Trace parse artifacts from raw upload to vector index.

This view is for operators diagnosing why one document is delayed or degraded. It surfaces parser settings, semantic structure, chunk generation, and Milvus insertion as separate observable stages.

Current run

Battery density addendum review

Uploaded 09:14 by Battery Safety Team · parser backend `aliyun`

Pipeline progression

Live state

1. Object storage ingestion

Stored in bucket `upload-files` with artifact prefix `artifacts`

2. Aliyun parse layout extraction

Parsed pages, recovered tables, and OCR confidence summarize here once the run completes.

3. Semantic blocks

Semantic block persistence is tracked here after parse artifact storage completes.

4. Vector chunk build

Using overlap chunking with header-prefixed embedding text

5. Embedding generation

Target model `text-embedding-v3` · dimension 1024

6. Milvus insertion

Waiting for chunk vectors before collection sync

Run log

Recent events

09:18:11

Semantic block serialization completed

Stored block tree and section hierarchy in Postgres parse artifact store.

09:20:44

Chunk builder emitted overlap windows

Header context is prepended to vector chunks for downstream retrieval quality.

09:22:08

Embedding worker rate-limited temporarily

Retry budget still healthy. No manual action required unless latency exceeds 15 minutes.

Artifacts generated

Output layers

Layout JSON Page, table, and text-span counts populate from the parser artifact output.

Semantic blocks Semantic nodes are mapped into chapter and clause hierarchy here.

Vector chunks Overlap windows and embedding text populate after chunk generation.

Chunk profile

Top segments

4.2 Energy density threshold

Critical requirement clause

chunk count pending character count pending Linked

5.1 Thermal event test method

Supplier evidence cross-reference

chunk count pending character count pending Review

Appendix A formulas and tables

Dense table extraction from scan

chunk count pending character count pending Noisy

Configuration snapshot

Runtime values

Parser backend

Document extraction engine

aliyun 5 s poll 900 s timeout

Embedding target

Vector generation

text-embedding-v3 1024 dim top_k 10

Collection

Milvus destination

regulations_dense_1024_v2 dense-only ready