Sector Overview

Sector Overview

0. Data Sourcing & Provenance — NEVER FABRICATE

This section is the one to get right before any other. Market research lives or dies on credible numbers. A market-size, share, growth, or multiple figure with no traceable source is worse than no figure. A blank with a note beats an invented number.

Provenance has two mechanisms, and they divide the work. The clickable numbered source marker [n] on the figure, jumping to ## 来源 at the end, is the primary one: in flowing sector prose whose claims are all externally sourced, a [n] entry that names the firm, the date, and the basis already tells the reader the number is [披露] or [预期], so those two chips are not required in the narrative. But a citation cannot tell a reader that a number is ours or that a report is unconfirmed, so three tags are chipped inline even in prose:

• [测算] — anything we computed or assumed (a CAGR we derived, an implied share, a range midpoint).
• [推断] — our analytical read with no corresponding record ("this likely reads across to peers").
• [媒体] — reported by media and not corroborated by a disclosed record.

Chips ARE required throughout any table, KPI strip, or valuation summary, where disclosed actuals, our computations, and third-party forecasts sit side by side and the reader cannot otherwise tell them apart. The Chinese forms [披露] [测算] [预期] [推断] [媒体] in a Chinese deliverable, the English aliases [Reported] [Est.] [Consensus] [Inferred] [Media] in an English one; never mixed, and never a paraphrase such as [已披露] or [一致预期]. [预期] always names its provider inline.

Hard rules — violating any is a serious failure:

1. Cite firm + date + basis, via [n], for every market-size / share / CAGR / multiple. Not "market is $50bn" but "$50bn (Gartner, 2025; vendor-revenue basis)[n]". A TAM/CAGR with no named source is forbidden — delete the number. If a table cell must hold something, write n.d.; an unsourceable figure is not one we publish.
2. TAM hype vs. realistic addressable. Distinguish a vendor's promotional TAM from a defensible SAM/SOM; if you cite a big TAM, state the basis and whether it's realistic.
3. Conflicting third-party estimates → show the spread, don't cherry-pick. If Gartner says $50bn and IDC says $38bn, report both with sources and note the definitional difference — never silently pick the flattering one.
4. Don't pass off a derived number as a hard one. When you compute a CAGR, infer a share (revenue ÷ TAM), or take a range midpoint, that figure carries [测算], says so in the text ("we estimate", "implied", "midpoint"), and still cites the inputs' source [n]. Just don't dress a derivation up as a reported figure.
5. MCP first, then the finance vertical index, then web. Pull company financials/multiples from 同花顺/SEC MCPs; where a fact is not in a tool, 先走 finance-search.finance_search(金融垂搜,按档位召回并标注来源等级),它没有命中再落通用搜索。媒体档默认带日期窗口;官方档(weight>=3)反过来 —— 日期窗口会丢掉 publish_time 为空的记录(该档多数如此),取原始文件时不带窗口。 (web is last resort and must still be cited with publisher + date via [n]). Reconcile: segment/share pieces should tie to the stated total.

Workflow

Step 1: Define Scope

• Sector / subsector: What industry and how narrowly defined?
• Purpose: Client report, internal research, pitch material, idea generation
• Depth: High-level overview (5-10 pages) or deep dive (20-30 pages)
• Angle: Neutral landscape vs. thematic thesis (e.g., "AI infrastructure buildout")
• Universe: Public companies only, or include private?

Step 2: Market Overview

Market Size & Growth

• Total addressable market (TAM) with source — name the research firm, date, and basis, linked [n]
• Historical growth rate (5-year CAGR) — cite the source [n]; if you computed it, say "we estimate" and cite the underlying data
• Forecast growth rate and key assumptions — name the forecasting firm and date, linked [n]
• Market segmentation (by product, geography, end market, customer type)

Industry Structure

• Fragmented vs. consolidated — top 5 market share
• Value chain map — where does value accrue?
• Business model types (subscription, transaction, licensing, services)
• Barriers to entry (capital, regulatory, technical, network effects)

Key Trends & Drivers

• Secular tailwinds (3-5 major trends)
• Headwinds and risks
• Technology disruption vectors
• Regulatory developments
• M&A activity and consolidation trends

Step 3: Competitive Landscape

Company Profiles (for top 5-10 players):

This is a mixed-class table: chip every cell ([披露] for a filed figure, [测算] for a share you divided out, [预期] with its named provider for a forward number). A metric we could not source is n.d., never a filled-in guess. Numbers right-aligned, text left-aligned; document-table styling per the house formatting policy (no vertical rules, #1F3A5F header fill with white bold text, zebra white/#F4F7FA, thin #E5E9EF horizontal grid).

For each company, brief profile:

• Business description (2-3 sentences)
• Strategic positioning and moat
• Recent developments (earnings, M&A, product launches)
• Valuation snapshot (P/E, EV/EBITDA, EV/Revenue)

Competitive Dynamics

• How do companies compete? (price, product, service, distribution)
• Who is gaining/losing share and why?
• Disruption risk from new entrants or adjacent players

Step 4: Valuation Context

Where each number comes from — this step has a real source, use it rather than web-sourcing multiples.

• Sector multiples, current: hexin-index.sector_data gives 市盈率(TTM,整体法) / 总市值(合计) / 成份股个数 in one call. The subject must name the classification — 食品饮料板块(申万行业), not a bare 行业名 (which returns an empty table that reads as absence, not as a failed lookup). State the 板块名称 the call resolved to.
• Sector multiples, historical range and percentile: take them from the sector index, not from sector_data. Pull the index's daily PE(TTM) series with hexin-index.index_data (verified: a sector index returns dated daily PE rows), compute the range and percentile yourself, tag [测算], and state the window's start date and sampling frequency. sector_data is the current cross-section only — verified: given a two-month window it returned a header row with no data and echoed "交易日期":"最新" in indicators_params, i.e. it answers with the latest point and does not say it dropped your window. So always read indicators_params back: a single latest value dressed as a range is the failure mode here. Then compute the range/percentile yourself, [测算], with the window's start date stated. Do not use index_data's built-in 分位数 for a multi-year window — it defaults to 52 weeks and asking for 「近5年」 silently narrows it to two days and returns 100.0.
• Same 简称, different code across calls. One sector index resolved to its 中证 code in one call and a 深证 code in another, with a different PE on each. Record the code each figure came from and never splice two calls' series without checking they are the same index.
• Sector vs broader market: pull the broad index (沪深300 / 中证全指) the same way and show both multiples on the same date and 口径. A premium/discount quoted without saying which index and which date is not a number.
• Premium/discount drivers (growth, margins, market position) are [推断] unless a retrieved figure supports each one.
• Recent M&A transaction multiples have no source in this plugin — there is no precedent-transaction database here. Either cite the announcements you actually retrieved (wind-docs.get_company_announcements), or say the transaction multiples were not obtained. Do not present a remembered deal multiple.

Step 5: Investment Implications

• Where are the best risk/reward opportunities?
• What thematic bets can be expressed through this sector?
• Key debates in the sector (bull vs. bear arguments)
• Catalysts that could change the sector narrative

Step 6: Output

Deliver in the format the user requested; don't assume one. If unspecified, a sector primer is long-form, so build it as a PDF via report-render and say in one clause that you chose it. Include: 用户要 Word 时同样走 report-render（DocxReport，与 Report 同一套调用）——要出 Word 就先载入 report-render 技能，再动手建；手搓 python-docx / docx-js 会丢掉标签配色、[n] 跳转与封面页，机制与实测记录在那个技能里，不在这里重述。

• Market overview and sizing
• Competitive landscape map
• Company comparison table
• Valuation summary
• Key charts: market growth, share trends, valuation history
• A ## 来源 section (numbered, each entry a clickable link), starting on its own page
• A ## 覆盖范围与局限 block
• Excel appendix with detailed company data (optional)

## 来源 — that exact heading, nothing else. One entry per distinct [n], and the count of distinct markers must equal the count of entries:

[n] 〔一手|二手〕发布主体 · 文档或系统名 · 日期(发布日; 检索于) · URL

Primary is the issuing document itself — the research firm's own report, the filing, the exchange announcement, a database field sourced from those. Everything else is Secondary, and a Secondary entry must name what it relays: Secondary · Caixin · relaying IDC "China AI Market Tracker" · 2026-05-12 (published); retrieved 2026-07-25 · https://…. This is the common failure mode in sector work: a market-size or share number that resolves only to a news article quoting a report is a Secondary entry, and the body must say the underlying report was not obtained. Do not launder a news URL into an authoritative-looking citation.

## 覆盖范围与局限 — report it even when everything worked. State the retrieval timestamp, and for each of the three states — 有记录 / 检索范围内未发现 / 源不可用 — what it applies to:

## 覆盖范围与局限
Retrieved: [timestamp] · Universe: [the names and the sector definition used]

| 检查项 | 结论 | 源 | 检索于 |
|---|---|---|---|
| Financials for [names] | 有记录 | 同花顺 / SEC | [date] |
| Financials for [names] | 源不可用 | [system, why] | [date] |
| Consensus (revenue, EBITDA, EPS) | 有记录 for [n of N] names | [named provider] | [date] |
| Primary market-size report | 检索范围内未发现 — sized from a Secondary relay [n] | — | [date] |

本次未能覆盖: [sources that failed, and what they would have shown]
Data lag: [known lag for the sources used; sector overviews age fast]

"检索范围内未发现" is a statement about our search, not about the world — never render it as "there is no such data".

Format-agnostic quality bar — non-negotiable whatever the format (build it, then render/open it and look at every page/slide). report-render's references/pitfalls.md catalogues the silent failure modes behind these; read it rather than rediscovering them:

• Charts and their source notes never overlap. Caption "Figure X" above the chart; the Source: line below the plot. In matplotlib use a figure-level footnote (fig.text() + reserved bottom margin), not ax.text() at data coordinates (which collides with axis labels).
• State each chart's source exactly ONCE. If the source line is rendered inside the chart image (e.g. via fig.text()), do NOT also repeat it as a caption note below the figure — that duplicates the attribution. Pick one place (embedded-in-chart is fine) and don't echo it.
• Size charts by information density, not a fixed width — never let a chart dominate the page. A sparse chart (two or three bars) should render small; a dense or multi-panel chart can be wider. Don't hand-tune a width per chart, and don't apply one width to all. report-render implements this: it counts the drawn elements into a _chart_meta.json sidecar, maps density to a display width, and clamps by a height cap of roughly ⅓ of the text column. Use it rather than re-deriving the sizing.
• Bind each figure/table to its caption (e.g. reportlab KeepTogether([caption, image, note])) so a title never splits from its figure across a page/slide.
• Chinese / CJK must render, not blank, and wrap correctly — embed a real TTF (Regular + Bold) and set CJK word-wrap; built-in CID fonts render blank. report-render handles this, and references/pitfalls.md explains the several ways it fails silently if you build outside it.
• Links genuinely clickable where the format supports it: inline [n] jump to the Sources entry, entries open the source (PDF /Link annotations; HTML hyperlinks) — not plain blue text.
• Let content flow; don't force breaks. No one-section-per-page PageBreak() and no scattered CondPageBreak() stranding half-empty pages. The only legitimate hard breaks: after the cover, and before ## 来源 (own page). If a full-width chart spills and strands a gap, shrink its width (e.g. 174mm → ~138mm) so the bound caption+image fits the current page.
• No overflow, orphan, or half-empty pages/slides.

Important Notes

• Source ALL market-size/share/growth data — cite the research firm, date, and methodology/basis (see Section 0). No unsourced TAM or CAGR; an unsourceable figure is deleted, not published with a caveat.
• Distinguish between TAM hype and realistic addressable market; show the basis.
• Sector overviews age fast — stamp Retrieved: [date], note it in the coverage block, and flag data that may be stale.
• Charts are essential — market size waterfall, competitive positioning matrix, valuation scatter plot — each with its source.
• If for a client, tailor the "so what" to their specific situation (M&A target identification, competitive positioning, market entry).