Shidiao Notes and Quality Review
Working Principles
First read Sources and Rules to confirm the manuscript basis and scope of rule applicability. For items involving deadlines, page counts, weights, anonymity, AI, or submission methods, verify online using the official documents for the target edition, group, and phase. If verification fails, mark as 'pending verification' but continue research content without hindrance. Do not treat example manuscripts as hard competition rules.
Default language is Chinese. First review the user's existing materials and current progress; only ask for missing information that would change the plan, otherwise state assumptions and proceed. Without raw data, provide only plans, templates to be filled, or clearly marked demonstrations; do not fabricate samples, statistics, interviews, references, or award identities. Deliver output based on the depth of the user's task; partial revisions do not force a full redo of the entire process.
Objective
Based on Chapters 3–4 and 7–9 of the manuscript and the current edition's documents, transform notes into verifiable issues and remediation actions. Depth of review follows the current phase; do not mask design and data issues with format checks.
First Define the Review Scope
Check edition, group, school and region, free/corporate topic, current phase, deadline, and existing materials. Create a rule table: Item | Edition/Group/Phase | Original Text Location & Link | Date Published / Verified On | Official Requirement | Local Implementation Requirement | Status. If documents are not provided or unreadable, list as 'unverified'; do not automatically deem it a violation.
The undergraduate report framework of 20/30/30/20 can serve as a diagnostic reference, but not as official scoring for the 17th edition. Page counts, format, anonymity, plagiarism check, AI disclosure, file naming, and defense duration must reference the appendix of the applicable edition. The national completion date is not the school's material submission deadline; plan backwards from the earlier applicable operational milestone. Especially need to separately verify the international student group and knowledge competition participants.
Phase Gate Checkpoints
| Phase | Evidence to Check | Typical Issues & Fixes |
|---|---|---|
| Initiation | Problem definition, overall scope, channels and permissions, resource table | Scope too large or inaccessible: narrow topic/change design; cannot compensate by adding models |
| Design | Sampling steps, sample planning, problem-variable-item matrix, ethics statement | Claiming random sampling without a sampling frame, questions cannot answer objectives: rewrite sampling and measurement plan |
| Pre-test | Recruitment records, cognitive feedback, timing and skip logic, trial cleaning and analysis, revision table | Only test-filling without checking the full chain: supplement critical process checks; 10 classroom responses are not a generally valid pre-test sample threshold |
| Full Implementation | Training, sample contact and refusal, batches and channels, versions and follow-up records | Arbitrary substitution, induced responses, mixed versions: locate impact and revise or supplement data, retain changes |
| Data Validation | Raw files, data dictionary, transfer logs, cleaning rules & anomaly checks, frozen version | Bulk deletion of speeders/duplicate IP: review evidence and check cleaning sensitivity; do not directly modify raw values |
| Analysis | Problem correspondence, assumption diagnostics, measurement quality, code, robustness | Method stacking, leakage, failed measurements still interpreted in path: simplify or suspend that conclusion |
| Writing | Result indexing, references, conclusion-recommendation table, authentic sources | Conclusion overreach, fabricated ROI, software screenshot stacking: backtrack results and rewrite |
| Submission/Defense | Official checklist, anonymity & version, file openability, PPT & backup, timed rehearsal | Deadline confusion, anonymous omissions, mismatched numbers between versions: check item by item and retain submission proof |
Authenticity & Process Quality
Verify whether the sample actually satisfies screening conditions, whether survey channels match the plan, whether interviews can be traced back, and whether photos genuinely belong to the project. Contact counts, collection counts, exclusions, effective n, and n for each model must be traceable; check for duplicates and coding issues when combining multiple channels. When questioning authenticity, document visible anomalies and evidence to be checked; do not directly treat same-device, rapid responses, or a certain pattern as fabrication.
Do not generate survey responses to fake collection, fabricate corporate adoption proof, cooperation authorizations, interview recordings, on-site photos, or advisor signatures. If critical evidence gaps are found, complete other parts first; mark affected conclusions as 'pending verification,' and if necessary, suggest supplementary data collection, removal of conclusions, or adjustment of scope, rather than covering up.
Ethics & AI
Only collect information necessary to complete the task; state that recruitment is voluntary, explain usage and withdrawal procedures, and verify institutional requirements for vulnerable populations and sensitive information based on the actual scenario. Check permissions and identification risks for public texts, screenshots, recordings, photos, and third-party scales. Visibility on the internet does not imply free collection and publication. For privacy or compliance matters requiring specific legal judgment, separately verify current official authorities; do not base legal conclusions on the manuscript.
Restrict access to raw sensitive data, de-identify public reports, and avoid re-identification through combinations of quotes, locations, and identities. Retain storage and deletion arrangements appropriate to the project; do not indefinitely collect device/location information just to prove the survey.
Distinguish between research integrity advice and official AI regulations for the current edition. AI can assist in drafting items, code, pre-coding text, and language polishing, but must be verified; record the tool/model, date, purpose, type of input data, human review, and impact on results, and disclose according to competition requirements. Do not upload unauthorized identifiable materials; do not disguise generated content as field evidence; students must understand and be able to independently explain core investigation and analysis.
Pre-Submission Executable Checklist
According to the current edition's checklist, confirm each required file (main text, appendix, PPT, declarations, etc.) one by one; do not assert which files are required a priori. Check file names, versions, page count scope, format, references, anonymity scope, metadata/comments/revision marks/headers/footers/identifiers in figures; if anonymity requirements are not yet available, list as 'pending confirmation'; do not proactively delete research sources. Open the final exported file to check tables, formulas, table of contents page numbers, and fonts; prepare PPT with an offline backup. Clearly distinguish between the official review draft and the archived version with identity.
Ensure consistent sample size, indicator caliber, and core conclusions across all materials. Check whether upload was successful and whether receipts are retained; the skill's review conclusion does not equal actual submission; actual submission is still performed based on user request and tool capabilities.
Review Deliverables
Output: Number | Phase/Location | Issue | Visible Evidence or Gap | Nature | Impact | Remediation Action | Verification Evidence. Nature categories: confirmed rule violation, research methodology issue, missing material, pending rule verification, expression improvement. Prioritize based on impact on authenticity, interpretability, formal acceptance, and remediation cost; do not give estimated score deductions without official basis.
Use actionable statements such as 'can proceed to next phase,' 'review after supplementing specific evidence,' or 'key conclusion currently unsupported.' List the scope reviewed and the scope not yet verifiable; do not declare full compliance or award guarantees if materials are incomplete. The re-review is based on the actual results after remediation.