N1 2017-12 — audit report

70 questions (問題1–13, the non-listening half). Generated from the OCR and correction audit trails.

OCR confidence

Source: retypeset-digital — quality 3/5: Clean, fully legible retypeset digital text (PDF has a complete embedded text layer matching the page images), with underlines for target words preserved. Marked down for: bailitop Chinese ad header/footer on every page plus a large gray watermark crossing the text; loss of original layout (no boxed question numbers, re-flowed line breaks, options split across pages); two content elements surviving only as lower-quality embedded scans (問題13 tables on p24, slightly cropped, and the 聴解問題1 2番 map on p26); scattered retypesetting typos (バスワード, 再立付, 所届学部, ロボット論理, 選らんで, 確認者 for 確認書 in Q46 option 1); and the missing listening 問題5 3番 printed options.

Every section was transcribed twice independently (a full-page pass and a double-resolution half-page pass) and reconciled. Of 13 sections, 2 agreed on all content between the two passes (reconciled automatically, differing only in transcriber notes) and 11 had at least one content difference resolved by a third adjudication pass that re-read the page images. 38 leaf-level differences were examined in total.

Residual OCR risks (19):

• 問題1: transcribed from text witnesses, not the booklet image; see the section-specific audit notes
• 問題2: transcribed from text witnesses, not the booklet image; see the section-specific audit notes
• 問題3: transcribed from text witnesses, not the booklet image; see the section-specific audit notes
• 問題4: transcribed from text witnesses, not the booklet image; see the section-specific audit notes
• 問題5: transcribed from text witnesses, not the booklet image; see the section-specific audit notes
• 問題6: transcribed from text witnesses, not the booklet image; see the section-specific audit notes
• 問題7: transcribed from text witnesses, not the booklet image; see the section-specific audit notes
• 問題8: The こ-for-ご typos (こざいます/こ希望) and the redundant 宛あて are preserved as printed; they are genuine source-retype artifacts, not OCR errors, confirmed against the high-res half-page renders.
• 問題8: Passage labels （一）/(2)/(3)/(4) are inconsistent in the source but dropped per output shape; only noted.
• 問題9: The exact presence/absence of a single space after each (注N) marker is hard to read under the watermark; followed the clearest reading (space after passage 2 注1 and passage 3 注1, none elsewhere). Low impact.
• 問題9: The long dash in 育んだ産業― rendered as a single U+2015 as printed (it appears as one dash, not the ―― 2-em form); kept as both transcribers had it.
• 問題9: The half-width '?' after でしょうか? / のでしょうか? is kept exactly as printed (the source uses ASCII ?, not full-width ？); both transcribers agreed.
• 問題10: no human/agent spot-check was performed for this section (A/B byte-agreement on content was treated as sufficient)
• 問題11: The opening and closing quote glyphs around おれもやれるな are both rendered as slanted double-prime marks; transcribed as U+2033 on both sides. B noted a left/right directional difference, but at available resolution both read as U+2033; the precise intended codepoint (″ vs “”) is a retypeset artifact.
• 問題11: The half-width vs full-width parentheses distinction for (注) inline vs （注）gloss follows the print as rendered; OCR cannot fully rule out a font-metric rendering artifact, but the inline mark reads consistently narrower.
• 問題12: no human/agent spot-check was performed for this section (A/B byte-agreement on content was treated as sufficient)
• 問題13: Table-2 5-a middle character: the raster best-reads as 調育書 (月 component visible) but the heavy scan leaves residual doubt between 育 and a degraded 査; recorded as 調育書 with a note flagging the 調査書/調書 inconsistency across the document.
• 問題13: Table cell spanning is modeled with plain <td>s (content in the first row of a merged group, blanks below); the exact vertical extent of the merged 受付時間/所要日数 cells is inferred from divider lines in a low-resolution raster.
• 問題13: メ-ル/センタ- are transcribed with a half-width hyphen as literally printed; if the project later prefers normalizing the obvious intended long-vowel mark ー, these would change.

Source-text quirks the OCR transcribed verbatim (printed typos/recall artifacts; the ones judged to be retype errors were corrected in the layer below) — 12 noted at OCR time:

• 問題8: Passage (1) email contains likely retype typos, transcribed as printed: 「ありがとうこざいます」 (こ for ご in ございます), 「こ希望の便」「こ希望に添えなかった」 (こ for ご in ご希望), and redundant 「ご自宅宛あてに」 (宛 immediately followed by あて).
• 問題8: Passage (2) and its （注） gloss both spell the reading of 渇望 as かつぼう; transcribed as printed.
• 問題9: The retype prints inline kana readings as HALF-WIDTH parentheses on the baseline (e.g. 物見(ものみ)遊山(ゆさん), 富士山(ふじさん), 稀(まれ), 天邪鬼(あまのじゃく), 包摂(ほうせつ), 依拠(いきょ)); normalized to <ruby> markup in output to match JLPT ruby style. The full-width section markers （1）（2）（3） are omitted as passage labels.
• 問題11: The （注）こなす：処理する gloss is printed below passage B but glosses the word こなせる in passage A; placed at end of passage B as printed.
• 問題11: Passage A's こなせる(注) uses a half-width parenthesized 注 printed inline; transcribed as printed.
• 問題12: The 注 gloss line is printed as a partial re-typeset that erroneously repeats the inline furigana and the （注) marker: it reads '（注）空疎（くうそ）（注）な常套句（じょうとうく）：ここでは、中身のない形だけの言葉'. Normalized in output to ruby and with the duplicated inline （注） removed.
• 問題12: Possible source typo: 'ロボット論理' (リ) where 'ロボット倫理' might be expected given the surrounding list of ○○倫理 terms; transcribed as printed.
• 問題12: The 注 gloss as printed has a duplicated 注 marker and a doubled occurrence of 空疎: it reads 「（注）空疎（くうそ）（注）な常套句（じょうとうく）：…」. This is a retype artifact (the original passage has 空疎（くうそ）（注）な常套句（じょうとうく）, where 常套句 is the term being glossed, not 空疎), so the output normalizes the gloss to 「（注）空疎な常套句：…」 with ruby.
• 問題12: In the passage body, the inline furigana parentheses 空疎（くうそ） and 常套句（じょうとうく） were converted to <ruby> form. The retype's note marker placement is corrected in the layer below to follow 常套句.
• 問題13: Table 2 raster oddities: the row label '5－a' uses a long full-width dash while '5-b' uses a short half-width hyphen; transcribed as seen. The middle character of the 5-a item reads 調育書 in the raster (three characters, middle has a 月 component) — a likely retype garble; 5-b reads 調書 (two characters). Note the question 69 stem (clean digital text) writes 調査書, so the source is internally inconsistent. All kept exactly as printed in each location.
• 問題13: 再立付 (in パスワードを忘れた…再立付を願い出てください) is an unusual/likely-garbled retypeset word (perhaps for 再発行); transcribed as printed.
• 問題13: メ-ル and センタ- in the clean digital body use a short half-width hyphen rather than the long vowel mark ー; transcribed as printed. (Note: the question-70 answer text uses a real long vowel mark, メール, and is kept as such.)

Corrections applied (10)

Evidence-based reconstructions of the original exam text. This section is the canonical correction log for the sitting.

• 問3 q17 question: 「<u>撤回</u>した」 → 「<u>撤回した</u>」
  • All four choices already end in した, so leaving the source retype's した outside the tested span duplicates the inflection. The full target is 撤回した.
  • evidence: structural: 問題3 choices replace the underlined span.
• 問3 q18 question: 「<u>張り合っている</u>」 → 「<u>張り合って</u>いる」
  • All four choices are て-forms, so いる must remain outside the tested span for grammatical substitution (競い合っている). The source retype extended the underline too far, exactly as in the repeated N1 2010-07 q16 item.
  • evidence: structural: 問題3 choices replace the underlined span; repeated item in N1 2010-07 q16.
• 問8 q46 passage: 「ありがとうこざいます」 → 「ありがとうございます」
  • こざいます is not a Japanese word; the polite verb is ございます. A retype こ-for-ご artifact (quirks confirms it is a retype artifact, which is exactly this task's target). The true exam read ありがとうございます.
  • evidence: Category typo, confidence high. [retype-corrections-pass2 (propose + adversarial verify), 2026-06-20]
• 問8 q46 passage: 「こ希望の便が確保」 → 「ご希望の便が確保」
  • こ希望 is not a word; the honorific prefix is ご (ご希望). Retype こ-for-ご artifact. True exam read ご希望.
  • evidence: Category typo, confidence high. [retype-corrections-pass2 (propose + adversarial verify), 2026-06-20]
• 問8 q46 passage: 「万が一、こ希望に添えなかった」 → 「万が一、ご希望に添えなかった」
  • こ希望 is not a word; honorific prefix ご (ご希望). Retype こ-for-ご artifact. True exam read ご希望.
  • evidence: Category typo, confidence high. [retype-corrections-pass2 (propose + adversarial verify), 2026-06-20]
• 問8 q46 passage: 「ご自宅宛あてに確認書」 → 「ご自宅あてに確認書」
  • 宛あて redundantly writes the same word twice (宛 plus its kana reading あて). Polite business Japanese normally writes ご自宅あてに in kana; the doubled 宛 is a retype artifact.
  • evidence: Category added-char, confidence medium. [retype-corrections-pass2 (propose + adversarial verify), 2026-06-20]
• 問12 q65 passage: 「ロボット論理といった」 → 「ロボット倫理といった」
  • The entire list enumerates ○○倫理 terms (生命倫理、環境倫理、情報倫理、工学倫理…遺伝子倫理、脳神経倫理、ナノ倫理); 論理 (logic, りんり homophone of 倫理) breaks the pattern and must be ロボット倫理.
  • evidence: Category homophone, confidence high. [retype-corrections-pass2 (propose + adversarial verify), 2026-06-20]
• 問13 q69 passage: 「5－a.調育書(北山大学の書式)」 → 「5－a.調査書(北山大学の書式)」
  • 調育書 is not a real document name; the matching 5-b cell is 調書 and question 69 references 調査書(書式は会社指定のもの). 育 is a misread/retype of 査; the intended term is 調査書.
  • evidence: Category typo, confidence high. [retype-corrections-pass2 (propose + adversarial verify), 2026-06-20]
• 問13 q69 passage: 「再立付を願い出て」 → 「再発行を願い出て」
  • 再立付 is not a Japanese word; in context (forgotten password, request a new one) the intended term is 再発行 (reissue). A garbled retype.
  • evidence: Category typo, confidence medium. [retype-corrections-pass2 (propose + adversarial verify), 2026-06-20]
• 問12 q65 passage: <ruby>空疎<rt>くうそ</rt></ruby>（注）な<ruby>常套句<rt>じょうとうく</rt></ruby>にすぎない → <ruby>空疎<rt>くうそ</rt></ruby>な<ruby>常套句<rt>じょうとうく</rt></ruby>（注）にすぎない
  • （注N）note marker repositioned to follow its glossed term (the retype placed the marker before the word).
  • evidence: Marker-only move; the marker-stripped text is byte-identical. [note-marker relocation pass, 2026-06-20]