N1 2018-07 — 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: Grayscale scan of a retypeset (horizontal, Chinese-prep-site) printout: text fully legible throughout but soft, with dense small type on pp.4-6 and 10, light speckling, a header watermark on every content page, and a handwritten name on p2 (not overlapping text). Cover (p1) and final ad (p18) are crisp digital.

Every section was transcribed twice independently (a full-page pass and a double-resolution half-page pass) and reconciled. Of 13 sections, 1 agreed on all content between the two passes (reconciled automatically, differing only in transcriber notes) and 12 had at least one content difference resolved by a third adjudication pass that re-read the page images. 48 leaf-level differences were examined in total.

Residual OCR risks (26):

• 問題1: Source is a soft, speckled grayscale retypeset scan; the stray dots before Q1 opt3 and after Q6 opt3 are scan artifacts rather than printed glyphs, judged so but not 100% certain.
• 問題1: Instruction line uses ＿＿＿ (three full-width low lines) per convention; the printed blank width is approximate and normalized.
• 問題2: Scan is a softer grayscale retypeset; no suspicious printed typos detected in this section. All values are short fill-in-blank stems with no markup, so formatting-convention ambiguity is minimal.
• 問題3: Scan is a speckled grayscale retypeset; underline endpoints were judged by where the continuous rule terminates relative to characters. Q15's long underline is the one non-obvious call, but it is well-supported by both the image and the answer format.
• 問題3: No furigana is printed anywhere in this section, consistent with both transcribers.
• 問題4: Underline endpoints on conjugated verbs/adjectives are inherently fuzzy in this soft grayscale scan; Q22 option 2 (乗り出している, whole-form underline) vs options 1/3/4 (乗り出 only) was judged from magnified crops and is the one genuinely asymmetric case, but a reviewer relying on the low-res full-page render could read it differently.
• 問題4: Q20 option 4's absent sentence-final 。 is transcribed as printed; if the source actually has a faint 。 lost to scan speckle it would be a print/scan artifact, not a transcription choice.
• 問題5: Q28 number printed full-width (１０分) was normalized to 10分 per convention; correct but a deviation from the literal glyphs.
• 問題5: Speckled grayscale scan; small kana like 'だ' vs absence (the source of two of the diffs) are the main legibility hazard, but the double-resolution halves rendered them unambiguously.
• 問題6: Q39 「北市の住民」 may be a typeset truncation of a fuller city name (e.g. 〜北市); the scan shows exactly 北市 so it is transcribed as printed.
• 問題6: Q36 「来月初」 is unusual (expected 来月初旬?); printed glyphs are legible as 初, transcribed as-is.
• 問題6: Convention's intent for the 問題6 instruction-line star blank is inferred (treated as a canonical star slot); if the project later prefers a literal compact representation, the instruction line would need revisiting.
• 問題7: 程違い vs 程遠い: the scan is a soft grayscale retypeset; the inner radical reads as 韋 (違) rather than 袁 (遠), so kept as 程違い per 'transcribe typos as-is', but at this resolution 遠 cannot be 100% excluded.
• 問題7: The inline-kana artifact 上うわの空そら is unusual; normalized to <ruby>上<rt>うわ</rt></ruby>の<ruby>空<rt>そら</rt></ruby> to match JLPT ruby style and avoid duplicating kana in rendered text.
• 問題8: no human/agent spot-check was performed for this section (A/B byte-agreement on content was treated as sufficient)
• 問題9: Faint scan; full-resolution half-page images were legible at all decided points, but the 主なる/主な and のだ、 insertions were the subtlest reads. No OCR tooling (PIL/ImageMagick) available locally for pixel-level zoom, so judgments rest on the supplied renders.
• 問題9: No furigana present in this section; conventions for ruby were not exercised here.
• 問題10: The equals sign in 個人の自己＝＜私＞ is printed in an ASCII-ish width; conventions do not cover '='. Retained full-width ＝ (B's choice). Could arguably be half-width '='.
• 問題10: Both transcribers shared the 重要性→重要度 error; corrected here. Other agreed spans were spot-checked but not 100% exhaustively, so a residual shared-error risk on un-checked passage spans is low but nonzero.
• 問題11: Whether （中略） should be its own <p> is a convention judgment call: it is physically inline in this scan but is conventionally an editorial omission marker. Resolved to inline (A) per the literal "own printed line" rule.
• 問題11: Soft grayscale scan; no furigana and no underlines/circled marks appear in this section, so nothing to mis-read there, but faint kana could in principle hide a stray character.
• 問題12: Body 注 reference marker is a narrow paren glyph; normalized to full-width （注） — a formatting judgment, not certain the source intended full-width.
• 問題12: Confirmed there is no source-citation （…による）line; the passage genuinely ends at the 注 gloss.
• 問題13: Table left-column labels (ボランティア①②③) rendered as <td>; they are arguably row-headers, so a <th> reading is defensible. They carry no visual header styling in the source, hence <td>.
• 問題13: The stray closing '）' in the 活動日時 row and the '売ってり' / '1日目' wording are preserved as printed; these look like source-retypeset typos but are intentionally not corrected.
• 問題13: '③50円' in the 登録費用 line is faint/speckled but reads as ③ consistent with the ①②③ bullet pattern.

Source-text quirks the OCR transcribed verbatim (printed typos/recall artifacts; the ones judged to be retype errors were corrected in the layer below) — 13 noted at OCR time:

• 問題6: Source is a speckled grayscale retypeset scan; text legible throughout.
• 問題6: Q36 stem reads 来月初に (printed as-is); 初 without 旬 may be a source typo but transcribed verbatim.
• 問題6: Q39: text reads 「北市の住民」 (likely a typeset truncation of a city name such as 〜市); transcribed as printed.
• 問題7: Inline-furigana retypeset artifact: the gloss word is printed as '上うわの空そら' (kanji 上/空 each followed by their inline reading うわ/そら, i.e. 上の空 read うわのそら). Normalized to <ruby>/<rt> markup in output.
• 問題7: '程違い身の処し方' is a likely typo for 程遠い ('far from'); the printed glyph reads 違 (韋 component), so transcribed as printed and flagged.
• 問題9: Source is a Chinese-retypeset grayscale scan with speckling; 日语轻松考 headers and page-number footers (第6页 etc.) ignored per conventions.
• 問題9: Passage (2): 「会社の利益が増えれば全体の年俸の原資（注3）は増えますが」 confirmed against the image as 増えれば (an earlier draft suspected a 増えば typo; that was an OCR slip, not in the source).
• 問題10: Paragraph 5: the print clearly reads マスメディアに対抗する (対抗, a clean glyph — there is no 対将 typo); the underline on 強力な情報発信ツール is present. Text continues ...さまざまな自分の活動... そしてときには心境や悩み...公表するようになった...主導権を確保する...どの程度か...自力で公に情報発信する...手に入れたことに変わりはないだろう...自らの手でつくった.
• 問題11: Source is a speckled grayscale retypeset scan (小程序：日语轻松考 header); no furigana printed in this section.
• 問題13: ボランティア③ description cell prints '花を売ってりします' (apparent source typo for '売ったりします'); transcribed as printed.
• 問題13: The 活動日時 row prints '...11月9日収穫祭）各イベント...' with a stray closing full-width parenthesis '）' that has no matching opening on that line; transcribed as printed (suspected source typo).
• 問題13: 締切 line reads '締切①、②登録説明会開催日1日目の1週間前、③各イベント実施日の1週間前必着' — '締切①' (the marker is ①, not 'り') and '1日目'; transcribed as printed.
• 問題13: Scan is a soft grayscale Chinese retypeset with speckling; all digits/Latin normalized to half-width per conventions.

Corrections applied (13)

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

• 問3 q14 question: 「<u>すみやか</u>に片づけて」 → 「<u>すみやかに</u>片づけて」
  • The choices are complete adverbial phrases; leaving に outside the tested span duplicates it for choices such as 元の通りに and できるだけきれいに. The full target is すみやかに.
  • evidence: structural: 問題3 choices replace the underlined span.
• 問3 q18 question: 「<u>つかの間</u>の休息」 → 「<u>つかの間の</u>休息」
  • The choices are complete adnominal phrases, including the correct 短い. Leaving の outside the tested span produces 短いの休息; the full target is つかの間の.
  • evidence: structural: 問題3 choices replace the underlined span.
• 問4 q22 answers[0]: 「新たな道に<u>乗り出</u>した」 → 「新たな道に<u>乗り出した</u>」
  • Headword 乗り出す; the printed underline covers 乗り出した (including した). Our span drops した.
  • evidence: underline-boundary-vs-image
• 問4 q22 answers[2]: 「政府が調査に<u>乗り出</u>した」 → 「政府が調査に<u>乗り出した</u>」
  • Headword 乗り出す; the printed underline covers 乗り出した (including した). Our span drops した.
  • evidence: underline-boundary-vs-image
• 問4 q22 answers[3]: 「すぐに料理に<u>乗り出</u>した」 → 「すぐに料理に<u>乗り出した</u>」
  • Headword 乗り出す; the printed underline covers 乗り出した (including した). Our span drops した.
  • evidence: underline-boundary-vs-image
• 問6 q36 question: 「妹は、来月初に」 → 「妹は、来月初めに」
  • 来月初 is not a word; the idiom is 来月初め ('early next month'). The め was dropped in retyping. The witness answer-explanation reconstructs the full sentence as 「妹は、来月初めに...」.
  • evidence: jlpt-freq witness 解析 line (input/N1_2018-07/5Mcv4He-jlpt-freq-exam-and-answers.txt:1174): 「妹は、来月初めに 3 引越しするのを機に…」; the witness exam-prompt line (135) drops め identically to the corpus, while the assembled-answer analysis preserves it. [correct-retypes workflow (propose + adversarial verify), 2026-06-17]
• 問7 q41 passage: 「自分自身の実態には程違い身の処し方である」 → 「自分自身の実態には程遠い身の処し方である」
  • 程違い is a non-word; the established collocation is 〜には程遠い ('far from ~'). 遠 was misretyped as 違.
  • evidence: tools/find-source.py: 「実態には程遠い」 returns multiple Google Books hits (世界「民族」全史 / 宇山卓栄, 残業ゼロのノート術 / 石川和男, 人権と部落問題); 「実態には程違い」 returns zero hits. [correct-retypes workflow (propose + adversarial verify), 2026-06-17]
• 問4 q20 answers[3]: 「この器は作り出される」 → 「この器は作り出される。」
  • Dropped sentence-final 。. This is the natural-usage correct option (not a deliberate distractor), and the missing period is punctuation only.
  • evidence: Internal consistency: all of q20 options 0-2 and every option in q21-q25 end in 。; only this option lacks it. Note the direct witness line also omits the period, so support is sibling-consistency rather than a positive witness; change is minimal and low-risk. [correct-retypes workflow (propose + adversarial verify), 2026-06-17]
• 問13 q69 passage: 「花を売ってりします」 → 「花を売ったりします」
  • 売ってり is a non-word. The parallel ～たり…～たり structure with the immediately adjacent 野菜を売ったり requires 売ったり. The flyer is genuine reading material, not a usage-distractor.
  • evidence: Passage reads 「…公園内で収穫した野菜を売ったり、花を売ってりします。」 — adjacent 野菜を売ったり establishes the parallel structure; 売ってり is an unambiguous typo for 売ったり. [correct-retypes workflow (propose + adversarial verify), 2026-06-17]
• 問3 q17 question: 「<u>エレガント</u>な」 → 「<u>エレガントな</u>」
  • 問題3 underline boundary: synonym options are adnominal, so the trailing な belongs inside the underlined span (same pattern as N1_2020-12 q16 架空の). Without it the answer leaves a dangling な.
  • evidence: structural: 問題3 options replace the underlined adnominal span.
• 問9 q0 passage: 「（注1）な箱」 → 「な（注1）箱」
  • （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]
• 問10 q0 passage: 「（注1）をひそめる人」 → 「をひそめる（注1）人」
  • （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]
• 問6 q40 question/correctOrder: 「政府が発表した数字によると、昨年4月1日 _ _ ★ _ 一昨年と比べて15万人少なく、過去最低となった。; correctOrder null」 → 「政府が発表した数字によると、昨年4月1日 _ ★ _ _ 一昨年と比べて15万人少なく、過去最低となった。; correctOrder [1,2,3,0]」
  • The accepted assembly is 昨年4月1日現在における15歳未満の人口は..., order 2341. Since the answer key gives q40 answer 3, ★ belongs in the second slot.
  • evidence: input/N1_2018-07/5Mcv4He-jlpt-freq-exam-and-answers.txt lines 1203-1205: q40 正解:3 and assembled order 2 現在 3 における 4 15歳未満の人口 1 は. [order-conflict cleanup, 2026-06-24]