ThinkIdiot PDF RAG tutorial validation
Date: 2026-10-13
Python: 3.12.13, Linux ARM64
Ollama: 0.32.13
Execution: CPU, RAG_NUM_GPU=0, RAG_NUM_THREADS=4
Deterministic tests: 8 passed
Fixture: original fictional four-page sample-handbook.pdf
These are smoke checks, not a general accuracy benchmark.
Model digests:
nomic-embed-text:latest: 0a109f422b47e3a30ba2b10eca18548e944e8a23073ee3f3e947efcf3c45e59f
qwen3:4b: 359d7dd4bcdab3d86b87d73ac27966f4dbb9f5efdfcc75d34a8764a09474fae7

Live check output:
Indexed 4 passages into temporary index.json
PASS: expected page 1, answer fact 25, matching quotation

25

[1] sample-handbook.pdf, PDF page 1
    Quote: Annual leave Full-time employees receive 25 days of paid annual leave per calendar year.
PASS: expected page 2, answer fact 160, matching quotation

EUR 160

[1] sample-handbook.pdf, PDF page 2
    Quote: Business travel The hotel reimbursement limit is EUR 160 per night.
PASS: expected page 3, answer fact 300, matching quotation

EUR 300

[1] sample-handbook.pdf, PDF page 3
    Quote: The annual equipment allowance is EUR 300.

PASS: missing revenue information produced an abstention