Ship DeepSeek roadmap P0-5/P2 integrations.

Smoke simple_project offline, reviewer shortest_path, library promotion
gate, PDF drop logs, and per-stage cache hit-rate helper — without
touching Layout placement packing.

Co-authored-by: Cursor <cursoragent@cursor.com>

backend/services/api_logs.py

@@ -104,3 +104,27 @@ class ApiLogger:
 
         key = f"{project_prefix(user_id, project_id)}/api_logs.jsonl"
         storage.write_text(key, text)
+
+
+def cache_stats_by_stage(entries: list[dict]) -> dict[str, dict]:
+    """Roll up prompt-cache hit rate per pipeline stage.
+
+    Returns ``{stage: {calls, input_tokens, cache_read_tokens, hit_ratio}}``.
+    ``hit_ratio`` is cache_read / input when input > 0, else 0.
+    """
+    out: dict[str, dict] = {}
+    for e in entries:
+        stage = str(e.get("stage") or "unknown")
+        bucket = out.setdefault(
+            stage,
+            {"calls": 0, "input_tokens": 0, "cache_read_tokens": 0, "hit_ratio": 0.0},
+        )
+        bucket["calls"] += 1
+        bucket["input_tokens"] += int(e.get("input_tokens") or 0)
+        bucket["cache_read_tokens"] += int(e.get("cache_read_input_tokens") or 0)
+    for bucket in out.values():
+        inp = bucket["input_tokens"]
+        bucket["hit_ratio"] = (
+            round(bucket["cache_read_tokens"] / inp, 4) if inp else 0.0
+        )
+    return out

backend/services/llm/pdf_ingest.py

@@ -133,6 +133,16 @@ def render_pdf_page_jpegs(
         return []
 
     try:
+        total = len(doc)
+        omitted = [i + 1 for i in range(total) if i not in set(page_indices)]
+        if omitted:
+            log.info(
+                "PDF page images: %s rendering %d/%d pages; omitted e.g. %s",
+                pdf_path.name,
+                min(len(page_indices), max_pages),
+                total,
+                omitted[:12],
+            )
         matrix = fitz.Matrix(zoom, zoom)
         for i in page_indices:
             if i < 0 or i >= len(doc):

backend/services/pipeline.py

@@ -839,10 +839,30 @@ async def _stage_ic_extraction(ctx: PipelineContext) -> None:
                     api_logger=private,
                 )
 
-                # Upload to storage, then copy to library
+                # Upload to storage, then copy to library only if pintable is usable
                 extracted_key = f"{ctx.ws.prefix}/extracted/{safe}.json"
                 ctx.storage.upload_from_local(json_path, extracted_key)
-                proj_svc.save_to_library(ctx.storage, extracted_key, "extracted", f"{safe}.json")
+                try:
+                    from backend.pinscopex.library_gate import should_promote_extraction
+
+                    payload = json.loads(json_path.read_text(encoding="utf-8"))
+                    ok, reason = should_promote_extraction(payload)
+                    if ok:
+                        proj_svc.save_to_library(
+                            ctx.storage, extracted_key, "extracted", f"{safe}.json",
+                        )
+                    else:
+                        logger.warning(
+                            "Skipping library promote for %s: %s (kept project-local)",
+                            mpn, reason,
+                        )
+                except Exception:
+                    logger.exception(
+                        "Library gate failed for %s — promoting anyway", mpn,
+                    )
+                    proj_svc.save_to_library(
+                        ctx.storage, extracted_key, "extracted", f"{safe}.json",
+                    )
 
                 # Upload source datasheet PDF to library (content-addressed)
                 store_datasheet(ctx.storage, pdf_path, mpn)