/research-report-interpretation

name: research-report-interpretation
description: Interpret sell-side, institutional, expert, or investor research by extracting claims, assumptions, valuation drivers, verification needs, and thesis impact.

Research Report Interpretation Skill

这个 skill 用于解读券商、卖方、机构、专家或投资者研究报告。它的目标不是复述研报，而是把一份报告拆成可验证 claim、隐含假设、预期变量、估值驱动和对 Mira thesis 的增量影响。

研报在 Mira 中默认是 `sellside_and_expert_research`，通常是 `L3 secondary / signal`。它可以帮助识别框架、预期差、变量优先级、估值方法和叙事变化，但不能替代公司披露、监管文件、官方数据、市场数据或可复算模型。

Use When

• 用户提供或指定一份券商研报、机构研报、rating change、target price update、initiation note、industry note、专家研究、投资者信或研究 PDF。
• 用户问“这篇研报怎么看 / 靠谱吗 / 有什么新东西 / 对 thesis 有什么影响”。
• 需要从研报中提取可复用方法、变量框架或预期差线索，但当前任务仍围绕某个研究对象，而不是纯方法论研究。
• 需要把研报结论与已有 Mira thesis、company filing、earnings package、consensus proxy 或市场定价做差异检查。

如果用户问的是“这个研究方法本身是否值得纳入 Mira”，优先进入 `loops/methodology-research-loop.md`。如果研报解读触发 thesis 状态变化，再 handoff 到 `loops/thesis-update-loop.md` 或 `loops/event-delta-loop.md`。

Required Inputs

• `research_object`: ticker、company、industry、macro asset 或 methodology object。
• `market_scope`
• `time_boundary`
• `report_title`
• `provider_or_submitter`
• `author_or_team`，如可得
• `report_date` 或明确的 `source_gap`
• `report_type`: initiation、update、rating_change、target_price_update、industry_note、thematic_note、expert_research、investor_letter、other
• `access_route`: public_on_demand、user_material、authorized_provider 或 transient_only
• `license_scope`
• `storage_scope`
• `redistribution_allowed`
• `user_goal`: summarize、challenge、extract_claims、compare_to_thesis、update_thesis、reverse_engineer_method
• `completeness_status`: full_report、excerpt、screenshot、summary_only、metadata_only

Required Source Types

• Report source or user material intake record.
• `restricted_source_note` when the report is paid, confidential, licensed, user-provided, expert-network, or otherwise restricted.
• At least one independent `L1` / `L2` / `L5` source when a report claim is used for a durable conclusion.
• Existing Mira thesis package, evidence log or expectation map when the user asks for thesis impact.
• Consensus or estimate source, or explicit `source_gap`, when the interpretation depends on market expectation baseline.

Ingestion And Permission Gate

Before using a newly supplied report, run `data/ingestion-layer.md`.

Default treatment:

• Public report URL: `ingestion_route=public_on_demand`; cite metadata and short summaries only.
• User upload, screenshot, clipped PDF or exported note: `ingestion_route=user_material`; keep private unless user explicitly approves promotion.
• Licensed vendor or paid research: `ingestion_route=authorized_provider` or `user_material`; tracked artifacts may contain metadata, short compliant effect notes and claim categories, not raw report content.
• Unknown permission: `storage_scope=transient_only` or `private`; `public_case_use=blocked`.

Do not commit full paid reports, substantial excerpts, expert-network transcripts, or vendor raw data. Do not quote long passages. Extract claim-level summaries instead.

Analysis Flow

1. Report Metadata And Source Posture

Record:

• report identity, date, provider, author/team and source URL/path
• permission and storage boundary
• whether the report is complete or only a fragment
• whether it is company-specific, industry-wide, macro, thematic or method-focused
• stale boundary: report date, event covered, next earnings, next data release or target price/rating update

If `report_date`, permission, completeness or object identity is missing, keep the output at `working_view` and mark the gap.

2. Thesis Of The Report

Separate:

• headline conclusion
• core variable the author believes matters most
• time horizon of the view
• catalyst or revision path
• base case, bull case and bear case
• rating / target price / valuation output
• evidence the report itself cites
• what the report asks the market to change its mind about

Never treat rating, target price or author conviction as evidence by itself.

3. Claim Extraction

Extract claim-level records into `report-claim-map.csv`.

Each row should identify:

• `claim_type`: fact, reported_metric, guidance, target, forecast, assumption, interpretation, opinion, market_pricing, sentiment or derived_calculation
• `source_speaker`: sellside, buyside, expert, company, market, media or mira
• `variable`: revenue, margin, pricing, volume, capex, cash flow, multiple, risk premium, market share, demand, supply, regulation or other
• `time_horizon`: current, next_quarter, FY1, FY2, long_term or unknown
• `evidence_cited_by_report`
• `independent_verification_status`
• `treatment`: use_normally, attribute, monitor, needs_cross_check, source_gap, exclude

Facts and reported metrics require independent verification before they can support a durable Mira conclusion. Forecasts, assumptions, interpretations and opinions remain attributed to the report.

4. Expectation And Variant-Perception Bridge

Ask:

• Is the report describing consensus, challenging consensus, or updating consensus?
• Which variable is the expectation baseline: revenue, EPS, margin, FCF, capex, TAM, unit volume, ASP, multiple or risk premium?
• Does the report cite consensus provider/data, or is it using author estimates?
• Is the claimed edge from better facts, better interpretation, different time horizon, or different risk weighting?
• What would prove the author right or wrong?

If no reliable consensus proxy exists, write `source_gap`; do not replace consensus with a single analyst view, media tone or price action.

5. Valuation And Model Decomposition

If the report includes target price, rating, valuation multiple, DCF, SOTP or scenario math, decompose:

• forecast driver: revenue, margin, EPS, FCF, capex, share count, net cash/debt
• valuation anchor: P/E, EV/Sales, EV/EBITDA, EV/FCF, DCF, SOTP, NAV, replacement cost or other
• terminal assumptions and discount rate if available
• multiple change versus estimate change
• sensitiv