dsh-paper-doc-adapter — Parsed exam paper structural self-consistency check across question type, answer, score and yotta identifier

dsh-paper-doc-adapter reads one parsed exam paper — the paper header plus one row per question — and checks that parse result's own structural self-consistency: that each question records its number and stem, that an answer or an explanation is present, that the question scores total the paper score the header states, that the question type comes from the vocabulary you configure, that the difficulty coefficient falls inside your range, that question identifiers are present and unique, that a knowledge point is recorded, and that no template placeholder survives in the stem.

What it looks like

Real output from this plugin over its own PD-006 test fixture — not a mock-up. The rule pack ships no invented quotations, so a finding names both the clause it applied and the fact that the clause text was not obtained.

What it answers

You ask	What it answers
Both the question number and the stem are blank on a row — does the check say anything?	Yes. PD-001 reports a row whose questionNo and stem are both empty, because it asks for at least one of the two. It checks that something is filled, not whether the question is well written or the stem rigorous.
A question has its number and stem, but neither an answer nor an explanation.	PD-002 reports that row: it asks for at least one of answer and explanation. It does not check whether the answer is correct — that is subject-expert review, and the pack says as much.
The question scores add up to 98 while the header says 100 — is that caught?	Yes. PD-003 adds the score column and compares the total with the totalScore the paper header states, allowing the configured tolerance of 0.01; if the header total is not a parseable number it reports itself in skipped instead. It only does the addition — whether the marks are distributed sensibly is not judged.
We never listed our own question types. What does the tool do with the question type column?	PD-004 reports itself in skipped, because its values list ships empty: naming question types is your institution's convention and the engine hard-codes no list. Once you configure your values, a questionType outside them is reported; the rule then checks membership only, and it performs no question-type-specific structure check.
The difficulty column holds 1.2 in one row and 中等 in another — what comes back?	PD-005 reports the value it cannot parse as a number and the value outside the configured min/max range of 0–1. That range is your institution's convention, not the only scale — 1–5 or a percentage needs min/max changed or the rule disabled. It checks the range only, never whether the difficulty is appropriate, and it is capped at info.
Two questions carry the same yotta identifier. Why is that worse than a blank one?	PD-006 reports the repeated value, comparing with whitespace ignored, because importing that parse result would silently overwrite an existing question; a blank identifier only keeps that one question out. With no identifier filled anywhere the rule reports itself in skipped. It checks presence and uniqueness, and nothing beyond that.

Standards it follows

Document	Number	Cited by rules
《试题元数据规范》	JY/T 0606 系列（本次未取得条文）	PD-001, PD-002, PD-003, PD-006, PD-007, PD-008
本机构题库与命题口径（本机构配置）	无统一标准（本条依据为本机构配置的题型口径）	PD-004
本机构题库与命题口径（本机构配置）	无统一标准（本条依据为本机构配置的难度口径）	PD-005

Boundary: this plugin checks a 试卷解析结果 for structural self-consistency — that each question records its number and stem, that an answer or an explanation is present, that the question scores total the paper's score, that the question type comes from your vocabulary, that the difficulty coefficient falls inside your range, that question identifiers are present and unique, that a knowledge point is recorded, and that no placeholder survives. It does not decide whether a question is well written, whether an answer is correct, whether the difficulty is appropriate, or whether the paper meets its assessment objectives.

⚠️ What this plugin deliberately does not do

It treats the paper as a generic table and performs no question-type-specific checking. So it does not verify that an answer letter falls within the options offered, that a fill-in-the-blank has the right number of blanks, or that a multiple-choice question has exactly one correct option. Those need the structure of each question type parsed, which this adapter does not attempt. That boundary is stated in the pack's header and in the troubleshooting section.

It never checks whether data is fit to import into a question bank beyond key presence and uniqueness. A duplicate identifier is the dangerous case — importing would silently overwrite an existing question — which is why PD-006 is warn and why it reports itself in skipped when no identifier column is present rather than checking the wrong column.

The difficulty range 0–1 is a local convention, not the only scale. It is the common convention, larger meaning easier; a unit using 1–5 or a percentage can change min/max or disable the rule.

Every excerpt in the rule pack says, in so many words, that the clause text was not obtained — the verification pass retrieved neither JY/T 0606's text nor the yotta platform's field specification. When the texts are in hand, replace each excerpt with the real clause and raise kind to direct.

Compatibility

Surface	Status
Harness	Peer range >=0.1.2-rc.1 <0.2.0 || >=0.2.0-0 <0.3.0 — verified to accept both 0.2.0-rc.2 and 0.2.1-alpha.1. engines.dsh is deliberately not declared: it has no reader and cannot reject a host
Node	`^22.19.0
Platforms	All (plain ESM; no native code, no network, no model call)
Tool mode	Works in native, ptc and both; for a full paper use ptc

What it does

Registers the paper_doc_adapter tool. It reads one parsed paper — the paper header plus one row per question — applies a versioned rule pack, and returns a report.

Rule	Check	Severity	Basis kind
PD-001	the question number and stem are recorded	warn	principle
PD-002	an answer or an explanation is present	warn	principle
PD-003	the question scores total the paper score	warn	principle
PD-004	the question type comes from your vocabulary (off by default)	info	local
PD-005	the difficulty coefficient is inside your range	info	local
PD-006	question identifiers are present and unique	warn	principle
PD-007	a knowledge point is recorded	warn	principle
PD-008	the stem holds no unreplaced placeholder	warn	principle

Install

dsh plugin --profile <name> add dsh-paper-doc-adapter
dsh --profile <name> --dump-config | grep 'dsh-paper-doc-adapter'

Configuration

Key	Type	Default	Description
rulesFile	string	rules/paper-doc-adapter.yaml	Rule-pack path, relative to the package root
disabledRules	string[]	[]	Rule ids to stop running; each appears in skipped
onlyRules	string[]	[]	Run only these rule ids; empty runs every rule
skipNotes	string	""	Note appended to every skipped reason
timeoutMs	number	120000	Cooperative tool timeout budget

Rule-level parameters worth knowing:

• PD-003 field / headerField / tolerance — the score column and the header total it must match.
• PD-004 values — your question types, e.g. [单选题, 多选题, 判断题, 填空题, 简答题, 计算题, 作文题]. Empty means no check.
• PD-005 min / max — the difficulty range, 0 to 1 by default. A local convention, not a standard figure.
• PD-008 terms — the placeholders to look for.

Material format

The tool accepts JSON or YAML:

paperTitle: 某某科目 2026 年模拟试卷
subject: 某某科目
grade: 高二
paperVersion: 2026 版
totalScore: '3'
durationMin: '90'
rows:
  - { 序号: '1', 题号: '1', 题型: 单选题, 题干: 下列关于……的说法，正确的是（　　）,
      选项: A. ……；B. ……；C. ……；D. ……, 答案: B,
      解析: 本题考查……，依据……，故 B 正确。, 分值: '3', 难度: '0.72',
      知识点: 某某章节 某某知识点, 题目ID: YOTTA-2026-0001, 标签: 基础题 }

Column names are matched case-insensitively and ignoring spaces, underscores and hyphens; the paper's own column names are kept, so a finding names the column it read. Note that 序号 is read as the row's display number — give the question number its own column (题号).

Rule sources

Rule data lives in rules/paper-doc-adapter.yaml. Its header explains what the plugin does not check, and each rule's note repeats the part that matters for that rule. The load-time guard still requires a document, clause, excerpt and source per rule, and still forbids a locally configured check from being error.

Troubleshooting

• It did not notice that an answer letter is outside the options. By design: the plugin treats the paper as a generic table and performs no question-type-specific validation.
• PD-006 reports itself as skipped. No identifier column is present, so the uniqueness check has nothing to key on. It reports that rather than checking the wrong column.
• PD-005 fires on a difficulty of 1.8. The default range is 0–1; change min/max if your bank uses another scale.
• PD-003 fires although the paper adds up. The header total and the sum of the rows disagree — usually a missing question or a mistyped score.
• PD-004 never runs. Its vocabulary is empty; fill it with your bank's question types.
• The plugin installs but the tool never appears. Check that main resolves to lib/index.mjs and that pnpm run build produced it; a wrong main makes the loader skip the entry silently.
• dsh plugin add refuses the package as incompatible. The peer range covers 0.1.x and 0.2.x; if your runtime sits outside it, grant an explicit exemption: dsh plugin --profile <name> allow-version [email protected] --dsh-version <runtime> --accept-risk
• check reports manifest-peers as failed. The static checker compares against a hard-coded peer range that predates the 0.2 line. The runtime enforces peer compatibility at install time, so the declared range is the correct one; this is a known upstream issue in dsh-plugin-dev.

Development

pnpm install
pnpm run typecheck   # tsc --noEmit
pnpm test            # vitest, the shared table-plugin suite plus paired fixtures
pnpm run build       # tsdown -> lib/index.mjs + lib/index.d.mts
node ../scripts/sync-shared.mjs dsh-paper-doc-adapter   # refresh src/shared from ../_shared

The plugin is data-only: src/model.ts declares the table shape, the shared kit supplies the reader and the check engine, and the rule pack declares every check.

License

Apache License 2.0 © 2026 dsh-paper-doc-adapter contributors.