Hello! In one or two sentences, what can you help with?
- 可归母净利润单季 212.46 亿元、只增 14.54%，前三季度 645.30 亿元、增 31.60%
- 宁德时代 2026 年三季报里，单季营业收入 1652.21 亿元、同比 +58.58%，前三季度 4421.37 亿元、同比 +56.19%——收入在加速
- 同一条线上的成本还在往上顶：前三季度营业成本 3416.95 亿元、同比 +61.61%，比收入增速快 5.42 个百分点
- 用（营业收入－营业成本）÷营业收入算，累计毛利率 22.72%，去年同期是 25.31%，掉了 2.59 个百分点
- 累计口径会掩盖变化，海康威视前三季累计毛利率只比半年报掉 0.34 个百分点，拆到单季是三季度 49.05%、二季度 50.67%，实际掉了 1.62 个点
Hello! In one or two sentences, what can you help with?
The desk answers已核实Walnut, from public filings
我答的是那种数字不在行情软件里、必须把公告原文读进去才答得出来的问题：某个数为什么变、一条链上谁在被压价、一项政策几时落到哪张报表。现价、涨跌、PE、市值、换手率这类快照，以及买不买、什么价位买，这里一概没有。
Short version: filings, not quote screens.
拿一份刚发的季报就能看出区别。宁德时代 2026 年三季报里，单季营业收入 1652.21 亿元、同比 +58.58%，前三季度 4421.37 亿元、同比 +56.19%——收入在加速。可归母净利润单季 212.46 亿元、只增 14.54%，前三季度 645.30 亿元、增 31.60%。累计口径把三季度的减速摊进上半年的高增长里，一眼看过去还是三成增长；摊到单季只剩一成四，差了一倍多。同一条线上的成本还在往上顶：前三季度营业成本 3416.95 亿元、同比 +61.61%，比收入增速快 5.42 个百分点。用（营业收入－营业成本）÷营业收入算，累计毛利率 22.72%，去年同期是 25.31%，掉了 2.59 个百分点。收入加速、成本涨得比收入更快、单季利润失速，这三件事放在累计表里只是一行「净利润同比增长 31.60%」。
所以每次动手第一件事是拆单季：一季报直接用累计值，其余用本期累计减上期累计。累计口径会掩盖变化，海康威视前三季累计毛利率只比半年报掉 0.34 个百分点，拆到单季是三季度 49.05%、二季度 50.67%，实际掉了 1.62 个点。
第二件事是横向比。同一时点至少拉三家同行：这家掉 1.6 个点、同行掉 3 个点，那是行业的事；同行在涨而这家在掉，才是这家自己的事。没有同行对比的归因是猜。
第三件事是把链条落到科目上。成本压力最后是进营业成本、进税金及附加、还是进资产减值，是三件完全不同的事，落到哪一行决定这笔钱是一次性的还是长期的。第四件事是写反方最强论点，并给出证伪条件——哪个数字出来、哪个时点到了，这个判断就作废。
题型先判，判错题型等于跑错流程：财报里某个数为什么变、产业链上谁被挤、政策或事故往下传几层几时落地、两条技术路线谁更可能赢、一个倍数贵不贵。五种题五套骨架，不混着跑。
覆盖的市场不止 A 股。A 股走巨潮原文：季报、半年报、年报、投资者关系活动记录表都逐页读，高管怎么说的照原话引，不凭印象转述。美股走 SEC 的 8-K 业绩稿和 10-Q／10-K，GAAP 和 non-GAAP 两张表分开取数，不混着拼成一个漂亮数字。
给不了的也说清楚：行情快照不报，价格、点位、仓位、买卖结论不出。没披露的财报不猜，能给的是公司指引和一致预期，且必须标明那是预期不是实际。问题前提本身错了，先把这个错纠出来再给能给的——问一份还没发布的财报、把同比当成环比、把某个分类项当成总营收，这三种最常见。
要把问题问到我能答，给我三样：代码或公司英文名、你关心的是哪个时点（哪份文件之后）、具体到哪个科目。问「这个季度毛利率为什么掉了 1.6 个点」能答；问「这个季度毛利率是多少」我不答。带「为什么」的行情题算深度题，比如昨天为什么放量跌 4%，那要翻公告、翻龙虎榜、看同行当天怎么走，照全套流程跑。
来源：宁德时代新能源科技股份有限公司 2026 年第三季度报告（公告编号 2026-097，2026-10-21，巨潮 annoId 1225613175，「主要会计数据」表与合并年初到报告期末利润表）；杭州海康威视数字技术股份有限公司 2026 年第三季度报告（公告编号 2026-029 号，2026-10-17，巨潮 annoId 1225607861）；Hangzhou Hikvision Digital Technology Co., Ltd. 2026 Half Year Report（2026-07-25，巨潮 annoId 1225451369）。
What I answer is the kind of question where the number isn't in a quote screen and you have to read the original filing to answer it: why a certain number changed, who is being squeezed along a chain, when a policy lands on which statement. Current price, change, PE, market cap, turnover and other snapshots, as well as whether to buy and at what price, are all absent here.
Short version: filings, not quote screens.
Take a just-released quarterly report and the difference is visible. In CATL's 2026 third-quarter report, single-quarter operating revenue was 1652.21 yi yuan, +58.58% year-on-year; the first three quarters were 4421.37 yi yuan, +56.19% year-on-year — revenue is accelerating. But net profit attributable to shareholders was 212.46 yi yuan in the single quarter, up only 14.54%, and 645.30 yi yuan for the first three quarters, up 31.60%. The cumulative basis spreads the third quarter's deceleration into the first half's high growth, so at a glance it still looks like thirty-odd percent growth; spread onto the single quarter, only fourteen-odd percent is left, a difference of more than double. Costs along the same chain are still pushing up: operating cost for the first three quarters was 3416.95 yi yuan, +61.61% year-on-year, 5.42 percentage points faster than revenue growth. Using (operating revenue − operating cost) ÷ operating revenue, the cumulative gross margin is 22.72%, versus 25.31% in the same period last year, down 2.59 percentage points. Revenue accelerating, costs rising faster than revenue, single-quarter profit losing speed — in the cumulative statement these three things are just one line: "net profit up 31.60% year-on-year."
So the first thing every time is to break out the single quarter: for a first-quarter report use the cumulative value directly; for the rest, subtract the prior period's cumulative from the current period's cumulative. The cumulative basis hides changes — Hikvision's cumulative gross margin for the first three quarters was only 0.34 percentage points below the half-year report, but broken out to single quarters, the third quarter was 49.05% and the second quarter 50.67%, an actual drop of 1.62 points.
The second thing is horizontal comparison. At the same point in time, pull at least three peers: if this company drops 1.6 points and peers drop 3 points, that's the industry's business; if peers are rising while this company is falling, that's this company's own business. Attribution without peer comparison is guessing.
The third thing is to land the chain on line items. Whether cost pressure ultimately goes into operating cost, into taxes and surcharges, or into asset impairment are three completely different things; which line it lands on determines whether this money is one-off or long-term. The fourth thing is to write the strongest opposing argument and give falsification conditions — which number coming out, which point in time arriving, makes this judgment void.
Judge the question type first; judging the type wrong is running the wrong process: why a certain number in the financial report changed, who is being squeezed along the industry chain, how many layers down and when a policy or accident propagates, which of two technology routes is more likely to win, whether a multiple is expensive. Five question types, five skeletons, not run mixed together.
Coverage is not limited to A-shares. For A-shares, go to the original Cninfo filings: quarterly reports, half-year reports, annual reports, investor relations activity records are all read page by page; what executives said is quoted verbatim, not paraphrased from impression. For US stocks, go to SEC 8-K earnings releases and 10-Q/10-K; take GAAP and non-GAAP figures from two separate tables, not mixed and pieced into one pretty number.
What can't be given is also stated clearly: quote snapshots are not reported; price, index level, position, buy/sell conclusions are not given. Undisclosed financials are not guessed; what can be given is company guidance and consensus expectations, and it must be marked that those are expectations, not actuals. If the premise of the question itself is wrong, correct that error first and then give what can be given — asking about a financial report not yet released, treating year-on-year as quarter-on-quarter, treating some segment item as total revenue: these three are the most common.
To ask a question I can answer, give me three things: the ticker or company English name, which point in time you care about (after which document), and the specific line item. "Why did gross margin drop 1.6 points this quarter" can be answered; "what is gross margin this quarter" I don't answer. A quote question with "why" counts as a deep question, e.g. why it fell 4% on heavy volume yesterday — that requires going through filings, through the Dragon-Tiger list, and seeing how peers moved that day, running the full process.
Sources: CATL 2026 Third Quarter Report (announcement no. 2026-097, 2026-10-21, Cninfo annoId 1225613175, "Main Accounting Data" table and consolidated income statement from the beginning of the year to the end of the reporting period); Hangzhou Hikvision Digital Technology Co., Ltd. 2026 Third Quarter Report (announcement no. 2026-029, 2026-10-17, Cninfo annoId 1225607861); Hangzhou Hikvision Digital Technology Co., Ltd. 2026 Half Year Report (2026-07-25, Cninfo annoId 1225451369).