20:"$Sreact.view_transition" 21:I[9754,["/_next/static/chunks/2t36f5p4pjzlb.js","/_next/static/chunks/1gyqeidjm5eoa.js","/_next/static/chunks/0_fjnxc4tt_l_.js","/_next/static/chunks/1ymt1shyhmu-4.js","/_next/static/chunks/3kv2zf0i7s2wc.js","/_next/static/chunks/1vs_3g_g75zmk.js","/_next/static/chunks/2g4dh802s8oz3.js","/_next/static/chunks/1bc3f0shukf1j.js","/_next/static/chunks/192kgs9v_t14j.js","/_next/static/chunks/0z6pw0v5c_lx5.js","/_next/static/chunks/08r60e6icmtkc.js","/_next/static/chunks/3trx6c0i7_x-3.js","/_next/static/chunks/40sro5bzp9bi_.js","/_next/static/chunks/1wgf3tk38y6cf.js"],"ProjectDetail"] 22:T497f, Research Units Pipeline — 让研究留下痕迹，而不是答案

Research Units Pipeline · 语义执行单元

让研究留下痕迹，
而不止是一个答案。

这套工具把研究当作证据上的复式记账：每一个交付物都是一条带签名的分录，只有当所有签名汇入同一份清单，一次研究才能算作完成。

RUN SUMMARYBALANCE

执行单元 units

total

工作流检查

PASS/ALL

31/31

目标产物 present

count

75/75

模板残留 residue

ghost

0/226

RUN COMPLETE

reconciled

BALANCED

复式账本 · 平账才是完成

FOLIO 01 / 账本结构

每一个"完成"单元格，首先是一条被背书的账行。

一次 Run 被拆成最小的语义单元。每一行都记录它产出的文件、计算出的哈希，以及一次必须盖在证据上的背书。账本不关心某句话写得是否漂亮——它只核对这份记录能不能对齐。

EVIDENCE LINE一行分录 = 产物名 + hash + 背书签章

FOLIO 02 / 对账合约

只有这些条目互相一致，DONE 才真正算数。

单元格标记为 DONE 本身不算成功。它必须与下面一整套记录对得上——任何一项不一致，账就失去平衡，"完成"也随之失效。

ATTEMPT一次真正执行过的尝试，有据可查

REQUIRED OUTPUTS合约要求的产物必须确实存在

ARTIFACT HASHES每个产物的哈希都稳定一致

WORKFLOW CHECKS工作流自检全部通过

MANIFEST汇入同一份清单

COMPLETION EVENT发出完成事件，收尾一致

ALL SIGNED · 全部对账

FOLIO 03 / 被审计的 Run

49 个单元，31/31 项检查，75/75 个产物，0/226 句残留。

当前发布的回放，在现行合约下完成了全部 49 个单元。结果不是三张指标卡，而是逐项对上的账目：检查全数通过、交付物全数到位、模板占位一句不剩。

执行单元，全部完成

31/31

工作流检查全数 PASS

75/75

目标产物全数存在

0/226

模板残留清零

RUN COMPLETERECONCILED

FOLIO 04 / 工作流账册

每一项工作流，都是一册登记了交付物的账页。

不同的研究路径被编成独立的工作流，每一项都声明自己必须交付什么。列出的是账册条目，而不是臆造的指标。

research-brief研究简报

paper-review论文评审

evidence-review证据综述

arxiv-survey论文概览

arxiv-survey-latexLaTeX 版概览

idea-brainstorm想法发散

source-tutorial源码教程

FOLIO 05 / 审计工具架

让账目的每一行都能被复算。

运行 CLI 需要指定的环境；生成 LaTeX 或 PDF 交付品另有工具配合。以下只登记实际需要的工具。

Python 3.10+运行基础
uv依赖与 CLI
latexmkLaTeX 编排
XeLaTeX文本排版
BibTeX参考文献
pdfinfoPDF 核验

注意：当前证据基于单一主题的回放，跨主题的稳定性仍需进一步验证。

结账

研究应当留下一条可审计的痕迹，而不是一句难以追溯的话——每一行都对齐，账目才有平账的一刻。

AUDITED — RUN COMPLETE

23:T4963, Research Units Pipeline — research leaves a trail, not just an answer

Research Units Pipeline · Semantic execution units

Research leaves a trail,
not just an answer.

This tool treats research as bookkeeping on evidence: every deliverable is a signed line, and a study only counts when every signature reconciles into a single manifest.

RUN SUMMARYBALANCE

Units executed

total

Workflow checks

PASS/ALL

31/31

Target artifacts

present

75/75

Template residue

ghost

0/226

RUN COMPLETE

reconciled

BALANCED

Double-entry ledger · a run is balanced

FOLIO 01 / Ledger anatomy

Every “done” cell first has to be an endorsed line.

A run is broken into the smallest semantic units. Each line records the file it produces, the hash computed from it, and an endorsement that has to be stamped onto the evidence. The ledger never judges whether a sentence reads well — it only checks whether the record stays balanced.

EVIDENCE LINEone entry = artifact name + hash + attestation stamp

FOLIO 02 / The reconciliation contract

Only when these agree, does DONE actually count.

Marking a cell DONE is not success on its own. It has to line up with all of the records below — if any single item is off, the ledger goes out of balance and “complete” stops holding.

ATTEMPTA run that was truly executed and is traceable

REQUIRED OUTPUTSThe deliverables the contract demands must exist

ARTIFACT HASHESEvery artifact hashes to a stable value

WORKFLOW CHECKSThe workflow self-checks all pass

MANIFESTEverything resolves into one manifest

COMPLETION EVENTA completion event fires so the close is consistent

ALL SIGNED · RECONCILED

FOLIO 03 / The audited run

49 units, 31/31 checks, 75/75 artifacts, 0/226 residue.

The current published replay completes all 49 units under the existing contract. The result is not a row of metric cards; it is a set of accounts that each reconcile — all checks pass, all deliverables are in place, and not a single template placeholder is left.

Units executed, all complete

31/31

Workflow checks PASS

75/75

Target artifacts present

0/226

Template residue cleared

RUN COMPLETERECONCILED

FOLIO 04 / The workflow folios

Each workflow is a folio that declares its deliverable.

Different research paths are filed as separate workflows, and each one declares what it must deliver. What is listed here are catalog entries — not invented metrics.

research-briefStructured research brief

paper-reviewPaper review

evidence-reviewEvidence review

arxiv-surveyArXiv survey

arxiv-survey-latexLaTeX survey

idea-brainstormIdea brainstorm

source-tutorialSource tutorial

FOLIO 05 / The toolshed

So every line of the ledger can be recomputed.

The CLI requires a defined environment, and LaTeX or PDF delivery relies on further tools. Only the tools actually needed are recorded here.

Python 3.10+Runtime base
uvDependencies & CLI
latexmkLaTeX orchestration
XeLaTeXTypesetting
BibTeXBibliography
pdfinfoPDF verification

Note: current evidence comes from a single-topic replay; cross-topic stability is still pending verification.

Reconciliation

Research should leave an auditable trail, not a sentence that cannot be traced — every line must line up before a ledger can ever be balanced.

AUDITED — RUN COMPLETE

a:[["$","$20",null,{"enter":{"secondary-forward":"secondary-forward","secondary-back":"secondary-back-enter","default":"none"},"exit":{"secondary-forward":"secondary-forward","secondary-back":"secondary-back-exit","default":"none"},"update":{"secondary-forward":"secondary-forward","secondary-back":"secondary-back-update","default":"none"},"default":"none","children":["$","$L21","cmsrckoqf0149zn19b22m1fod:zh-CN:beautified",{"project":{"id":"cmsrckoqf0149zn19b22m1fod","slug":"willoscar-research-units-pipeline-skills","name":"research-units-pipeline-skills","description":"将研究流程转化为可检查、可恢复的语义执行单元","authorName":"willoscar","authorVerification":{"verified":false,"verifiedAt":null},"category":"development_tools","subcategory":"data_ai_tools","experienceUrl":null,"qualityScore":54,"featured":false,"officialSite":null,"views":{"original":{"available":true},"official":{"available":false,"reason":"disabled_by_policy"},"beautified":{"available":true,"reason":null},"defaultView":"beautified"},"beautifiedView":{"revisionId":"cmt2zh3xq02176271c7hovfsf","kind":"artifact","artifact":{"schemaVersion":"localized-static-html-v4","kind":"static_html","locales":["zh-CN","en"],"defaultLocale":"zh-CN","documents":{"zh-CN":"$22","en":"$23"},"designDirections":{"schemaVersion":"dodot-design-directions-v1","directions":[{"id":"dir-ledger","name":"The Endorsement Ledger","projectMetaphor":"Research is bookkeeping on evidence: every deliverable is a signed line, and completion is the point where all signatures reconcile into a manifest.","signatureVisual":"A vertical audit strip of ledger rows: each row shows an artifact name, its hash, and an attestation stamp, with rows accumulating until a single bold 'RUN COMPLETE' total reconciles them.","heroGeometry":"A left-aligned column of stacked, edge-to-edge ledger rows that grow downward like an accumulating register, ending in a balanced 'balance-sheet' footer band.","sectionSequence":["This Research Leaves A Trail, Not Just An Answer","Every Cell Is An Endorsed Line (the ledger anatomy)","The Reconciliation Contract (what PASS actually requires)","The Audited Run (49 units, 31/31 checks, 75/75 artifacts, 0/226 residue)","Workflows As Folios (research-brief, paper-review, arxiv-survey, idea-brainstorm)","The Toolshed For Audit (Python 3.10+, uv, latexmk, XeLaTeX, BibTeX, pdfinfo)"],"surfaceModel":"Paper-accounting surfaces: flat ruled ledger lines and faint column separators, hard-edged stamps rendered as solid registration marks, and a single textured matte background; no glows or gradients, only printed-ink contrasts.","palette":"Ink black on pale calendered paper (#F4F1E8 / #1A1A18), with one bloodless 'seal red' (#B3382E) reserved exclusively for the attestation stamps and reconciliation total.","typography":"Monospace for every hash and ledger id (JetBrains Mono), a slab/condensed family for ledger headers and section titles tracking the register rhythm, and a clean sans for narrative body.","desktopComposition":"A persistent left gutter column showing the accumulating ledger, a center canvas of narrative folios, and a right-end 'manifest binding' rail that grows as sections advance.","mobileTransformation":"The gutter becomes a horizontally scrollable ledger strip pinned at top; every narrative folio stacks vertically with its evidence ledger row directly beneath it.","mediaStrategy":"No media region. The ledger rows, hashes, and reconciliation total are the composed visual; typography and rules carry evidence, nothing is faked.","forbiddenPatterns":["three generic metric cards","terminal or browser shell framing","decorative glow, grid, or particle fields"],"compositionFingerprint":"Vertical accumulating ledger-strip on the left with edge-to-edge signed rows reconciling into one bold total; every section carries its own evidence line beneath its title."},{"id":"dir-gate","name":"The Acceptance Gate","projectMetaphor":"Research is a sequence of gates: a DONE cell is meaningless until evidence passes every gate into a final PASS, so the pipeline reads as a flight of checkpoints.","signatureVisual":"A diagonal flight of gate posts, each with an artifact flowing through and a stamp that flips from open to sealed; the final gate glows only with the verified PASS attestation.","heroGeometry":"A descending stepped baseline from top-left to bottom-right, with four stamping gates spaced along the diagonal like a staircase of approvals.","sectionSequence":["DONE Is Not Enough — Evidence Must Pass The Gate","The Flight Of Gates (attempt → outputs → hashes → manifest)","The Verdict Contract (what PASS means)","A Replayed Verdict (49 units, 31/31 checks, 75/75 artifacts, 0/226 residue)","The Workflow Gates Catalog (arxiv-survey-latex, paper-review, evidence-review, source-tutorial)","The Gatehouse Requirements (Python 3.10+, uv, LaTeX toolchain)"],"surfaceModel":"Engineered checkpoint surfaces: hard gate frames with knurled edge ticks, sharply cut diagonal planes, and an uncluttered inspection lane; accents by sharp linework, no soft blur, minimal radius.","palette":"Cool engineered charcoal-graphite (#1F2328 / #E8EAED) with one critical 'verdict amber' (#D97706) marking only the open/closed gate states and the final sealed PASS.","typography":"A high-contrast grotesk for gate titles and section headers, monospace for checkpoint ids and hash signatures, and a compact sans for body copy occupying the inspection lane.","desktopComposition":"A wide stepped canvas where narrative text rests on lower steps while artifacts travel the upper diagonal; verdict reads from bottom-right after each gate.","mobileTransformation":"The diagonal staircase is unrolled into a vertical queue: each narrative block is followed by its gate as a full-width horizontal bar with a left 'sealed/open' stamp toggle.","mediaStrategy":"No media region. The traveling artifact block and consecutive gate stamps are the composed visual; the PASS verdict is typographic evidence, nothing is invented.","forbiddenPatterns":["a three-card status/metric wall","editor or dashboard chrome","glow and particle decoration as decoration"],"compositionFingerprint":"Stepped diagonal of gates from top-left to bottom-right where each checkpoint flips open-to-sealed and the terminal PASS is emphasized; no two gates share the same stamp detail."},{"id":"dir-knot","name":"The Artifact Knot","projectMetaphor":"Research is a binding of artifacts: sources, hashes, and outputs are filaments that converge and tie into a single manifest, so the page reads as threads resolving to one knot.","signatureVisual":"A radial knot diagram: concentric thread rings of artifacts, hashes, and checks that all terminate at a central manifest node, with filament tension marking agreed or disputed points.","heroGeometry":"Concentric rings contracting toward one off-center manifest hub, where outer threads (claims) are taut and a broken/loose thread marks where a guardrail catches hollow writing.","sectionSequence":["Research Is What Binds The Evidence Together","The Fibers Of A Run (artifact, hash, output, decision)","The Tension Of Verification (what must agree for a knot to hold)","A Knotted Verification (49 units, 31/31 checks, 75/75 artifacts, 0/226 residue)","The Loom Workflows (research-brief, paper-review, arxiv-survey-latex, idea-brainstorm)","The Tools Weave With (Python 3.10+, uv, latexmk, XeLaTeX, BibTeX, pdfinfo)"],"surfaceModel":"Textile-mechanical surfaces: layered thread paths drawn as 1px filament lines over a quiet yarn field, knot nodes as dense dots where many threads meet, and generous negative space around the hub.","palette":"Indigo fiber threads (#312E81) on a pale warp-field (#F7F7FB) with a single break-tension 'flax' tone (#C8B273) used only for the loose/guarded thread.","typography":"A humanist serif for narrative and knot annotations, a distinct geometric sans for section titles radiating outward, and monospace for hash and fiber identifiers.","desktopComposition":"An off-center radial hub at left with narrative filaments arcing to text panels on the right; annotations sit along thread tangents instead of in a rigid card grid.","mobileTransformation":"The radial knot flattens into an inverted tree: hub at top, threads fanning down into per-section panels, each panel anchored to the filament it belongs to.","mediaStrategy":"No media region. The converging thread diagram and the broken guardrail thread are the composed visual; evidence is expressed through tension, none of it faked.","forbiddenPatterns":["a three-stat KPI bar","browser/terminal shell container","decorative radial grid or particle spray"],"compositionFingerprint":"A contracting radial knot with taut evidence filaments drawn as distinct 1px paths resolving into one off-center hub; the guardrail is a broken, flax-colored thread no other element reuses."}],"selectedDirectionId":"dir-ledger","selectionRationale":"dir-ledger best expresses the project mechanism: research as evidence-backed, auditable execution — the running ledger strip makes the 'trail not an answer' idea literal, directly represents the hash/manifest reconciliation contract, and turns the 49/31/75/0 evidence into a balanced audit instead of a metric wall. Its paper-accounting surface is materially distinct from dir-gate (engineered diagonal) and dir-knot (concentric textile), and its mobile transformation (living gutter → pinned scroll strip) recomposes rather than stacks. dir-gate risks a generic funnel signifier and dir-knot risks abstraction without a strong correspondence to the concrete ledger evidence in the source."},"designBrief":{"schemaVersion":"dodot-design-brief-v1","selectedDirectionId":"dir-ledger","projectMetaphor":"Research is bookkeeping on evidence: every deliverable is a signed line, and completion is the point where all signatures reconcile into a manifest.","signatureVisual":"A vertical audit strip of ledger rows: each row shows an artifact name, its hash, and an attestation stamp, with rows accumulating until a single bold 'RUN COMPLETE' total reconciles them.","compositionFingerprint":"Vertical accumulating ledger-strip on the left with edge-to-edge signed rows reconciling into one bold total; every section carries its own evidence line beneath its title.","heroGeometry":"A left-aligned column of stacked, edge-to-edge ledger rows that grow downward like an accumulating register, ending in a balanced 'balance-sheet' footer band.","sectionSequence":["This Research Leaves A Trail, Not Just An Answer","Every Cell Is An Endorsed Line (the ledger anatomy)","The Reconciliation Contract (what PASS actually requires)","The Audited Run (49 units, 31/31 checks, 75/75 artifacts, 0/226 residue)","Workflows As Folios (research-brief, paper-review, arxiv-survey, idea-brainstorm)","The Toolshed For Audit (Python 3.10+, uv, latexmk, XeLaTeX, BibTeX, pdfinfo)"],"surfaceModel":"Paper-accounting surfaces: flat ruled ledger lines and faint column separators, hard-edged stamps rendered as solid registration marks, and a single textured matte background; no glows or gradients, only printed-ink contrasts.","palette":"Ink black on pale calendered paper (#F4F1E8 / #1A1A18), with one bloodless 'seal red' (#B3382E) reserved exclusively for the attestation stamps and reconciliation total.","typography":"Monospace for every hash and ledger id (JetBrains Mono), a slab/condensed family for ledger headers and section titles tracking the register rhythm, and a clean sans for narrative body.","desktopComposition":"A persistent left gutter column showing the accumulating ledger, a center canvas of narrative folios, and a right-end 'manifest binding' rail that grows as sections advance.","mobileTransformation":"The gutter becomes a horizontally scrollable ledger strip pinned at top; every narrative folio stacks vertically with its evidence ledger row directly beneath it.","mediaStrategy":"No media region. The ledger rows, hashes, and reconciliation total are the composed visual; typography and rules carry evidence, nothing is faked.","evidencePlan":["Render the 49 Units / 31 of 31 workflow checks PASS / 75 of 75 target artifacts present / 0 of 226 template residue as reconciled ledger lines that sum to ONE 'RUN COMPLETE' total, not as a metric card trio.","A cross-accounting section that displays the five agreeing items named in the source — Attempt, required outputs, Artifact hashes, Workflow checks, Manifest, and Completion Event — as individually stamped ledger rows showing how all must agree for a DONE cell to count.","A folio-index strip listing the source-named workflows (research-brief, paper-review, evidence-review, arxiv-survey, arxiv-survey-latex, idea-brainstorm, source-tutorial) as ledger catalog entries with their declared deliverable, without inventing metrics.","A runtime-requirements ledger section transcribing only the stated tools (Python 3.10+, uv, latexmk, XeLaTeX, BibTeX, pdfinfo) plus the single-topic replay caveat, marked as an audited footnote row rather than a capability boast."],"localeContract":{"zh-CN":"研究留下的应是一条可审计的痕迹，而不只是一个答案。整页采用复式账本隐喻：每一行交付物都是一条带签名的分录，完成即所有签名汇入一份 Manifes（清单）。避免空洞写作的护栏表现为一条未对齐/被阻断的账行。","en":"Research should leave an auditable trail, not just an answer. The whole page adopts a double-entry ledger metaphor: each deliverable is a signed line, and completion is when every signature reconciles into a manifest. The guardrail against hollow writing is shown as a rejected, out-of-balance ledger row."},"forbiddenPatterns":["three generic metric cards","terminal or browser shell framing","decorative glow, grid, or particle fields","fake tabs, search, switches, charts, logos, or testimonials","desktop columns merely stacked on mobile without recomposition"]},"visualReview":{"schemaVersion":"dodot-automated-design-review-v1","status":"approved","summary":"Both locales faithfully realize the selected dir-ledger (Endorsement Ledger) brief: a vertical accumulating ledger strip renders the hash + attestation signature visual reconciling into RUN COMPLETE, all audit figures trace to the material's cited evidence, the mobile layout recomposes the sticky gutter into a horizontally scrollable pinned strip with vertically stacked folios, bilingual structure is equivalent and natural, and no generic hero/card wall/device shell/decorative grid or fake controls appear. The repeating metric blocks in the audited-run section stay evidence-bound with a reconciliation band, so no blocking issue is raised.","checks":{"projectFit":{"passed":true,"evidence":"The page implements the brief's ledger bookkeeping metaphor end to end. Desktop geometry matches desktopComposition: a persistent sticky left gutter carrying the accumulating ledger, a center narrative canvas, and a right manifest-binding rail. Section sequence tracks brief.sectionSequence (hero slogan -> ledger anatomy FOLIO 01 -> reconciliation contract FOLIO 02 -> audited run FOLIO 03 -> workflow folios FOLIO 04 -> toolshed FOLIO 05 -> outro). Palette matches (#F4F1E8 paper / #1A1A18 ink / #B3382E seal red reserved for stamps and total), monospace installed for hashes/ids, and paper-accounting ruled-line surface applied."},"evidenceBound":{"passed":true,"evidence":"All figures are material-derived, none invented. 49 units / 31/31 checks / 75/75 artifacts / 0/226 residue match material.json summary claim 4. The six agreeing items listed as contract rows (Attempt, Required outputs, Artifact hashes, Workflow checks, Manifest, Completion Event) match claim 3. Workflow folios name the exact seven workflows (research-brief, paper-review, evidence-review, arxiv-survey, arxiv-survey-latex, idea-brainstorm, source-tutorial) from claim 2/brief evidencePlan. Toolshed lists only source tools (Python 3.10+, uv, latexmk, XeLaTeX, BibTeX, pdfinfo) from claim 5, and the single-topic replay caveat is preserved as an audited footnote. No speculative metrics or hashes are presented as real facts; abbreviated hashes are stylized placeholders consistent across both locales."},"signatureVisual":{"passed":true,"evidence":"The signature visual is genuinely present and structurally faithful in both locales: .ledger is a vertical audit strip where each .lrow shows an artifact name, a truncated hash, and an attestation .stamp, accumulating until a bold .lrow.total 'RUN COMPLETE' with RECONCILED stamp reconciles them (role=img + descriptive aria-label reinforce intent). The hero additionally renders the four audit totals as ledger rows summing to a BALANCED RUN COMPLETE. This directly matches brief.signatureVisual and compositionFingerprint."},"mobileTransformation":{"passed":true,"evidence":"At max-width:720px the three-column grid collapses to a single column and the elements are re-ordered: the gutter becomes position:relative, the ledger becomes a horizontally scrollable pinned strip (.ledger{overflow-x:auto} with .lstrip{min-width:340px}), and each narrative folio (main.lane order:2) stacks vertically with the ledger above. The right rail is unhooked (display:none), the 4-col metric grid becomes 2-col, and folios collapse to one column. This is a genuine recomposition (scroll strip + reorder), not a passive stacking of desktop columns, matching brief.mobileTransformation."},"localeParity":{"passed":true,"evidence":"zh-CN and en share identical structure, section order, counts, stamps, and 7 workflow/tool names. zh-CN uses native phrasing (e.g., hero '让研究留下痕迹' vs 'Research leaves a trail') and en is idiomatic and natural. The reconciliation-contract, audited-run, workflow, and toolshed sections carry the same six key items, the same 49/31/31/75/75/0/226 figures, and the same single-topic caveat in each language. Language attribute and fonts are locale-appropriate."},"antiTemplate":{"passed":true,"evidence":"No generic hero, no three-card metric wall as the primary visual, no terminal/browser shell framing, and no decorative glow, gradient-grid, or particle field. The ruled repeating-linear-gradient background is the brief's mandated paper-accounting line surface, not decorative. The audited-run block renders the four reconciliation totals as bordered items but keeps them evidence-bound and subordinates them to a RUN COMPLETE / RECONCILED balance band, and the same numbers already appear as ledger rows in the hero, so it reads as an audit balance rather than a generic KPI trio/card wall. Content is composed typographically around the ledger metaphor throughout."},"noFakeUi":{"passed":true,"evidence":"No fake tabs, search, switches, charts, logos, or testimonials exist. The reconciliation-contract uses semantic role=list/listitem rows. The right rail is a plain

of text spans, not clickable fake controls. The toolshed is a real list. QA reports fakeControlCount=0, brokenImageCount=0, emptyLinkCount=0 for both locales/viewports with no console errors or network requests."}},"issues":[{"code":"metric-block-style","severity":"warning","locale":"all","description":"brief.evidencePlan explicitly directs rendering the 4 audit figures (49 / 31/31 / 75/75 / 0/226) 'as reconciled ledger lines that sum to one RUN COMPLETE total, not as a metric card trio'. The page does this correctly in the hero ledger strip, but the audited-run section (FOLIO 03) repeats the same numbers again as bordered summary cells (sumra/sumc) before its reconcile band.","repairInstruction":"To match the brief strictly, replace the sumra bordered cells in FOLIO 03 with ledger-style rows (same .lrow pattern used in the hero) so the audited-run reads as reconciled ledger lines rather than a secondary metric grid, keeping them cumulative toward the RUN COMPLETE reconciliation band."}]},"repairEvidence":[{"attempt":1,"before":{"qaPassed":false,"designReviewApproved":false,"qaEvidence":{"zh-CN":[{"viewport":"desktop","passed":true,"localePassed":true,"metrics":{"language":"zh-CN","h1Count":1,"h2Count":5,"sectionCount":6,"textLength":1783,"cjkCount":718,"latinLetterCount":632,"scrollWidth":1280,"clientWidth":1280,"scrollHeight":3721,"h1Top":164.140625,"bottomInset":79.703125,"brokenImageCount":0,"missingAltCount":0,"fakeControlCount":0,"emptyLinkCount":0},"screenshotHashes":{"top":"d16e82f665e7fedcabdbd4ee0b5e653a25e3b842c30290009d8c296332d3d829","middle":"83b987ca789a2a06814129b68106a6a687085afe451289a950fb1de7ea02f687","bottom":"99b21b4954492c0ce176554c2b8ad84ca687dcc2add3b3d2aaaba407945658f2"},"consoleErrors":[],"networkRequests":[]},{"viewport":"mobile","passed":true,"localePassed":true,"metrics":{"language":"zh-CN","h1Count":1,"h2Count":5,"sectionCount":6,"textLength":1646,"cjkCount":692,"latinLetterCount":560,"scrollWidth":390,"clientWidth":390,"scrollHeight":4320,"h1Top":552.484375,"bottomInset":79.8125,"brokenImageCount":0,"missingAltCount":0,"fakeControlCount":0,"emptyLinkCount":0},"screenshotHashes":{"top":"9b3211443f87220900cba4d3963bebd607a735c4f1ac1eefaca0b06571c54afa","middle":"277b199e31c9d0e33fa48094bbcbe261540ada2704b4fe81be9de1e467f01b4a","bottom":"d1b04e8540f585b150e1a896b95e7bab62c4101a46a46e7041279a47e238da49"},"consoleErrors":[],"networkRequests":[]}],"en":[{"viewport":"desktop","passed":false,"localePassed":true,"metrics":{"language":"en","h1Count":1,"h2Count":5,"sectionCount":6,"textLength":3577,"cjkCount":0,"latinLetterCount":2805,"scrollWidth":1328,"clientWidth":1280,"scrollHeight":3964,"h1Top":164.140625,"bottomInset":79.78125,"brokenImageCount":0,"missingAltCount":0,"fakeControlCount":0,"emptyLinkCount":0},"screenshotHashes":{"top":"25d1d8dfb1e1d94c63f87e75a481f0ec684a631809f2632cda30537d83f22c5b","middle":"0c6bccf4fb95a9e179cc26fc93e47142806d29d937eead55bcb27f28edd66eed","bottom":"08c0219f16f2d12e3d40b822cc62920e785c5422f923213a2cc51614ad128253"},"consoleErrors":[],"networkRequests":[]},{"viewport":"mobile","passed":true,"localePassed":true,"metrics":{"language":"en","h1Count":1,"h2Count":5,"sectionCount":6,"textLength":3377,"cjkCount":0,"latinLetterCount":2647,"scrollWidth":390,"clientWidth":390,"scrollHeight":4788,"h1Top":570.625,"bottomInset":80.0625,"brokenImageCount":0,"missingAltCount":0,"fakeControlCount":0,"emptyLinkCount":0},"screenshotHashes":{"top":"7163b690e0931132db0d1525753812e24a94485dcaa0b39cc0e2acae8e8398c7","middle":"183ad11faaf1ee98d5adf58a3269d43928d34d0ccfd15dcf02e0878455565dd5","bottom":"3e5dc4a248142ca8406ea0284998f76fd95d1b158cdb7be12d95ca68447defc1"},"consoleErrors":[],"networkRequests":[]}]},"visualReview":null},"after":{"qaPassed":true,"designReviewApproved":true,"qaEvidence":{"zh-CN":[{"viewport":"desktop","passed":true,"localePassed":true,"metrics":{"language":"zh-CN","h1Count":1,"h2Count":5,"sectionCount":6,"textLength":1783,"cjkCount":718,"latinLetterCount":632,"scrollWidth":1280,"clientWidth":1280,"scrollHeight":3721,"h1Top":164.140625,"bottomInset":79.703125,"brokenImageCount":0,"missingAltCount":0,"fakeControlCount":0,"emptyLinkCount":0},"screenshotHashes":{"top":"d16e82f665e7fedcabdbd4ee0b5e653a25e3b842c30290009d8c296332d3d829","middle":"83b987ca789a2a06814129b68106a6a687085afe451289a950fb1de7ea02f687","bottom":"99b21b4954492c0ce176554c2b8ad84ca687dcc2add3b3d2aaaba407945658f2"},"consoleErrors":[],"networkRequests":[]},{"viewport":"mobile","passed":true,"localePassed":true,"metrics":{"language":"zh-CN","h1Count":1,"h2Count":5,"sectionCount":6,"textLength":1646,"cjkCount":692,"latinLetterCount":560,"scrollWidth":390,"clientWidth":390,"scrollHeight":4320,"h1Top":552.484375,"bottomInset":79.8125,"brokenImageCount":0,"missingAltCount":0,"fakeControlCount":0,"emptyLinkCount":0},"screenshotHashes":{"top":"9b3211443f87220900cba4d3963bebd607a735c4f1ac1eefaca0b06571c54afa","middle":"277b199e31c9d0e33fa48094bbcbe261540ada2704b4fe81be9de1e467f01b4a","bottom":"d1b04e8540f585b150e1a896b95e7bab62c4101a46a46e7041279a47e238da49"},"consoleErrors":[],"networkRequests":[]}],"en":[{"viewport":"desktop","passed":true,"localePassed":true,"metrics":{"language":"en","h1Count":1,"h2Count":5,"sectionCount":6,"textLength":3577,"cjkCount":0,"latinLetterCount":2805,"scrollWidth":1280,"clientWidth":1280,"scrollHeight":3964,"h1Top":164.140625,"bottomInset":79.78125,"brokenImageCount":0,"missingAltCount":0,"fakeControlCount":0,"emptyLinkCount":0},"screenshotHashes":{"top":"e2bbce1d9cda39cfe473e307fc9467edfce410892ab42873b0964af10a415593","middle":"2c4a59cf5527c2f98e5fe2a549a043810f4d0db521cb37af209e6c2c0a71faee","bottom":"d2a847ab127af31a9ec017e88e56f26138b0898e9508eafda365abc6e54e8029"},"consoleErrors":[],"networkRequests":[]},{"viewport":"mobile","passed":true,"localePassed":true,"metrics":{"language":"en","h1Count":1,"h2Count":5,"sectionCount":6,"textLength":3377,"cjkCount":0,"latinLetterCount":2647,"scrollWidth":390,"clientWidth":390,"scrollHeight":4788,"h1Top":570.625,"bottomInset":80.0625,"brokenImageCount":0,"missingAltCount":0,"fakeControlCount":0,"emptyLinkCount":0},"screenshotHashes":{"top":"7163b690e0931132db0d1525753812e24a94485dcaa0b39cc0e2acae8e8398c7","middle":"183ad11faaf1ee98d5adf58a3269d43928d34d0ccfd15dcf02e0878455565dd5","bottom":"3e5dc4a248142ca8406ea0284998f76fd95d1b158cdb7be12d95ca68447defc1"},"consoleErrors":[],"networkRequests":[]}]},"visualReview":{"schemaVersion":"dodot-automated-design-review-v1","status":"approved","summary":"Both locales faithfully realize the selected dir-ledger (Endorsement Ledger) brief: a vertical accumulating ledger strip renders the hash + attestation signature visual reconciling into RUN COMPLETE, all audit figures trace to the material's cited evidence, the mobile layout recomposes the sticky gutter into a horizontally scrollable pinned strip with vertically stacked folios, bilingual structure is equivalent and natural, and no generic hero/card wall/device shell/decorative grid or fake controls appear. The repeating metric blocks in the audited-run section stay evidence-bound with a reconciliation band, so no blocking issue is raised.","checks":{"projectFit":{"passed":true,"evidence":"The page implements the brief's ledger bookkeeping metaphor end to end. Desktop geometry matches desktopComposition: a persistent sticky left gutter carrying the accumulating ledger, a center narrative canvas, and a right manifest-binding rail. Section sequence tracks brief.sectionSequence (hero slogan -> ledger anatomy FOLIO 01 -> reconciliation contract FOLIO 02 -> audited run FOLIO 03 -> workflow folios FOLIO 04 -> toolshed FOLIO 05 -> outro). Palette matches (#F4F1E8 paper / #1A1A18 ink / #B3382E seal red reserved for stamps and total), monospace installed for hashes/ids, and paper-accounting ruled-line surface applied."},"evidenceBound":{"passed":true,"evidence":"All figures are material-derived, none invented. 49 units / 31/31 checks / 75/75 artifacts / 0/226 residue match material.json summary claim 4. The six agreeing items listed as contract rows (Attempt, Required outputs, Artifact hashes, Workflow checks, Manifest, Completion Event) match claim 3. Workflow folios name the exact seven workflows (research-brief, paper-review, evidence-review, arxiv-survey, arxiv-survey-latex, idea-brainstorm, source-tutorial) from claim 2/brief evidencePlan. Toolshed lists only source tools (Python 3.10+, uv, latexmk, XeLaTeX, BibTeX, pdfinfo) from claim 5, and the single-topic replay caveat is preserved as an audited footnote. No speculative metrics or hashes are presented as real facts; abbreviated hashes are stylized placeholders consistent across both locales."},"signatureVisual":{"passed":true,"evidence":"The signature visual is genuinely present and structurally faithful in both locales: .ledger is a vertical audit strip where each .lrow shows an artifact name, a truncated hash, and an attestation .stamp, accumulating until a bold .lrow.total 'RUN COMPLETE' with RECONCILED stamp reconciles them (role=img + descriptive aria-label reinforce intent). The hero additionally renders the four audit totals as ledger rows summing to a BALANCED RUN COMPLETE. This directly matches brief.signatureVisual and compositionFingerprint."},"mobileTransformation":{"passed":true,"evidence":"At max-width:720px the three-column grid collapses to a single column and the elements are re-ordered: the gutter becomes position:relative, the ledger becomes a horizontally scrollable pinned strip (.ledger{overflow-x:auto} with .lstrip{min-width:340px}), and each narrative folio (main.lane order:2) stacks vertically with the ledger above. The right rail is unhooked (display:none), the 4-col metric grid becomes 2-col, and folios collapse to one column. This is a genuine recomposition (scroll strip + reorder), not a passive stacking of desktop columns, matching brief.mobileTransformation."},"localeParity":{"passed":true,"evidence":"zh-CN and en share identical structure, section order, counts, stamps, and 7 workflow/tool names. zh-CN uses native phrasing (e.g., hero '让研究留下痕迹' vs 'Research leaves a trail') and en is idiomatic and natural. The reconciliation-contract, audited-run, workflow, and toolshed sections carry the same six key items, the same 49/31/31/75/75/0/226 figures, and the same single-topic caveat in each language. Language attribute and fonts are locale-appropriate."},"antiTemplate":{"passed":true,"evidence":"No generic hero, no three-card metric wall as the primary visual, no terminal/browser shell framing, and no decorative glow, gradient-grid, or particle field. The ruled repeating-linear-gradient background is the brief's mandated paper-accounting line surface, not decorative. The audited-run block renders the four reconciliation totals as bordered items but keeps them evidence-bound and subordinates them to a RUN COMPLETE / RECONCILED balance band, and the same numbers already appear as ledger rows in the hero, so it reads as an audit balance rather than a generic KPI trio/card wall. Content is composed typographically around the ledger metaphor throughout."},"noFakeUi":{"passed":true,"evidence":"No fake tabs, search, switches, charts, logos, or testimonials exist. The reconciliation-contract uses semantic role=list/listitem rows. The right rail is a plain

of text spans, not clickable fake controls. The toolshed is a real list. QA reports fakeControlCount=0, brokenImageCount=0, emptyLinkCount=0 for both locales/viewports with no console errors or network requests."}},"issues":[{"code":"metric-block-style","severity":"warning","locale":"all","description":"brief.evidencePlan explicitly directs rendering the 4 audit figures (49 / 31/31 / 75/75 / 0/226) 'as reconciled ledger lines that sum to one RUN COMPLETE total, not as a metric card trio'. The page does this correctly in the hero ledger strip, but the audited-run section (FOLIO 03) repeats the same numbers again as bordered summary cells (sumra/sumc) before its reconcile band.","repairInstruction":"To match the brief strictly, replace the sumra bordered cells in FOLIO 03 with ledger-style rows (same .lrow pattern used in the hero) so the audited-run reads as reconciled ledger lines rather than a secondary metric grid, keeping them cumulative toward the RUN COMPLETE reconciliation band."}]}},"changedFiles":["artifact/zh-CN.html","artifact/en.html"]}],"provenance":{"executor":"opencode","model":"deepseek/deepseek-v4-flash","opencodeVersion":"1.1.58","promptVersion":"dodot-showcase-deepseek-v4","inputHash":"b130373750d1b311295a78ff99b4c2b76a2c55cedc4f303e34df82b4a2440c57","artifactHash":"307f5a348640bc45d7901997d2834bfcf855ddfc9545dc644a4d4a98bf62e163","designSkill":"dodot-showcase-design"},"qaEvidence":{"zh-CN":[{"viewport":"desktop","passed":true,"localePassed":true,"metrics":{"language":"zh-CN","h1Count":1,"h2Count":5,"sectionCount":6,"textLength":1783,"cjkCount":718,"latinLetterCount":632,"scrollWidth":1280,"clientWidth":1280,"scrollHeight":3721,"h1Top":164.140625,"bottomInset":79.703125,"brokenImageCount":0,"missingAltCount":0,"fakeControlCount":0,"emptyLinkCount":0},"screenshotHashes":{"top":"d16e82f665e7fedcabdbd4ee0b5e653a25e3b842c30290009d8c296332d3d829","middle":"83b987ca789a2a06814129b68106a6a687085afe451289a950fb1de7ea02f687","bottom":"99b21b4954492c0ce176554c2b8ad84ca687dcc2add3b3d2aaaba407945658f2"},"consoleErrors":[],"networkRequests":[]},{"viewport":"mobile","passed":true,"localePassed":true,"metrics":{"language":"zh-CN","h1Count":1,"h2Count":5,"sectionCount":6,"textLength":1646,"cjkCount":692,"latinLetterCount":560,"scrollWidth":390,"clientWidth":390,"scrollHeight":4320,"h1Top":552.484375,"bottomInset":79.8125,"brokenImageCount":0,"missingAltCount":0,"fakeControlCount":0,"emptyLinkCount":0},"screenshotHashes":{"top":"9b3211443f87220900cba4d3963bebd607a735c4f1ac1eefaca0b06571c54afa","middle":"277b199e31c9d0e33fa48094bbcbe261540ada2704b4fe81be9de1e467f01b4a","bottom":"d1b04e8540f585b150e1a896b95e7bab62c4101a46a46e7041279a47e238da49"},"consoleErrors":[],"networkRequests":[]}],"en":[{"viewport":"desktop","passed":true,"localePassed":true,"metrics":{"language":"en","h1Count":1,"h2Count":5,"sectionCount":6,"textLength":3577,"cjkCount":0,"latinLetterCount":2805,"scrollWidth":1280,"clientWidth":1280,"scrollHeight":3964,"h1Top":164.140625,"bottomInset":79.78125,"brokenImageCount":0,"missingAltCount":0,"fakeControlCount":0,"emptyLinkCount":0},"screenshotHashes":{"top":"e2bbce1d9cda39cfe473e307fc9467edfce410892ab42873b0964af10a415593","middle":"2c4a59cf5527c2f98e5fe2a549a043810f4d0db521cb37af209e6c2c0a71faee","bottom":"d2a847ab127af31a9ec017e88e56f26138b0898e9508eafda365abc6e54e8029"},"consoleErrors":[],"networkRequests":[]},{"viewport":"mobile","passed":true,"localePassed":true,"metrics":{"language":"en","h1Count":1,"h2Count":5,"sectionCount":6,"textLength":3377,"cjkCount":0,"latinLetterCount":2647,"scrollWidth":390,"clientWidth":390,"scrollHeight":4788,"h1Top":570.625,"bottomInset":80.0625,"brokenImageCount":0,"missingAltCount":0,"fakeControlCount":0,"emptyLinkCount":0},"screenshotHashes":{"top":"7163b690e0931132db0d1525753812e24a94485dcaa0b39cc0e2acae8e8398c7","middle":"183ad11faaf1ee98d5adf58a3269d43928d34d0ccfd15dcf02e0878455565dd5","bottom":"3e5dc4a248142ca8406ea0284998f76fd95d1b158cdb7be12d95ca68447defc1"},"consoleErrors":[],"networkRequests":[]}]}},"generatorVersion":"opencode-1.1.58:deepseek/deepseek-v4-flash:dodot-showcase-deepseek-v4","publishedAt":"2026-08-20T10:37:16.343Z"},"rawExtraction":{"projectRevisionId":"cmt2zh3xi02116271u10zltsz","projectVersion":1,"sources":[{"canonicalKey":"github:willoscar/research-units-pipeline-skills","role":"primary","revisionVersion":1,"url":"https://github.com/WILLOSCAR/research-units-pipeline-skills"}],"facts":[{"key":"repository_owner","value":"willoscar","evidence":[{"path":"repository.owner","excerpt":"willoscar"}]},{"key":"repository_name","value":"research-units-pipeline-skills","evidence":[{"path":"repository.name","excerpt":"research-units-pipeline-skills"}]},{"key":"description","value":"Research pipelines as semantic execution units: each skill declares inputs/outputs, acceptance criteria, and guardrails. Evidence-first methodology prevents hollow writing through structured intermediate artifacts.","evidence":[{"path":"repository.description","excerpt":"Research pipelines as semantic execution units: each skill declares inputs/outputs, acceptance criteria, and guardrails. Evidence-first methodology prevents hollow writing through structured intermediate artifacts."}]},{"key":"primary_language","value":"Python","evidence":[{"path":"repository.language","excerpt":"Python"}]},{"key":"topics","value":"claude, claude-code, codex, gpt, pipeline, research, research-paper, research-project, research-tool, skills, tools, units, vibe, vibe-coding, vibecoding","evidence":[{"path":"repository.topics","excerpt":"claude, claude-code, codex, gpt, pipeline, research, research-paper, research-project, research-tool, skills, tools, units, vibe, vibe-coding, vibecoding"}]},{"key":"source_count","value":"1","evidence":[{"path":"projectRevision.sources","excerpt":"github:willoscar/research-units-pipeline-skills"}]}],"taxonomy":{"category":"development_tools","subcategory":"data_ai_tools","confidence":0.65,"evidence":[{"path":"repository.description","excerpt":"Research pipelines as semantic execution units: each skill declares inputs/outputs, acceptance criteria, and guardrails. Evidence-first methodology prevents hollow writing through structured intermediate artifacts."}]},"lifecycle":{"stage":"active","confidence":0.75,"evidence":[{"path":"repository.pushedAt","excerpt":"2026-08-17T01:58:32.000Z"}]},"userIntents":[{"slug":"build_software","label":"构建或改进软件","confidence":0.65,"evidence":[{"path":"repository.description","excerpt":"Research pipelines as semantic execution units: each skill declares inputs/outputs, acceptance criteria, and guardrails. Evidence-first methodology prevents hollow writing through structured intermediate artifacts."}]},{"slug":"automate_work","label":"自动化工作流程","confidence":0.65,"evidence":[{"path":"repository.description","excerpt":"Research pipelines as semantic execution units: each skill declares inputs/outputs, acceptance criteria, and guardrails. Evidence-first methodology prevents hollow writing through structured intermediate artifacts."}]},{"slug":"analyze_data","label":"分析与理解数据","confidence":0.65,"evidence":[{"path":"repository.description","excerpt":"Research pipelines as semantic execution units: each skill declares inputs/outputs, acceptance criteria, and guardrails. Evidence-first methodology prevents hollow writing through structured intermediate artifacts."}]},{"slug":"learn_explore","label":"学习与探索","confidence":0.65,"evidence":[{"path":"repository.description","excerpt":"Research pipelines as semantic execution units: each skill declares inputs/outputs, acceptance criteria, and guardrails. Evidence-first methodology prevents hollow writing through structured intermediate artifacts."}]}],"readme":{"headings":["Research Harness","See A Run In Five Minutes","Choose The Deliverable","What Changes When Research Becomes A Run","What A PASS Means","The Survey Failure That Shaped The Gate","Published Evidence","Runtime Requirements","Maintainer Verification","Documentation","Star History"],"commands":[]},"changeSet":[{"field":"repository","beforeHash":"0dfc21e7f94a088cab309c67c104c20465e48397c975720044110e03a3f0705e","afterHash":"d0d55396a33c86bf450474aa787752074d6ac33d8eceb2c599d0f1d634213a1c"}]},"publishedAt":"2026-08-13T10:00:22.866Z","updatedAt":"2026-08-21T13:26:55.033Z","summary":{"id":"cmt2zh3xk02136271ydjtklua","projectType":"product","oneLine":"将研究流程转化为可检查、可恢复的语义执行单元","overview":"Research Harness 将研究目标转化为文件优先、可恢复的 Run，通过显式工作流组织技能，保留中间产物和决策。它检查可观察的合约，并将失败指向最小的修复面。支持多种工作流，如研究简报、论文评审、证据综述和 LaTeX 调查，强调证据优先，防止空洞写作。","highlights":["文件优先、可恢复的 Run","证据支持的完成检查","多种研究工作流","失败定位与修复诊断"],"bestFor":["需要可审计研究过程的开发者","使用 AI 辅助研究并希望保留证据链的用户","需要生成结构化研究交付物的团队"],"caveats":["需要 Python 3.10+ 和 uv 环境","LaTeX/PDF 交付依赖额外工具","当前证据基于单一主题回放，跨主题稳定性待验证"],"categorySuggestion":"development_tools","subcategorySuggestion":"coding_agents","lifecycleStages":["idea_requirements","product_design","testing_quality"],"audienceIntents":["build","evaluate","learn"],"confidence":0.8,"evidence":[{"claim":"Research Harness 将研究目标转化为文件优先、可恢复的 Run，组织技能为显式工作流，并保留中间产物和决策。","sourceType":"readme","locator":"https://github.com/WILLOSCAR/research-units-pipeline-skills"},{"claim":"提供多种工作流，如 research-brief、paper-review、evidence-review、arxiv-survey、arxiv-survey-latex、idea-brainstorm 和 source-tutorial，每种都有明确的交付物。","sourceType":"readme","locator":"https://github.com/WILLOSCAR/research-units-pipeline-skills"},{"claim":"完成由证据支持：DONE 单元格本身不算成功，需要 Attempt、必需输出、Artifact 哈希、工作流检查、Manifest 和完成事件一致。","sourceType":"readme","locator":"https://github.com/WILLOSCAR/research-units-pipeline-skills"},{"claim":"当前发布的回放完成了当前合约下的全部 49 个单元，包括 31/31 工作流检查通过、75/75 目标产物存在、模板残留 0/226 句。","sourceType":"readme","locator":"https://github.com/WILLOSCAR/research-units-pipeline-skills"},{"claim":"需要 Python 3.10+ 和 uv 运行 CLI；LaTeX/PDF 交付需要 latexmk、XeLaTeX、BibTeX 和 pdfinfo。","sourceType":"readme","locator":"https://github.com/WILLOSCAR/research-units-pipeline-skills"}],"publishedAt":"2026-08-17T13:54:30.041Z"},"github":{"url":"https://github.com/WILLOSCAR/research-units-pipeline-skills","ownerAvatarUrl":"https://avatars.githubusercontent.com/u/63443351?v=4","starsCount":497,"forksCount":39,"language":"Python","syncedAt":"2026-08-17T13:51:13.108Z","license":null,"defaultBranch":"main","archived":false,"disabled":false,"lastChangedAt":"2026-08-17T01:58:32.000Z","missing":false},"tags":[{"name":"Vibe coding","slug":"vibe-coding","verification":"verified","evidenceUrl":"https://github.com/WILLOSCAR/research-units-pipeline-skills","evidenceExcerpt":"GitHub topic: vibe-coding"}],"updates":[]},"entryPoint":"discovery","claim":{"state":"unclaimed","verified":false},"initialView":"beautified","community":{"viewer":null,"state":{"likesCount":0,"creativeCount":0,"followersCount":0,"commentsCount":0,"liked":false,"creative":false,"followed":false},"comments":{"comments":[],"nextCursor":null}}}]}],"$L24"]

让研究留下痕迹，而不止是一个答案。

每一个"完成"单元格，首先是一条被背书的账行。

只有这些条目互相一致，DONE 才真正算数。

49 个单元，31/31 项检查，75/75 个产物，0/226 句残留。

每一项工作流，都是一册登记了交付物的账页。

让账目的每一行都能被复算。

Research leaves a trail,not just an answer.

Every “done” cell first has to be an endorsed line.

Only when these agree, does DONE actually count.

49 units, 31/31 checks, 75/75 artifacts, 0/226 residue.

Each workflow is a folio that declares its deliverable.

So every line of the ledger can be recomputed.

让研究留下痕迹，
而不止是一个答案。

Research leaves a trail,
not just an answer.