CN IB Regulatory Case Research
Build a defensible precedent set. Do not return a list of documents that merely share keywords.
Principles
- Search primary regulatory and issuer filings before media, databases, or summaries.
- Separate question similarity, fact-pattern similarity, business similarity, and procedural-stage similarity.
- Preserve the exact question, response, intermediary procedure, conclusion, source URL/file, and page.
- Do not treat a precedent as binding law or assume a past outcome predicts the current review result.
- Distinguish what the issuer stated from what the sponsor, counsel, or accountant verified.
- Report negative and imperfect precedents when they materially inform the risk.
Workflow
1. Decompose the research question
Write an issue tree containing:
- explicit regulatory questions;
- implicit review concerns;
- relevant facts and transaction structure;
- responsible intermediary;
- target board and review stage;
- expected evidence and possible follow-up questions.
Do not search the full user sentence as the only query.
2. Build a search vocabulary
Read references/search-strategy.md. Create combinations of:
- regulatory terminology;
- business and product terminology;
- accounting/legal synonyms;
- issuer characteristics;
- board and filing stage;
- known comparable issuers.
Record every material query so the search is reproducible.
3. Search in source order
Prioritize:
- exchange and CSRC filing pages;
- issuer-filed prospectuses, inquiry responses, sponsor reports, and legal/accounting opinions;
- official rules, guidance, and disciplinary decisions;
- licensed databases or structured filing repositories;
- professional summaries only as discovery aids.
Use a secondary source to find a case, then return to the primary filing for evidence.
If cn-ib-source-verifier is available, use its source-connector registry to plan the access path. In particular, use 见微数据 as a case-discovery and prospectus-comparison aid, not as a substitute for the underlying exchange or CSRC document.
4. Create the candidate matrix
Use assets/case-matrix-template.csv. One row must identify:
textcase_id, issuer, board, regulator, filing_date, filing_type, question_topic, exact_question, response_summary, intermediary_procedures, outcome, source_url, page, direct_quote, issue_similarity, fact_pattern_similarity, business_similarity, regulatory_stage_similarity, evidence_quality, recency
Do not score a case until the primary filing and page have been confirmed.
5. Rank candidates
Read references/relevance-rubric.md, score each dimension from 0 to 5, then run:
bashpython3 scripts/rank_cases.py cases.csv --output ranked-cases.csv
The default weighting favors issue and fact-pattern similarity over industry similarity. Review the ranking manually; a score does not replace judgment.
6. Extract the regulatory pattern
For the strongest cases, separate:
- regulator's explicit concern;
- issuer's factual explanation;
- quantitative support;
- sponsor procedures;
- counsel procedures;
- accountant procedures;
- unresolved limitations;
- follow-up questions and final status if available.
Do not merge statements from different parties into one unattributed answer.
7. Apply precedents to the current task
Explain:
- which precedents are directly comparable;
- which are analogous only;
- what evidence the current issuer would need;
- which response structures recur;
- what differences may prevent direct reliance;
- likely follow-up questions.
Use cautious language: 可供参考, 具有一定可比性, 事实基础存在差异, or 不能直接类推.
Output contract
Deliver:
- one-paragraph answer to the research question;
- ranked precedent table;
- source-backed extracts with page references;
- recurring regulatory concerns and response patterns;
- recommended evidence and intermediary procedures;
- gaps, non-comparable facts, and search limitations.
Do not download or retain unnecessary filing copies in the user's deal directory. Use temporary storage and retain only requested final artifacts.
