@
 // AUTO-GENERATED — do not edit by hand.
 // Run `bun run scripts/build-admin-embedded.ts` to regenerate.
-// Source: admin/dist/ at 2026-07-21.
+// Source: admin/dist/ at 2026-05-27.
 //
 // Bun resolves the file: imports to a path that works at runtime even
 // inside a compiled binary (`bun build --compile`). The manifest maps
 // the request path the express handler sees to (resolved-path, mime).

 // @ts-ignore — type: 'file' is Bun ESM, not in lib.d.ts
-import A_0_assets_index_CzLjRij__js from '../admin/dist/assets/index-CzLjRij_.js' with { type: 'file' };
+import A_0_assets_index_CoGEje3__js from '../admin/dist/assets/index-CoGEje3-.js' with { type: 'file' };
 // @ts-ignore — type: 'file' is Bun ESM, not in lib.d.ts
 import A_1_assets_index_GxkWX7v3_css from '../admin/dist/assets/index-GxkWX7v3.css' with { type: 'file' };
 // @ts-ignore — type: 'file' is Bun ESM, not in lib.d.ts
@@ -19,7 +19,7 @@ export interface AdminAsset {
 }

 export const ADMIN_ASSETS: Record<string, AdminAsset> = {
-  "/admin/assets/index-CzLjRij_.js": { path: A_0_assets_index_CzLjRij__js as unknown as string, mime: "application/javascript; charset=utf-8" },
+  "/admin/assets/index-CoGEje3-.js": { path: A_0_assets_index_CoGEje3__js as unknown as string, mime: "application/javascript; charset=utf-8" },
  "/admin/assets/index-GxkWX7v3.css": { path: A_1_assets_index_GxkWX7v3_css as unknown as string, mime: "text/css; charset=utf-8" },
  "/admin/index.html": { path: A_2_index_html as unknown as string, mime: "text/html; charset=utf-8" },
 };
@@ -1,5 +1,5 @@
 import type { BrainEngine } from '../core/engine.ts';
-import { embedBatch, currentEmbeddingSignature } from '../core/embedding.ts';
+import { embedBatch, currentEmbeddingSignature, resolveEmbeddingModelLabel } from '../core/embedding.ts';
 import type { ChunkInput } from '../core/types.ts';
 import { chunkText } from '../core/chunkers/recursive.ts';
 import { createProgress, type ProgressReporter } from '../core/progress.ts';
@@ -581,11 +581,16 @@ async function embedPage(
  for (let j = 0; j < toEmbed.length; j++) {
    embeddingMap.set(toEmbed[j].chunk_index, embeddings[j]);
  }
+  // #1717: label each (re)embedded chunk with the model that actually
+  // produced its vector. Preserved chunks (not re-embedded this pass) keep
+  // their existing model so a mixed-model page isn't relabeled wholesale.
+  const embedModelLabel = resolveEmbeddingModelLabel();
  const updated: ChunkInput[] = chunks.map(c => ({
    chunk_index: c.chunk_index,
    chunk_text: c.chunk_text,
    chunk_source: c.chunk_source,
    embedding: embeddingMap.get(c.chunk_index),
+    model: embeddingMap.has(c.chunk_index) && embedModelLabel ? embedModelLabel : c.model,
    token_count: c.token_count || Math.ceil(c.chunk_text.length / 4),
  }));

@@ -717,12 +722,16 @@ async function embedAll(
      for (let j = 0; j < toEmbed.length; j++) {
        embeddingMap.set(toEmbed[j].chunk_index, embeddings[j]);
      }
+      // #1717: stamp the resolved embedding model on (re)embedded chunks;
+      // preserve the existing model on chunks left untouched.
+      const embedModelLabel = resolveEmbeddingModelLabel();
      // Preserve ALL chunks, only update embeddings for stale ones
      const updated: ChunkInput[] = chunks.map(c => ({
        chunk_index: c.chunk_index,
        chunk_text: c.chunk_text,
        chunk_source: c.chunk_source,
        embedding: embeddingMap.get(c.chunk_index) ?? undefined,
+        model: embeddingMap.has(c.chunk_index) && embedModelLabel ? embedModelLabel : c.model,
        token_count: c.token_count || Math.ceil(c.chunk_text.length / 4),
      }));
      await observed(pacer, () => engine.upsertChunks(page.slug, updated, pageOpts));
@@ -1012,11 +1021,15 @@ async function embedAllStale(
          for (let j = 0; j < stale.length; j++) {
            staleIdxToEmbedding.set(stale[j].chunk_index, embeddings[j]);
          }
+          // #1717: label the re-embedded (stale) chunks with the resolved
+          // model; preserve the existing model on the non-stale chunks.
+          const embedModelLabel = resolveEmbeddingModelLabel();
          const merged: ChunkInput[] = existing.map(c => ({
            chunk_index: c.chunk_index,
            chunk_text: c.chunk_text,
            chunk_source: c.chunk_source,
            embedding: staleIdxToEmbedding.get(c.chunk_index) ?? undefined,
+            model: staleIdxToEmbedding.has(c.chunk_index) && embedModelLabel ? embedModelLabel : c.model,
            token_count: c.token_count || Math.ceil(c.chunk_text.length / 4),
          }));
          await observed(pacer, () => engine.upsertChunks(slug, merged, { sourceId: keySourceId }));
@@ -39,7 +39,6 @@ import * as db from '../core/db.ts';
 import { sqlQueryForEngine, executeRawJsonb } from '../core/sql-query.ts';
 import { MinionQueue } from '../core/minions/queue.ts';
 import { isRetryableError } from '../core/retry-matcher.ts';
-import { normalizeAdminClientSourceScope } from '../core/admin-source-scope.ts';
 import {
  computeContentHash,
  validateIngestionEvent,
@@ -1087,7 +1086,6 @@ export async function runServeHttp(engine: BrainEngine, options: ServeHttpOption
      const oauthClients = await sql`
        SELECT c.client_id as id, c.client_name as name, 'oauth' as auth_type,
          c.grant_types, c.scope, c.created_at, c.token_ttl,
-          c.source_id, c.federated_read,
          CASE WHEN c.deleted_at IS NOT NULL THEN 'revoked' ELSE 'active' END as status,
          (SELECT max(created_at) FROM mcp_request_log WHERE token_name = c.client_id) as last_used_at,
          (SELECT count(*)::int FROM mcp_request_log WHERE token_name = c.client_id) as total_requests,
@@ -1097,7 +1095,6 @@ export async function runServeHttp(engine: BrainEngine, options: ServeHttpOption
      const legacyKeys = await sql`
        SELECT a.id, a.name, 'api_key' as auth_type,
          '{"bearer"}' as grant_types, 'read write admin' as scope, a.created_at, null as token_ttl,
-          null as source_id, null as federated_read,
          CASE WHEN a.revoked_at IS NOT NULL THEN 'revoked' ELSE 'active' END as status,
          a.last_used_at,
          (SELECT count(*)::int FROM mcp_request_log WHERE token_name = a.name) as total_requests,
@@ -1110,17 +1107,6 @@ export async function runServeHttp(engine: BrainEngine, options: ServeHttpOption
    }
  });

-  app.get('/admin/api/sources', requireAdmin, async (_req: Request, res: Response) => {
-    try {
-      const sources = await engine.listAllSources({ includeArchived: false });
-      const options = sources.map(s => ({ id: s.id, name: s.name }));
-      if (!options.some(s => s.id === 'default')) options.unshift({ id: 'default', name: 'default' });
-      res.json(options);
-    } catch (e) {
-      res.status(503).json({ error: 'service_unavailable' });
-    }
-  });
-
  // v0.38 Slice 4 — per-OAuth-client agent spend viewer. Pre-computes today's
  // spend (committed + pending reservations) per client so the Agents tab
  // can render a "$X / $Y today" cell. Read-side endpoint only — no mutation.
@@ -1454,8 +1440,6 @@ export async function runServeHttp(engine: BrainEngine, options: ServeHttpOption
      // missing, empty) and rejects the rest with a structured 400.
      const { name, tokenTtl, grantTypes, redirectUris, tokenEndpointAuthMethod } = req.body;
      const rawScopes = (req.body as Record<string, unknown>).scopes ?? (req.body as Record<string, unknown>).scope;
-      const rawSourceId = (req.body as Record<string, unknown>).source_id ?? (req.body as Record<string, unknown>).sourceId;
-      const rawFederatedRead = (req.body as Record<string, unknown>).federated_read ?? (req.body as Record<string, unknown>).federatedRead;
      if (!name) { res.status(400).json({ error: 'Name required' }); return; }
      let scopeString: string;
      try {
@@ -1486,23 +1470,8 @@ export async function runServeHttp(engine: BrainEngine, options: ServeHttpOption
        });
        return;
      }
-      let sourceScope: ReturnType<typeof normalizeAdminClientSourceScope>;
-      try {
-        const sources = await engine.listAllSources({ includeArchived: false });
-        sourceScope = normalizeAdminClientSourceScope(
-          rawSourceId,
-          rawFederatedRead,
-          sources.map(s => s.id),
-        );
-      } catch (e) {
-        res.status(400).json({
-          error: 'invalid_source_scope',
-          message: e instanceof Error ? e.message : String(e),
-        });
-        return;
-      }
      const result = await oauthProvider.registerClientManual(
-        name, grants, scopeString, uris, sourceScope.sourceId, sourceScope.federatedRead, validatedAuthMethod,
+        name, grants, scopeString, uris, 'default', undefined, validatedAuthMethod,
      );
      // Set per-client TTL if specified
      if (tokenTtl && Number(tokenTtl) > 0) {
@@ -1,51 +0,0 @@
-import { isValidSourceId } from './source-id.ts';
-
-export interface AdminClientSourceScope {
-  sourceId: string;
-  federatedRead: string[];
-}
-
-function parseFederatedReadInput(raw: unknown, sourceId: string): string[] {
-  if (raw === undefined || raw === null) return [sourceId];
-  if (typeof raw === 'string') {
-    const values = raw.split(',').map(s => s.trim()).filter(Boolean);
-    return values.length > 0 ? values : [sourceId];
-  }
-  if (Array.isArray(raw)) {
-    if (!raw.every(v => typeof v === 'string')) {
-      throw new Error('federated_read must be an array of source ids');
-    }
-    const values = raw.map(v => v.trim()).filter(Boolean);
-    return values.length > 0 ? values : [sourceId];
-  }
-  throw new Error('federated_read must be a string or array');
-}
-
-export function normalizeAdminClientSourceScope(
-  rawSourceId: unknown,
-  rawFederatedRead: unknown,
-  availableSourceIds: string[],
-): AdminClientSourceScope {
-  const knownSources = new Set(availableSourceIds);
-  const fallbackSourceId = knownSources.has('default') ? 'default' : (availableSourceIds[0] ?? 'default');
-  const sourceId = rawSourceId === undefined || rawSourceId === null || rawSourceId === ''
-    ? fallbackSourceId
-    : rawSourceId;
-  if (!isValidSourceId(sourceId)) {
-    throw new Error(`Invalid source_id: ${JSON.stringify(sourceId)}`);
-  }
-  if (knownSources.size > 0 && !knownSources.has(sourceId)) {
-    throw new Error(`Unknown source_id: ${sourceId}`);
-  }
-
-  const federatedRead = Array.from(new Set(parseFederatedReadInput(rawFederatedRead, sourceId)));
-  for (const id of federatedRead) {
-    if (!isValidSourceId(id)) {
-      throw new Error(`Invalid federated_read source_id: ${JSON.stringify(id)}`);
-    }
-    if (knownSources.size > 0 && !knownSources.has(id)) {
-      throw new Error(`Unknown federated_read source_id: ${id}`);
-    }
-  }
-  return { sourceId, federatedRead };
-}
@@ -20,6 +20,7 @@
 import type { BrainEngine } from './engine.ts';
 import type { ChunkInput } from './types.ts';
 import { embedBatchWithBackoff } from '../commands/embed.ts';
+import { resolveEmbeddingModelLabel } from './embedding.ts';
 import { type DbPacer, createNoopPacer, observed } from './db-pacer.ts';
 import { AbortError } from './abort-check.ts';

@@ -200,11 +201,17 @@ export async function embedStaleForSource(
        for (let j = 0; j < stale.length; j++) {
          staleIdxToEmbedding.set(stale[j].chunk_index, embeddings[j]);
        }
+        // #1717: label re-embedded chunks with the model that produced the
+        // vector; preserved chunks keep their existing model. Without this,
+        // upsertChunks falls back to DEFAULT_EMBEDDING_MODEL for every chunk
+        // (the same mislabel the embed.ts paths fixed).
+        const embedModelLabel = resolveEmbeddingModelLabel();
        const merged: ChunkInput[] = existing.map((c) => ({
          chunk_index: c.chunk_index,
          chunk_text: c.chunk_text,
          chunk_source: c.chunk_source,
          embedding: staleIdxToEmbedding.get(c.chunk_index) ?? undefined,
+          model: staleIdxToEmbedding.has(c.chunk_index) && embedModelLabel ? embedModelLabel : c.model,
          token_count: c.token_count || Math.ceil(c.chunk_text.length / 4),
          // Carry through per-chunk metadata. upsertChunks writes these as
          // EXCLUDED.<col> (not COALESCE), so omitting them here resets image
@@ -113,6 +113,21 @@ export async function embedBatch(
  return results;
 }

+/**
+ * Resolve the embedding model label (`provider:model`) to stamp onto
+ * `content_chunks.model`, so each chunk records the model that actually
+ * produced its vector instead of the engine's hardcoded default (#1717).
+ * Returns undefined if the gateway is unconfigured; callers then fall back
+ * to the chunk's existing model rather than mislabeling it.
+ */
+export function resolveEmbeddingModelLabel(): string | undefined {
+  try {
+    return gatewayGetModel();
+  } catch {
+    return undefined;
+  }
+}
+
 /** Currently-configured embedding model (short form without provider prefix). */
 export function getEmbeddingModelName(): string {
  return gatewayGetModel().split(':').slice(1).join(':') || 'text-embedding-3-large';
@@ -8,7 +8,7 @@ import { chunkText } from './chunkers/recursive.ts';
 import { chunkCodeText, chunkCodeTextFull, detectCodeLanguage, CHUNKER_VERSION } from './chunkers/code.ts';
 import { findChunkForOffset } from './chunkers/edge-extractor.ts';
 import { extractCodeRefs, imageOfCandidates } from './link-extraction.ts';
-import { embedBatch, embedMultimodal, currentEmbeddingSignature } from './embedding.ts';
+import { embedBatch, embedMultimodal, currentEmbeddingSignature, resolveEmbeddingModelLabel } from './embedding.ts';
 import { slugifyPath, slugifyCodePath, isCodeFilePath } from './sync.ts';
 import type { ChunkInput, PageInput, PageType } from './types.ts';
 import { computeEffectiveDate } from './effective-date.ts';
@@ -716,8 +716,12 @@ export async function importFromContent(
      ? chunks.map((c) => wrapChunkForEmbedding(c.chunk_text, prefix, c.chunk_source))
      : chunks.map((c) => c.chunk_text);
    const embeddings = await embedBatch(wrappedTexts);
+    // #1717: label each chunk with the model that actually produced its
+    // vector, not the engine's hardcoded default.
+    const embedModelLabel = resolveEmbeddingModelLabel();
    for (let i = 0; i < chunks.length; i++) {
      chunks[i].embedding = embeddings[i];
+      if (embedModelLabel) chunks[i].model = embedModelLabel;
      // token_count tracks the wrapped string length so cost reporting
      // reflects what we actually sent to the embedder.
      chunks[i].token_count = Math.ceil(wrappedTexts[i].length / 4);
@@ -1141,7 +1145,10 @@ export async function importCodeFile(
    const matched = existingByKey.get(key);
    if (matched && matched.embedding) {
      // Reuse the existing embedding verbatim. No API call, no cost.
+      // #1717: carry the existing model label along with the reused vector
+      // so the upsert doesn't relabel it with the engine default.
      chunks[i]!.embedding = matched.embedding as Float32Array;
+      chunks[i]!.model = matched.model ?? undefined;
      chunks[i]!.token_count = matched.token_count ?? undefined;
    } else {
      needsEmbedIndexes.push(i);
@@ -1153,9 +1160,12 @@ export async function importCodeFile(
    try {
      const textsToEmbed = needsEmbedIndexes.map((i) => chunks[i]!.chunk_text);
      const embeddings = await embedBatch(textsToEmbed);
+      // #1717: stamp the model that produced these vectors.
+      const embedModelLabel = resolveEmbeddingModelLabel();
      for (let j = 0; j < needsEmbedIndexes.length; j++) {
        const i = needsEmbedIndexes[j]!;
        chunks[i]!.embedding = embeddings[j]!;
+        if (embedModelLabel) chunks[i]!.model = embedModelLabel;
        chunks[i]!.token_count = Math.ceil(chunks[i]!.chunk_text.length / 4);
      }
    } catch (e: unknown) {
@@ -1,44 +0,0 @@
-import { describe, expect, test } from 'bun:test';
-import { normalizeAdminClientSourceScope } from '../src/core/admin-source-scope.ts';
-
-describe('admin register-client source scope normalization', () => {
-  const sources = ['default', 'team-a', 'team-b'];
-
-  test('defaults write and read scope to default source', () => {
-    expect(normalizeAdminClientSourceScope(undefined, undefined, sources)).toEqual({
-      sourceId: 'default',
-      federatedRead: ['default'],
-    });
-  });
-
-  test('accepts snake-case body fields with multiple read sources', () => {
-    expect(normalizeAdminClientSourceScope('team-a', ['team-a', 'team-b'], sources)).toEqual({
-      sourceId: 'team-a',
-      federatedRead: ['team-a', 'team-b'],
-    });
-  });
-
-  test('accepts comma-separated federated_read for API callers', () => {
-    expect(normalizeAdminClientSourceScope('team-a', 'team-b, default,team-b', sources)).toEqual({
-      sourceId: 'team-a',
-      federatedRead: ['team-b', 'default'],
-    });
-  });
-
-  test('falls back to selected write source when federated_read is empty', () => {
-    expect(normalizeAdminClientSourceScope('team-b', [], sources)).toEqual({
-      sourceId: 'team-b',
-      federatedRead: ['team-b'],
-    });
-  });
-
-  test('rejects malformed source ids before registration', () => {
-    expect(() => normalizeAdminClientSourceScope('../secret', undefined, sources)).toThrow(/Invalid source_id/);
-    expect(() => normalizeAdminClientSourceScope('team-a', ['snake_case'], sources)).toThrow(/Invalid federated_read/);
-  });
-
-  test('rejects valid-looking but unregistered source ids', () => {
-    expect(() => normalizeAdminClientSourceScope('ghost', undefined, sources)).toThrow(/Unknown source_id/);
-    expect(() => normalizeAdminClientSourceScope('team-a', ['ghost'], sources)).toThrow(/Unknown federated_read/);
-  });
-});
@@ -15,6 +15,7 @@ import { describe, test, expect, beforeAll, afterAll, beforeEach } from 'bun:tes
 import { PGLiteEngine } from '../src/core/pglite-engine.ts';
 import { resetPgliteState } from './helpers/reset-pglite.ts';
 import { embedStaleForSource } from '../src/core/embed-stale.ts';
+import { configureGateway, resetGateway } from '../src/core/ai/gateway.ts';
 import type { ChunkInput } from '../src/core/types.ts';

 let engine: PGLiteEngine;
@@ -276,4 +277,49 @@ describe('embedStaleForSource', () => {
    // The stale text row actually got its embedding.
    expect(txtRow.embedded_at).not.toBeNull();
  });
+
+  // #1717: the backfill path must label re-embedded chunks with the model
+  // that produced the vector, and preserve the existing label on chunks it
+  // did not touch (before the fix, both were reset to the engine default).
+  test('labels re-embedded chunks with the gateway model, preserves untouched labels (#1717)', async () => {
+    configureGateway({
+      embedding_model: 'openai:text-embedding-3-large',
+      env: { OPENAI_API_KEY: 'sk-test-embed-stale-1717' },
+    });
+    try {
+      await engine.putPage('notes/model-label', {
+        type: 'note',
+        title: 'model-label',
+        compiled_truth: '# model-label\n\nseeded',
+      });
+      await engine.upsertChunks('notes/model-label', [
+        {
+          chunk_index: 0,
+          chunk_text: 'already embedded elsewhere',
+          chunk_source: 'compiled_truth',
+          embedding: new Float32Array(1536).fill(0.01),
+          model: 'voyage:voyage-3',
+          token_count: 4,
+        },
+        {
+          chunk_index: 1,
+          chunk_text: 'stale chunk needing embed',
+          chunk_source: 'compiled_truth',
+          token_count: 5,
+          embedding: undefined, // stale
+        },
+      ]);
+
+      const result = await embedStaleForSource(engine, 'default', { embedFn: fakeEmbedFn });
+      expect(result.embedded).toBe(1);
+
+      const after = await engine.getChunks('notes/model-label');
+      const preserved = after.find((c) => c.chunk_index === 0)!;
+      const reembedded = after.find((c) => c.chunk_index === 1)!;
+      expect(reembedded.model).toBe('openai:text-embedding-3-large');
+      expect(preserved.model).toBe('voyage:voyage-3');
+    } finally {
+      resetGateway();
+    }
+  });
 });
@@ -37,6 +37,8 @@ mock.module('../src/core/embedding.ts', () => ({
  // setPageEmbeddingSignature / invalidateStaleSignatureEmbeddings resolve to
  // null via the Proxy default, so the signature value is inert here.
  currentEmbeddingSignature: () => 'test:model:1536',
+  // #1717: embed paths stamp this label on (re)embedded chunks.
+  resolveEmbeddingModelLabel: () => 'openai:text-embedding-3-large',
 }));

 // Import AFTER mocking.
@@ -803,3 +805,34 @@ describe('embedAllStale --source threading (D7)', () => {
    expect((firstCallOpts as { sourceId?: string }).sourceId).toBe('media-corpus');
  });
 });
+
+// #1717: content_chunks.model must record the model that actually produced
+// each vector, not the gateway/engine default.
+describe('content_chunks.model labeling (#1717)', () => {
+  test('stamps the resolved embedding model on re-embedded chunks, preserves it on untouched chunks', async () => {
+    let upserted: any[] | undefined;
+    // Chunk 0 is stale (no embedded_at) → gets re-embedded this pass.
+    // Chunk 1 is already embedded with a DIFFERENT model → must be preserved,
+    // not relabeled to the current model.
+    const chunks = [
+      { chunk_index: 0, chunk_text: 'a', chunk_source: 'compiled_truth', embedded_at: null, model: 'zeroentropyai:zembed-1', token_count: 1 },
+      { chunk_index: 1, chunk_text: 'b', chunk_source: 'compiled_truth', embedded_at: '2026-01-01', embedding: new Float32Array(1536), model: 'voyage:voyage-3', token_count: 1 },
+    ];
+    const engine = mockEngine({
+      getPage: async () => ({ slug: 'notes/x', compiled_truth: 'a', timeline: '', source_id: 'default' }),
+      getChunks: async () => chunks,
+      upsertChunks: async (_slug: string, c: any[]) => { upserted = c; },
+      setPageEmbeddingSignature: async () => null,
+    });
+
+    await runEmbedCore(engine, { slugs: ['notes/x'] });
+
+    expect(upserted).toBeDefined();
+    const byIdx = Object.fromEntries(upserted!.map(c => [c.chunk_index, c]));
+    // Re-embedded chunk carries the model that produced its vector (was
+    // mislabeled with the default before the fix).
+    expect(byIdx[0].model).toBe('openai:text-embedding-3-large');
+    // Untouched chunk keeps its original model — no wholesale relabel.
+    expect(byIdx[1].model).toBe('voyage:voyage-3');
+  });
+});
@@ -78,8 +78,6 @@ describe('v0.36.1.x #1077 — admin register-client supports PKCE public clients
    // (under either name) from req.body. Pin the fallback pattern so the
    // PKCE-fix regression contract stays load-bearing.
    expect(src).toMatch(/req\.body[^;]*scopes\s*\?\?\s*[^;]*scope\b/);
-    expect(src).toMatch(/req\.body[^;]*source_id\s*\?\?\s*[^;]*sourceId\b/);
-    expect(src).toMatch(/req\.body[^;]*federated_read\s*\?\?\s*[^;]*federatedRead\b/);
    // v0.41.3 (T4 atomicity fix, codex F4): admin endpoint now validates
    // tokenEndpointAuthMethod via the shared validator and passes it to
    // registerClientManual as a positional arg. Pre-v0.41.3 the route did
@@ -89,7 +87,6 @@ describe('v0.36.1.x #1077 — admin register-client supports PKCE public clients
    // UPDATE block (the regex deliberately asserts the post-insert UPDATE
    // is GONE).
    expect(src).toMatch(/validateTokenEndpointAuthMethod\(tokenEndpointAuthMethod\)/);
-    expect(src).toMatch(/normalizeAdminClientSourceScope\(/);
    expect(src).toMatch(/registerClientManual\([^)]*validatedAuthMethod[^)]*\)/);
    // Regression guard: post-insert UPDATE flipping client_secret_hash to
    // NULL based on a runtime check is exactly the non-atomic pattern T4
@@ -270,26 +267,3 @@ describe('v0.42.43.0 #2095 — volunteer-events sink + cycle purge wiring (struc
    expect(src).toMatch(/purged_volunteer_events_count/);
  });
 });
-
-describe('#1490 + #1036 — admin Register Agent form exposes source scope + redirect URIs', () => {
-  test('Agents.tsx RegisterModal sends source_id + federated_read in the POST body (#1490)', () => {
-    const src = readFileSync('admin/src/pages/Agents.tsx', 'utf8');
-    expect(src).toMatch(/source_id:\s*sourceId/);
-    expect(src).toMatch(/federated_read:\s*federatedRead/);
-    // The source list comes from the requireAdmin-gated endpoint, not a hardcoded 'default'.
-    expect(src).toMatch(/api\.sources\(\)/);
-  });
-
-  test('serve-http.ts serves /admin/api/sources behind requireAdmin (#1490)', () => {
-    const src = readFileSync('src/commands/serve-http.ts', 'utf8');
-    expect(src).toMatch(/app\.get\('\/admin\/api\/sources',\s*requireAdmin/);
-  });
-
-  test('Agents.tsx RegisterModal sends redirectUris + authorization_code grants when URIs are entered (#1036)', () => {
-    const src = readFileSync('admin/src/pages/Agents.tsx', 'utf8');
-    // One-per-line textarea split into an array...
-    expect(src).toMatch(/redirectUris\.split\('\\n'\)/);
-    // ...included in the body with the CLI's grant-type inference convention.
-    expect(src).toMatch(/redirectUris:\s*uris,\s*grantTypes:\s*\['authorization_code',\s*'refresh_token'\]/);
-  });
-});
@@ -73,4 +73,21 @@ describe('importFromContent embedding_signature stamping (F1)', () => {
    await importFromContent(engine, 'concepts/unstamped', '# Unstamped\n\nbody content.', { noEmbed: true });
    expect(await signatureOf('concepts/unstamped')).toBeNull();
  });
+
+  // #1717: content_chunks.model must record the model that produced the
+  // vector (the configured gateway model), not the engine's hardcoded
+  // default. The gateway here is configured to openai:text-embedding-3-large,
+  // which differs from DEFAULT_EMBEDDING_MODEL — so this fails without the
+  // import-path model stamping.
+  test('inline embed labels content_chunks.model with the configured model (#1717)', async () => {
+    await importFromContent(engine, 'concepts/labeled', '# Labeled\n\nsome body content to chunk and embed.', {});
+    const rows = await engine.executeRaw<{ model: string }>(
+      `SELECT cc.model FROM content_chunks cc
+        JOIN pages p ON p.id = cc.page_id
+       WHERE p.slug = $1 AND p.source_id = 'default'`,
+      ['concepts/labeled'],
+    );
+    expect(rows.length).toBeGreaterThan(0);
+    for (const r of rows) expect(r.model).toBe('openai:text-embedding-3-large');
+  });
 });
@@ -125,30 +125,6 @@ describe('client registration', () => {
    expect(client!.client_name).toBe('test-agent');
  });

-  test('registerClientManual persists source_id and federated_read into AuthInfo', async () => {
-    await sql`
-      INSERT INTO sources (id, name) VALUES (${'team-a'}, ${'Team A'}), (${'team-b'}, ${'Team B'})
-      ON CONFLICT DO NOTHING
-    `;
-    const { clientId, clientSecret } = await provider.registerClientManual(
-      'source-scoped-test',
-      ['client_credentials'],
-      'read write',
-      [],
-      'team-a',
-      ['team-a', 'team-b'],
-    );
-
-    const tokens = await provider.exchangeClientCredentials(clientId, clientSecret!, 'read');
-    const authInfo = await provider.verifyAccessToken(tokens.access_token) as {
-      sourceId?: string;
-      allowedSources?: string[];
-    };
-
-    expect(authInfo.sourceId).toBe('team-a');
-    expect(authInfo.allowedSources).toEqual(['team-a', 'team-b']);
-  });
-
  test('getClient returns undefined for unknown client', async () => {
    const client = await provider.clientsStore.getClient('nonexistent');
    expect(client).toBeUndefined();