Four-file answer import evaluation

20 September 2026 · Local SkillPixel · Real Gemini extraction

Extraction accuracy: PASS for these four inputs

145/145 question rows matched numbering, order, type and A–D options. 40/40 THPT answers matched its printed key. The other 105 answers remained blank because their files contain no explicit answer key.

Blank-answer cases are not scored as 100% answer accuracy: no ground-truth key was supplied for those files.

FileRowsStructureAnswer resultService time
Test 22 — entrance exam40/4040/4040 correctly blank22.46s
Unit 10 — Lifelong Learning15/1515/1515 correctly blank16.26s
THPT 2025 with answers40/4040/4040/40 key matches21.72s
HSA English test 750/5050/5050 correctly blank26.93s

Method and limits

Ran the current extract_answers service inside the running backend using gemini-flash-latest. Each original PDF was supplied in both the question-paper and answer-file roles; the prompt, schema and normalization were unchanged. No synthetic answers were supplied. The separate GUI run used normal R2 uploads and polling.

Reference counts: Test 22, 40 questions across 4 pages; Unit 10, 15 across 2 pages; THPT, 40 across 15 pages; HSA 7, 50 across 10 pages. Text extraction and rendered pages were inspected. The THPT key is an image table on page 15 and was transcribed independently for comparison.

One direct service run per file. This measures this model response, not statistical reliability across repeated runs or other file formats. All inputs use single-choice A–D questions, so this evaluation does not cover multi-select, true/false or short-answer keys.

Environment: localhost:3000 / backend:8123; branch feature/SP-732-document-quiz-import-v1; HEAD 64d66dd1 plus existing local changes. No product code was changed during this evaluation.

Browser verification

THPT: real question upload, answer upload, extraction, 40-row review, Save and reload passed. The 40 saved answers exactly matched the draft and source key. Unit 10 GUI outcome is recorded in the companion Markdown report.

Test 22 and HSA 7 were evaluated through the actual live extraction service; their GUI upload/save paths were not repeated. Do not interpret service evidence as four complete browser E2E runs.

Before publishing: THPT includes its answer key in the same PDF. The retained question PDF would expose that page to students. Provide a question-only PDF in the first upload slot and the original key-containing PDF in the answer slot. Both QA quizzes remain drafts.

Every imported row

Test 22 — entrance exam — 40 rows, no mismatches
QuestionTypeOptionsExpected answerImportedResult
1SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
2SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
3SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
4SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
5SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
6SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
7SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
8SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
9SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
10SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
11SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
12SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
13SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
14SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
15SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
16SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
17SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
18SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
19SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
20SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
21SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
22SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
23SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
24SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
25SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
26SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
27SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
28SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
29SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
30SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
31SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
32SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
33SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
34SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
35SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
36SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
37SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
38SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
39SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
40SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
Unit 10 — Lifelong Learning — 15 rows, no mismatches
QuestionTypeOptionsExpected answerImportedResult
1SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
2SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
3SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
4SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
5SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
6SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
7SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
8SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
9SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
10SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
11SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
12SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
13SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
14SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
15SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
THPT 2025 with answers — 40 rows, no mismatches
QuestionTypeOptionsExpected answerImportedResult
1SINGLE_CHOICEA B C DAAMatch
2SINGLE_CHOICEA B C DCCMatch
3SINGLE_CHOICEA B C DCCMatch
4SINGLE_CHOICEA B C DCCMatch
5SINGLE_CHOICEA B C DDDMatch
6SINGLE_CHOICEA B C DCCMatch
7SINGLE_CHOICEA B C DCCMatch
8SINGLE_CHOICEA B C DAAMatch
9SINGLE_CHOICEA B C DBBMatch
10SINGLE_CHOICEA B C DBBMatch
11SINGLE_CHOICEA B C DCCMatch
12SINGLE_CHOICEA B C DBBMatch
13SINGLE_CHOICEA B C DDDMatch
14SINGLE_CHOICEA B C DAAMatch
15SINGLE_CHOICEA B C DAAMatch
16SINGLE_CHOICEA B C DCCMatch
17SINGLE_CHOICEA B C DCCMatch
18SINGLE_CHOICEA B C DCCMatch
19SINGLE_CHOICEA B C DAAMatch
20SINGLE_CHOICEA B C DDDMatch
21SINGLE_CHOICEA B C DAAMatch
22SINGLE_CHOICEA B C DDDMatch
23SINGLE_CHOICEA B C DAAMatch
24SINGLE_CHOICEA B C DCCMatch
25SINGLE_CHOICEA B C DDDMatch
26SINGLE_CHOICEA B C DAAMatch
27SINGLE_CHOICEA B C DAAMatch
28SINGLE_CHOICEA B C DBBMatch
29SINGLE_CHOICEA B C DDDMatch
30SINGLE_CHOICEA B C DDDMatch
31SINGLE_CHOICEA B C DCCMatch
32SINGLE_CHOICEA B C DDDMatch
33SINGLE_CHOICEA B C DDDMatch
34SINGLE_CHOICEA B C DAAMatch
35SINGLE_CHOICEA B C DAAMatch
36SINGLE_CHOICEA B C DDDMatch
37SINGLE_CHOICEA B C DBBMatch
38SINGLE_CHOICEA B C DBBMatch
39SINGLE_CHOICEA B C DDDMatch
40SINGLE_CHOICEA B C DCCMatch
HSA English test 7 — 50 rows, no mismatches
QuestionTypeOptionsExpected answerImportedResult
1SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
2SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
3SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
4SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
5SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
6SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
7SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
8SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
9SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
10SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
11SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
12SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
13SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
14SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
15SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
16SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
17SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
18SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
19SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
20SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
21SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
22SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
23SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
24SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
25SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
26SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
27SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
28SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
29SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
30SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
31SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
32SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
33SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
34SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
35SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
36SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
37SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
38SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
39SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
40SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
41SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
42SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
43SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
44SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
45SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
46SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
47SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
48SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
49SINGLE_CHOICEA B C DBlank — no source keyBlankMatch
50SINGLE_CHOICEA B C DBlank — no source keyBlankMatch

Evidence

Source reference: printed THPT key, last page
Source reference: printed THPT key, last page
THPT: browser extraction draft
THPT: browser extraction draft
THPT: saved answers reopened
THPT: saved answers reopened
Unit 10: missing answers block Save
Unit 10: missing answers block Save

Raw service outputs, normalized comparisons and GUI snapshots are stored alongside this report. Temporary tenant AI settings are restored after QA.