Document Intelligence Benchmark Workbench V3.1

BIC sample references · Excel importer · immutable prompts · suites · per-case and summary accuracy

qwen3-vl-plus-2025-12-19

1. Prompt versions

Create a new named version when you change a prompt. Once used, that exact version is locked for benchmark integrity.

Create prompt version

Each benchmark case references one BIC sample. The BIC content is snapshotted into the case, so later Excel edits cannot change historical results.

Import BIC samples from Excel

Create / replace one BIC sample manually

Each case stores any uploaded document files plus a BIC reference snapshot. You can create a case with only one document type.

Test case Model Application prompt eKTP prompt RIPLAY prompt FX Threshold prompt FX Jumlah prompt FORM_LDD prompt Slip Aplikasi 2 prompt

Layout crop/retry UNCERTAIN Verify critical identifiers

Create a suite from any cases, then run all cases using exactly the selected prompt versions and each case's own BIC reference.

Suite name

Suite

Run	Model	Case	BIC Reference	Prompts	Raw	Final	Status

First pass—

After retry—

Final—

Target≥99%

Field	Final	BIC	Result

Full immutable audit JSON