docs/pre-phase/ground_truth_source_analysis.md

Ground Truth Source Analysis

Project 02 | Pre-Phase 1 Activity

Overview

This document tracks which files from @data/raw were utilized to generate the initial 25 Ground Truth QA pairs (Track A).

Total Files in Corpus: 33 Files Analyzed for QA: 10 Status: Initial batch (25/60) complete.

Analysis Summary Table

Filename	Domain	Pages	QA Pairs Generated	Key Topics Covered
Access-to-Information-2023-annual-report.pdf	Financial	11	3	FY23 metrics, page views, data indicators.
2605.02520v1.pdf	Academic	15	3	RAG benchmarking, DeepEval metrics, retrieval strategies.
Python-tutorial-pdf1.pdf	Technical	~100	4	Author, Release date, Control flow, Coding style.
AtI-annual-report-2012.pdf	Financial	80	2	Transparency theme, data accuracy disclaimers.
AtI-annual-report-2014.pdf	Financial	84	1	Rights and permissions contact info.
2604.27415v1.pdf	Academic	22	2	ChipLingo framework, EDA-Bench accuracy.
Python-tutorial-pdf2.pdf	Technical	139	2	Function definitions, Web pages.
FAO_EMSTOT.pdf	Academic	10	2	Greenhouse gas types, IPCC global warming potentials.
WB_CLEAR.pdf	Technical	17	2	CLEAR Water Dashboard purpose, generation date.
Access-to-Information-2016-annual-report.pdf	Financial	7	2	ATI Policy 6th year anniversary, WBG Finances.
worldbankSECBOS-8b71cac3...pdf	Financial	67	2	President Ajay Banga, 2025 Fiscal Year period.

Clarification on Coverage

We did not read all 33 documents for question generation. To meet the "Track A" milestone efficiently, we focused on 10 representative documents that span all three required domains (financial, academic, technical).

The remaining 23 documents were:

1. Audited (for structure/parsing issues) in docs/audit_pre_phase.md.
2. Reserved for generating the remaining 35-45 QA pairs during Phase 1 development to ensure the dataset covers the full breadth of the corpus.

---

Last Updated: May 2026