@babav/knowledge-core-client 0.38.1 → 0.39.0

package/dist/index.d.ts

@@ -342,16 +342,22 @@ export interface ChunkHeat {
         bbox: [number, number, number, number];
     }[] | null;
 }
-/** Everything to overlay retrieval heat on the original document: a signed view-PDF URL
- * (render with pdf.js), its per-page geometry, and every chunk's heat + boxes. */
+/** Everything to overlay retrieval heat on the original document. Fetch the view-PDF BYTES
+ * via `analytics.documentViewPdf(id)` from YOUR BACKEND (KC serves them; the browser never
+ * touches GCS), render with pdf.js, and shade chunks by heat. `insufficient_data` (with
+ * `retrieval_events` as the sample size) is the document-level analogue of the corpus heat
+ * view's insufficient_data band — show a clear "not enough data yet" indication instead of
+ * misleading shading when it's true. */
 export interface ChunkHeatmapResponse {
     document_id: UUID;
-    view_pdf_url: string | null;
+    has_view_pdf: boolean;
     page_dims: {
         w: number;
         h: number;
         rotation: number;
     }[] | null;
+    retrieval_events: number;
+    insufficient_data: boolean;
     chunks: ChunkHeat[];
 }
 export interface MonthlyCorpus {
@@ -563,6 +569,8 @@ declare class HttpBase {
     protected url(path: string, query?: RequestOpts["query"]): string;
     protected raw(method: string, path: string, opts?: RequestOpts): Promise<Response>;
     protected request<T>(method: string, path: string, opts?: RequestOpts): Promise<T>;
+    /** GET binary bytes as a Blob (error bodies are JSON, so decode+throw on !ok). */
+    protected rawBytes(method: string, path: string, signal?: AbortSignal): Promise<Blob>;
     /** Auto-paginate a list endpoint into a single array. */
     protected pageAll<T>(path: string, query?: RequestOpts["query"]): Promise<T[]>;
 }
@@ -962,9 +970,15 @@ export declare class KnowledgeCoreClient extends HttpBase {
         document: (documentId: UUID) => Promise<DocumentAnalytics>;
         /** Parent-section grain for a document (never exposes chunk ids). */
         documentSections: (documentId: UUID) => Promise<SectionsResponse>;
-        /** Per-chunk retrieval heat + position in the document's view-PDF, plus a signed
-         * view-PDF URL — everything to overlay heat on the ORIGINAL document with pdf.js. */
+        /** Per-chunk retrieval heat + position + `has_view_pdf` + the `insufficient_data`
+         * indication — everything to overlay heat on the ORIGINAL document with pdf.js. Fetch the
+         * view-PDF bytes with `documentViewPdf(id)` (below). */
         documentChunks: (documentId: UUID) => Promise<ChunkHeatmapResponse>;
+        /** Fetch the document's canonical view-PDF BYTES (the surface for the heat overlay).
+         * Call this from YOUR BACKEND — KC serves the bytes, so the browser never fetches from
+         * GCS (no bucket CORS). Returns a Blob (Node: `Buffer.from(await blob.arrayBuffer())`;
+         * browser: `URL.createObjectURL(blob)`). Rejects 404 when `has_view_pdf` is false. */
+        documentViewPdf: (documentId: UUID, signal?: AbortSignal) => Promise<Blob>;
         /** Whole-corpus scatter payload: every document with x=retrieved_reach, y=conversion,
          *  bubble=chunk_count, and a server-computed quadrant. No pagination (`truncated` if capped). */
         corpusScatter: (corpusId: UUID) => Promise<ScatterResponse>;

package/dist/index.js

@@ -102,6 +102,15 @@ class HttpBase {
             throw new KnowledgeCoreError(res.status, parsed ?? text, path);
         return parsed;
     }
+    /** GET binary bytes as a Blob (error bodies are JSON, so decode+throw on !ok). */
+    async rawBytes(method, path, signal) {
+        const res = await this.raw(method, path, { signal });
+        if (!res.ok) {
+            const t = await res.text();
+            throw new KnowledgeCoreError(res.status, safeJson(t), path);
+        }
+        return await res.blob();
+    }
     /** Auto-paginate a list endpoint into a single array. */
     async pageAll(path, query = {}) {
         const out = [];
@@ -468,9 +477,15 @@ export class KnowledgeCoreClient extends HttpBase {
         document: (documentId) => this.request("GET", `/v1/analytics/documents/${documentId}`),
         /** Parent-section grain for a document (never exposes chunk ids). */
         documentSections: (documentId) => this.request("GET", `/v1/analytics/documents/${documentId}/sections`),
-        /** Per-chunk retrieval heat + position in the document's view-PDF, plus a signed
-         * view-PDF URL — everything to overlay heat on the ORIGINAL document with pdf.js. */
+        /** Per-chunk retrieval heat + position + `has_view_pdf` + the `insufficient_data`
+         * indication — everything to overlay heat on the ORIGINAL document with pdf.js. Fetch the
+         * view-PDF bytes with `documentViewPdf(id)` (below). */
         documentChunks: (documentId) => this.request("GET", `/v1/analytics/documents/${documentId}/chunks`),
+        /** Fetch the document's canonical view-PDF BYTES (the surface for the heat overlay).
+         * Call this from YOUR BACKEND — KC serves the bytes, so the browser never fetches from
+         * GCS (no bucket CORS). Returns a Blob (Node: `Buffer.from(await blob.arrayBuffer())`;
+         * browser: `URL.createObjectURL(blob)`). Rejects 404 when `has_view_pdf` is false. */
+        documentViewPdf: (documentId, signal) => this.rawBytes("GET", `/v1/analytics/documents/${documentId}/view-pdf`, signal),
         /** Whole-corpus scatter payload: every document with x=retrieved_reach, y=conversion,
          *  bubble=chunk_count, and a server-computed quadrant. No pagination (`truncated` if capped). */
         corpusScatter: (corpusId) => this.request("GET", `/v1/analytics/corpora/${corpusId}/scatter`),

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@babav/knowledge-core-client",
-  "version": "0.38.1",
+  "version": "0.39.0",
   "description": "TypeScript client for the Babav Knowledge Core API (Deno + Node 18+, zero deps). Includes the babav.visual grammar TYPES at the ./visual subpath (types only; all visual rendering is server-side).",
   "type": "module",
   "main": "./dist/index.js",

package/src/index.ts

@@ -346,12 +346,18 @@ export interface ChunkHeat {
   method: string | null;
   regions: { page: number; bbox: [number, number, number, number] }[] | null;
 }
-/** Everything to overlay retrieval heat on the original document: a signed view-PDF URL
- * (render with pdf.js), its per-page geometry, and every chunk's heat + boxes. */
+/** Everything to overlay retrieval heat on the original document. Fetch the view-PDF BYTES
+ * via `analytics.documentViewPdf(id)` from YOUR BACKEND (KC serves them; the browser never
+ * touches GCS), render with pdf.js, and shade chunks by heat. `insufficient_data` (with
+ * `retrieval_events` as the sample size) is the document-level analogue of the corpus heat
+ * view's insufficient_data band — show a clear "not enough data yet" indication instead of
+ * misleading shading when it's true. */
 export interface ChunkHeatmapResponse {
   document_id: UUID;
-  view_pdf_url: string | null;
+  has_view_pdf: boolean;      // a view-PDF exists -> fetch bytes with analytics.documentViewPdf(id)
   page_dims: { w: number; h: number; rotation: number }[] | null;
+  retrieval_events: number;   // distinct retrievals this doc appeared in (heat sample size)
+  insufficient_data: boolean; // too few retrievals to trust per-chunk heat yet
   chunks: ChunkHeat[];
 }
 export interface MonthlyCorpus {
@@ -589,6 +595,16 @@ class HttpBase {
     return parsed as T;
   }
 
+  /** GET binary bytes as a Blob (error bodies are JSON, so decode+throw on !ok). */
+  protected async rawBytes(method: string, path: string, signal?: AbortSignal): Promise<Blob> {
+    const res = await this.raw(method, path, { signal });
+    if (!res.ok) {
+      const t = await res.text();
+      throw new KnowledgeCoreError(res.status, safeJson(t), path);
+    }
+    return await res.blob();
+  }
+
   /** Auto-paginate a list endpoint into a single array. */
   protected async pageAll<T>(path: string, query: RequestOpts["query"] = {}): Promise<T[]> {
     const out: T[] = [];
@@ -1088,9 +1104,16 @@ export class KnowledgeCoreClient extends HttpBase {
     document: (documentId: UUID) => this.request<DocumentAnalytics>("GET", `/v1/analytics/documents/${documentId}`),
     /** Parent-section grain for a document (never exposes chunk ids). */
     documentSections: (documentId: UUID) => this.request<SectionsResponse>("GET", `/v1/analytics/documents/${documentId}/sections`),
-    /** Per-chunk retrieval heat + position in the document's view-PDF, plus a signed
-     * view-PDF URL — everything to overlay heat on the ORIGINAL document with pdf.js. */
+    /** Per-chunk retrieval heat + position + `has_view_pdf` + the `insufficient_data`
+     * indication — everything to overlay heat on the ORIGINAL document with pdf.js. Fetch the
+     * view-PDF bytes with `documentViewPdf(id)` (below). */
     documentChunks: (documentId: UUID) => this.request<ChunkHeatmapResponse>("GET", `/v1/analytics/documents/${documentId}/chunks`),
+    /** Fetch the document's canonical view-PDF BYTES (the surface for the heat overlay).
+     * Call this from YOUR BACKEND — KC serves the bytes, so the browser never fetches from
+     * GCS (no bucket CORS). Returns a Blob (Node: `Buffer.from(await blob.arrayBuffer())`;
+     * browser: `URL.createObjectURL(blob)`). Rejects 404 when `has_view_pdf` is false. */
+    documentViewPdf: (documentId: UUID, signal?: AbortSignal): Promise<Blob> =>
+      this.rawBytes("GET", `/v1/analytics/documents/${documentId}/view-pdf`, signal),
     /** Whole-corpus scatter payload: every document with x=retrieved_reach, y=conversion,
      *  bubble=chunk_count, and a server-computed quadrant. No pagination (`truncated` if capped). */
     corpusScatter: (corpusId: UUID) => this.request<ScatterResponse>("GET", `/v1/analytics/corpora/${corpusId}/scatter`),