N1 2023-12 — audit report

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

OCR confidence

Source: booklet-scan — quality 4/5: Clean, high-contrast scan of the original printed booklets; body text, boxed question numbers, and furigana are all crisply legible. Every page carries a light diagonal Vietnamese watermark ('Tôi Yêu Ngoại Ngữ Group / Yuuki Bùi') across the upper-left, overlapping text on most pages without impairing legibility. Minor issues: right-edge section tabs (文法/読解) are clipped on a few pages, faint halftone grain/dust specks throughout, and a small scanner smudge on listening page 2; not a retypeset digital text, so short of a 5.

Every section was transcribed twice independently (a full-page pass and a double-resolution half-page pass) and reconciled. Of 13 sections, 0 agreed on all content between the two passes (reconciled automatically, differing only in transcriber notes) and 13 had at least one content difference resolved by a third adjudication pass that re-read the page images. 60 leaf-level differences were examined in total.

Calibration: mondai1–6 matched the independently hand-curated reference on 196/200 fields; all four differences were errors in the reference, not the OCR.

Residual OCR risks (34):

• 問題1: The conventions' rule text (per-kanji ruby when the reading split is unambiguous) is in tension with its whole-word surname example (池田→いけだ) for question 3's 村上/むらかみ; the whole-word surname exemplar was followed, but a reviewer applying the rule text literally could prefer A's per-kanji split.
• 問題1: The ＿＿＿ representation of the instruction's blank underline is an uncovered-by-convention judgment call both transcribers happened to share; downstream consumers should treat its exact width as arbitrary.
• 問題2: The stray hyphen-like mark inside the Q13 blank was normalized away per the blank convention; if it were ever deemed intentional printed content the transcription would need revisiting, but at full resolution it reads as a typesetting/scan artifact.
• 問題3: The ＿＿＿ rendering of the instruction's leading underline rule is an ad-hoc convention-gap decision; other sections of this exam (and other exams) must use the same rendering for cross-section consistency.
• 問題3: Q17's inclusion of 。 inside <u>…</u> rests on reading the underline endpoint at double resolution; the reading is clear, but it is the only place in this section where a one-character-cell judgement changes the transcription.
• 問題3: Q16's whole-word ruby deliberately diverges from the printed per-kanji furigana placement because the conventions hard-code 池田 as the whole-word example; if the convention is later reinterpreted as per-kanji-when-split-is-printed, this field would change.
• 問題4: The Q22 options 1-3 underline stopping mid-conjugation (before いる／いた) is typographically unusual for this question type, but pixel measurement is unambiguous; if a higher-fidelity source ever surfaces, this is the one spot worth re-confirming.
• 問題4: Both transcribers could share a correlated error on lines not spot-checked, but six lines plus all diff sites were verified character-by-character against the double-resolution scans with no discrepancies found.
• 問題5: 山下 furigana is transcribed whole-word (<ruby>山下<rt>やました</rt></ruby>); the printed rubies span the whole name, and a per-kanji split 山(やま)下(した) would also be defensible, but both transcribers used whole-word per the name example (池田) in the conventions.
• 問題5: Q31 option 3 「買ったかにかかわらず」 is unusual Japanese and likely a retypesetting artifact in the source booklet; it is transcribed as printed per the conventions, so downstream consumers should not 'correct' it.
• 問題6: The printed slot blanks in the booklet are underlines of varying width; per conventions they are normalized to the ' _ _ ★ _ ' pattern, so the transcription does not preserve the printed blank widths (intentional).
• 問題6: p11-b.png is blank except the page number, confirming questions 39-40 are the entirety of mondai6's second page; no content was missed there.
• 問題7: The furigana over 鶯 is printed very small; read as うぐいす at 4x zoom and both transcribers agree, but individual glyph quality is marginal.
• 問題7: The lead-in line 「以下は、作家が書いたエッセイである。」 is printed outside the passage box but transcribed as the first <p> of the passage; downstream consumers should be aware of this placement convention.
• 問題8: 活き活き is printed with furigana い over each 活 only (き is okurigana); per the whole-word convention it was transcribed as <ruby>活き活き<rt>いきいき</rt></ruby>, which folds the okurigana into the ruby base. If the convention is later interpreted as 'whole kanji run' rather than 'whole word including okurigana', this should become <ruby>活<rt>い</rt></ruby>き<ruby>活<rt>い</rt></ruby>き.
• 問題8: The gap between 9月13日 and 13:30 in the 日時 cell was judged a full-width space from visual width; a half-width space is conceivable.
• 問題8: The justification spaces printed inside the header labels (件　名, 日　時) are dropped in the <table> labels per B's rendering; if header labels are later compared verbatim against the print, this normalization should be remembered.
• 問題8: correctAnswerIndex is null for all four questions; no answer key appears on these pages.
• 問題8: The diagonal Vietnamese watermark crosses several lines on p14-p17; all affected glyphs were legible at double resolution, but the watermark slightly degrades confidence on small furigana.
• 問題9: The 35〜40 wave dash in passage (2) is transcribed as 〜 (U+301C) per both transcribers; the printed glyph could be a fullwidth tilde ～ (U+FF5E). Both agreed, so it was not re-adjudicated.
• 問題9: Inline （注N） marker glyphs are tiny in the scan; their exact horizontal placement relative to surrounding kana (e.g., 鵜呑み（注1）に vs 鵜呑（注1）み) was judged by best reading and the mondai7/8 precedent, but sub-character placement carries minor uncertainty.
• 問題9: The full-width spaces after ？ and ！ in passage (3) are inferred from inter-glyph gaps in the scan; rendering ambiguity is possible though both transcribers and the source layout support them.
• 問題10: The printed parentheses around （中略） and the in-text （注） mark are small enough that half-width vs full-width cannot be distinguished with certainty; transcribed full-width per the convention that Japanese punctuation stays full-width, and both transcribers agree.
• 問題10: The leader after 「シュバイッツア博士」 is transcribed as …… (six-dot leader, two ellipsis characters); the exact dot count is consistent with the high-resolution image but small print leaves minor uncertainty.
• 問題10: The colon in the 注 gloss （注）アングル：角度 is transcribed as full-width ：; at print size a half-width colon would look nearly identical.
• 問題11: The furigana りちぎ is printed as a continuous run over 律儀; the per-kanji split 律(りち)儀(ぎ) follows the standard reading and seems unambiguous, but the print itself does not mark the boundary.
• 問題11: The diagonal watermark crosses several lines of passage A on p28; all overlapped characters were legible at double resolution, but the watermark slightly lowers contrast around 律儀さ and キリがない.
• 問題11: The phrasing 「…と考える人は「小説の書き方マニュアル」を信じる律儀さと同じで」 reads awkwardly and may itself be a retypeset artifact of the leaked booklet; it is transcribed exactly as printed per the conventions.
• 問題12: The exact terminus of the paragraph-2 underline relative to the following 。 is marginal at scan resolution; both transcribers and the question 62 quotation exclude the 。, which was adopted.
• 問題12: The 2-em dash was encoded as two U+2015 characters by both transcribers; the printed glyph is a single long dash, so a different but equally defensible encoding (e.g. U+2014 ×2) exists.
• 問題12: The booklet prints no source citation after the passage; this may be an omission in the retypeset source, but the transcription faithfully reflects the printed page.
• 問題13: The half-width space after circled digits ①-④ reflects a clearly visible printed gap, but it could be glyph side-bearing rather than a true typed space; rendering is unaffected either way.
• 問題13: The tab label マスダ買い取りサービス is printed in bold italic with an underline rule; italic and the rule cannot be represented in the allowed tag set, so only <b> is kept.
• 問題13: <th> vs <td> for the row-label cells is a semantic judgment based on the double rule setting off the label column; the labels themselves are printed in the same plain weight as body text.

Source-text quirks the OCR transcribed verbatim (printed typos/recall artifacts; the ones judged to be retype errors were corrected in the layer below) — 17 noted at OCR time:

• 問題1: Question 5 option 2 reads になって (likely the reading of 担って used as a distractor); verified against both image halves.
• 問題2: Question 13: a small stray hyphen-like mark is printed inside the blank, immediately before the closing parenthesis (looks like （　　-）). Both transcribers and the adjudicator confirmed it in the double-resolution scan; it appears to be a retypesetting or scan artifact rather than intended content, so the blank was transcribed as the standard （　　） per convention.
• 問題3: Question 16: the underline covers 肝心な including な (consistent with the な-adjective options); verified at high zoom.
• 問題3: Question 17: the printed underline extends a full character cell under the sentence-final 。, so the period is included inside <u>…</u>; verified at high zoom (this typesetter otherwise ends underlines precisely, e.g. 14/16/18/19).
• 問題5: Question 31 option 3 is printed as 「買ったかにかかわらず」 (not 「買ったにもかかわらず」; confirmed independently by both transcribers and re-verified against the double-resolution image during adjudication); possibly a retypesetting artifact, but transcribed exactly as printed.
• 問題6: Question 36: the ★ occupies the first of the four slots as printed (印刷の ★ ＿ ＿ ＿), unlike the other questions where it is the third slot. Verified against the page image.
• 問題7: In the second paragraph, the printed text has a full-width space after 「言いたいのかね？」 before 「という声が聞こえる」; it is transcribed as printed.
• 問題7: The furigana over 雁 in 「月に雁」 is printed as かりがね (not がん or かり); verified against the double-resolution scan at 4x zoom.
• 問題8: Passage (2): the email header is printed as 宛て先／件　名／日　時 label：value lines between double horizontal rules; rendered as a <table> per convention, dropping the printed ：separators (replaced by the cell boundary) and the full-width justification spaces inside the labels 件名 and 日時. The label is printed 宛て先 (with て), transcribed as printed. The double rule that ends the email after 担当：上田　映子 is likewise not representable and omitted. The lead-in sentence 「以下は、ある電気店から届いたメールである。」 is printed above the ruled block and included as the first <p>.
• 問題8: Passage (2): the time is printed 13：30 with a full-width colon; normalized to half-width 13:30 per convention, with digits normalized to half-width ASCII. Other printed full-width colons (e.g. 担当：) are kept as printed.
• 問題9: Passage (3) paragraph 4 prints a full-width space after both ？ and ！ (「君、広報が好きなの？　じゃあ、やってみなさい！　自分で言うなら…」); transcribed as printed.
• 問題10: The passage prints 「シュバイッツア博士」 (small ッ followed by full-size ツ and full-size ア, no small ァ or long vowel mark) for Eugene Smith's Schweitzer photo; transcribed as printed even though シュバイツァー would be the usual spelling.
• 問題10: 「多くの人たちによさが認めてもらえる環境が整った」 is printed with が after よさ (one might expect よさを); verified at high resolution; possibly a source typo but transcribed as printed.
• 問題11: Passage A, first paragraph: printed as 「書こうとしている人やすでに」 with no comma after 人や — transcribed as printed.
• 問題11: Passage A, second body paragraph: 「…を信じる律儀さと同じで、たしかに真面目で素直ないい人…」 — the phrasing 人は…律儀さと同じで is slightly awkward (one might expect 信じる人と同じで) but it is transcribed exactly as printed; 素直ないい人 (two い) confirmed at high zoom during adjudication (transcriber B had read a single い).
• 問題12: No source citation line (e.g. （○○『…』による）) is printed after the passage; the 注 glosses follow the final paragraph directly. Unusual for an N1 問題12 passage and possibly an omission in this retypeset.
• 問題12: The passage uses both 面倒がないところに変化はない and 面倒のない人間関係 (が vs の) — transcribed as printed and verified against the page image.