A-Share Pre-Market Brief | 2026-09-01 Tuesday

Scan window: 2026-08-31 (Monday) 15:00 close → 2026-09-01 07:00. Every item is tagged with its publication time; anything whose publication time cannot be confirmed is kept out of the main table.

Scope boundaries of this issue (judge the strength of the conclusions accordingly): ① Execution-price basis: this issue is written at 07:00, and the earliest price executable today is the opening price produced by the 09:25 call auction, not the 8/31 close. Measured yesterday: the 5 recommended names returned +0.40% on the 8/28 close basis but +2.74% executed at that day's opening price — a gap of 2.34pct. Unless otherwise stated, all price moves in this issue are on the 8/31 close basis; readers placing orders should convert to the opening price themselves. ② Overnight US equities are on a closing basis (readings after the 04:00 Beijing-time close on 9/1; the cited source was published 9/1 06:05). This issue also lists one intraday-basis source published at 23:58 — the two sets of numbers differ and both are correct, see the note in §1 — do not treat an intraday snapshot as a closing conclusion. ③ Sector price moves and fund flows are on the THS industry-sector 8/31 closing-snapshot basis, which differs from Eastmoney's "main-force net inflow" basis; the two cannot be subtracted across issues. ④ Limit-up / consecutive-limit-up data are on the Eastmoney official limit-up pool 8/31 basis**, and the number of entries in the pool has been verified to match the official count (both 88). ⑤ Dragon-Tiger List data are the formal exchange disclosures released after the 8/31 close (after 18:00), and have been added in this issue (yesterday's post-close recap left them blank because it was written at 15:30). ⑥ No northbound flow data: since 2024-08-19 daily net buying via the Shanghai/Shenzhen-Hong Kong Stock Connect is no longer disclosed. This issue gives no northbound figures and cites no interface value that returns 0.0. ⑦ Interim-report data are on the Eastmoney interim-report financial database, 2026-06-30 reporting-period basis (11,445 records market-wide), cross-checked name by name against the earnings-preannouncement database for first-disclosure dates. ⑧ Every ROE in this issue is a 2026 half-year figure, not annualised (e.g. 中际旭创 (Innolight, 300308) 37.62%, 恺英网络 (Kaiying Network, 002517) 13.74%, 天孚通信 (T&S Communications, 300394) 19.60% are all half-year figures), do not compare them directly with annualised ROE; every "year-on-year" is against the same period of 2025, and names where a low or high base distorts the percentage are flagged individually in their own entries.

⚠️ Data-pull and search log (this section is not sent to clients):

1. Every data pull went through this time; no interface failures

• Dragon-Tiger List ak.stock_lhb_detail_em('20260831','20260831') returned 84 records normally — the TypeError reported in yesterday's post-close issue is confirmed to be a scheduling problem ("15:30 is earlier than the 18:00 disclosure point"), not an interface failure; today's 07:00 pull succeeded on the first attempt. This should be hard-coded: the Dragon-Tiger List can only be back-filled in the next day's pre-market issue, it cannot be demanded in the same day's post-close issue.
• Limit-up pool push2ex getTopicZTPool (dpt=wz.ztzt, sort=fbt:asc) returned tc=88 / len=88, validation passed. Keep using sort=fbt:asc, do not use sort=fund:desc (it silently returns an empty pool).
• Quotes came from Tencent qt.gtimg.cn; the measured timestamp was of the form 20260831161406 (around 16:14), confirming post-close settled data, not an intraday snapshot.
• Interim reports ak.stock_yjbb_em('20260630') normal, 11,445 records. Earnings preannouncements ak.stock_yjyg_em('20260630') normal.
• Eastmoney clist was not used (502 for several consecutive trading days); this issue does not need it — sector data reuse the THS snapshot already validated in yesterday's post-close issue.

2. ⚠️ Two search contaminations intercepted this time (both of the "headline hides the publication date" family)

1. 华策影视 (Huace Film & TV)'s AI micro-drama plan: searching "华策 微短剧 AI 布局" returned a Cailianpress item, "Huace Group's Beijing short-drama team: two AI-created micro-dramas expected in September–October", highly consistent with today's main theme, and neither the headline nor the URL contains a year. After actually fetching the page it was confirmed that the publication time is 2024-06-25 11:02:11 — two-year-old news. Copying it would have produced a fabricated same-day catalyst such as "Huace announced today that it will launch AI micro-dramas in September–October". Discarded; §2.1 attributes no AI-related catalyst to Huace.
2. Two sets of overnight US numbers: one Sina Finance piece (published 8/31 23:58) gives "S&P −0.23%, Dow −0.39%, Nasdaq −0.18%", while a NetEase piece (published 9/1 06:05) gives "Dow −0.7% at 53,185.9, Nasdaq −0.12% at 26,370.89, S&P −0.33% at 7,686.14". At 23:58 Beijing time the US market had not yet closed (04:00 close), so the former is an intraday basis. This issue adopts the 9/1 06:05 closing basis and notes the source of the difference in §1. A further search summary gives "S&P 7,691.74, Nasdaq 26,402.42", which matches neither original source and is untraceable — discarded.

3. One cross-check that was actively performed on methodology and that changed the conclusion

• The assumption "8/31 is the interim-report deadline → there is a batch of new high-growth interim reports to mine today" was checked and rejected by myself. Screening for "disclosed 8/31 + net profit > RMB 100 million + revenue > RMB 500 million + net profit growth > 50% + revenue growth > 10%" gave 14 names; going back through the preannouncement table one by one showed that 10 of them had already issued earnings preannouncements between 7/3 and 7/20 (招商轮船 7/6, 天风证券 7/9, 中国船舶 7/14, 华正新材 7/14, 华创云信 7/15, 广西能源 7/15, 永杉锂业 7/15, 赤峰黄金 7/15, 富创精密 7/20, 广钢气体 7/3), so the interim report is not incremental information for them; of the remaining 4, 3 (华峰测控, 沐曦股份, 强一股份) were already covered in yesterday's pre-market issue. Conclusion: the interim-report-deadline line offers no tradable increment today, so this issue does not open a branch for it, keeping only one market-wide summary line in §1. This is exactly "trace it back to the first-disclosure date before talking about an expectation gap".

4. Sub-agents and self-checks (what changed in this issue)

• Ran two fundamentals-analyst agents (芒果超媒, and the three tanker names) and one risk-auditor. Both fundamentals agents changed the conclusion — this was not a formality:
  • 芒果超媒 (Mango Excellent Media): the first draft gave it "recommendation slot 2, 76 pts"; after verification it was cut to "watch only, 53 pts". The key point is that the first draft only read "net profit attributable −73.58%", whereas the primary interim report shows something more serious: operating profit of only RMB 30.2 million, an ex-non-recurring loss in Q2, and RMB 168 million of the RMB 202 million net profit attributable coming from the recognition of deferred tax assets, plus net selling of RMB 134 million by foreign capital + institutions on the Dragon-Tiger List, absorbed by hot money.
  • Tankers: the first draft wrote it up as "oil price up → tanker shipping benefits" and gave it a B grade with no recommendation slot. Verification proved the causal chain was backwards (the oil price is a deduction in TCE); the real driver is Hormuz. The branch was upgraded to A− and given a recommendation slot, but the name was switched from 招商轮船 to 中远海能.
• ⚠️ One agent conclusion was not adopted in this issue, and is recorded here: the agent argued that "芒果超媒's 6.41% turnover looks low, but free float is only 54.6% of total market cap, so it does not count as well-locked". On re-check this does not hold — turnover value RMB 1.087 billion ÷ free-float market cap RMB 17.348 billion = 6.26%, which matches the 6.41% the interface gives (on total market cap it would be 3.42%, which does not match), showing that this turnover rate is already on a free-float basis and should not be discounted again. The reason for non-adoption is written out in §5.11. Sub-agent conclusions must also be run past primary data.
• An arithmetic error caught by self-check (scoring table in the draft §4): the component scores did not add up to the totals (昆仑万维 components summed to 75 while the total was written as 79, 芒果超媒 components summed to 44 while the total was written as 66, 中文在线 components summed to 31 while the total was written as 46). Each has been recomputed so that the total strictly equals the sum of components, and the §3 master ranking has been re-scored and re-sorted accordingly. At the same time a flaw was found in the scoring model — "a collapsing fundamental picture is double-counted in both 'fundamentals' and 'risk deduction'" — which is explained in §4, where it is now deducted only once.

5. What this issue did not do (listed honestly)

• Did not obtain 9/1 call-auction or pre-market order data (A-shares have no continuous pre-market auction, and this data does not exist at 07:00), so all the auction signals in §8 are written as criteria to be verified, not as conclusions.
• Did not use concept-sector rankings (stock_board_concept_summary_ths upstream fields changed, as recorded in yesterday's post-close issue); the main-theme judgement is instead computed in-house from "the industry distribution of the 88 limit-up names + first-seal times + seal-order/turnover ratios".

---

0. Today in One Sentence

Today's hardest catalysts are the three pieces of post-close information that landed together between 20:00 and 21:30 last night — all of them occurred after the close, so yesterday's prices do not contain them. The first is 20:12: 中际旭创 (Innolight, 300308) announced a planned buyback of RMB 4 billion–8 billion with a price cap of RMB 1,200/share (Sina Finance 8/31 20:12¹); the second is 21:06: Hunan TV's official Weibo said that the premiere of 《后西游记》 (The Later Journey to the West), the country's first AI-generated long-form drama aired on a satellite channel, "ranked first among provincial satellite channels in real-time ratings in its time slot" (Sina 8/31 21:06²), followed at 21:20 by MIIT's release of a special action plan to cultivate artificial-intelligence application service providers (China News Service 8/31 21:20³).

The third post-close thread is one that almost no Chinese financial write-up gets right in terms of the causal chain: Hormuz continued to escalate overnight — the US military struck Iran's Larak Island on 8/30 (described as preventing it from firing mine-laying rockets into the strait), Iran counter-attacked with missiles, and trackable vessels transiting the strait fell to 5 over the weekend; Brent November contract settled +2.71% at USD 90.49, and that settlement happened at 02:30 Beijing time on 9/1 — when A-shares closed at 15:00 on 8/31 that move had not yet played out. ⚠️ But the chain "oil price up → tanker shipping benefits" is wrong: in the TCE formula the oil price is a cost that gets deducted. The real evidence is the route divergence as of 8/28 (the latest available reading) — TD3C, which must pass the strait, was +54 points week-on-week, while TD15, which does not, was −17 points and TD22 was −USD 1 million. See §2.5, the section this issue put the most work into.

The strongest branch is "AIGC film & TV / micro-drama", but it is a branch with an extremely strong catalyst and extremely poor fundamentals, and that split has to be stated up front. Three catalysts stack up: the ratings validation of the AI long-form drama (post-close increment), the 《微短剧发展管理办法》 (Measures for the Administration of Micro-Drama Development) formally taking effect today, 1 September (published 7/31, effective today), and MIIT's AI application special action (post-close increment). Meanwhile the sector was already the second-best performer market-wide yesterday (film & cinema +6.50%, culture & media +4.66%), and 12 of the 13 limit-ups were first-boards. At the same time, this issue checked the interim reports of 22 media names one by one: 芒果超媒 (300413) net profit attributable RMB 202 million, −73.58% YoY, Q2 −98.84% QoQ; 上海电影 (601595) −71.62%; 欢瑞世纪 (000892) a loss of RMB 144 million; 光线传媒 (300251) is not quoted on a YoY basis in this issue because the prior year contained 《哪吒 2 》 (Ne Zha 2) and is an abnormally high base; 儒意电影, 横店影视, 博纳影业, 中文在线, 掌阅科技, 电广传媒 and 中国电影 are all lossmaking. 12 of the 22 posted interim losses, and only 4 were both profitable and growing net profit attributable, of which only three had revenue above RMB 500 million — 华策影视 (300133, net profit RMB 167 million, +41.98%), 捷成股份 (300182) and 天娱数科 (002354) — and not one of those three hit limit-up yesterday. This is a textbook "concept ≠ beneficiary" scene.

On the direction of money, the judgement is: the divergence is happening inside hardware, not between "hardware vs applications" — this has to be stated precisely. The top three net inflows yesterday were all hardware: semiconductors +RMB 8.376 billion (1st), banks +RMB 6.729 billion (2nd), consumer electronics +RMB 5.052 billion (3rd); whereas the communications-equipment sector rose 1.56% yet saw net outflows of RMB 3.426 billion (the second-largest outflow market-wide), with the gain held up by a few limit-ups such as 星网锐捷 (Star-net Ruijie, 002396) and 移远通信 (Quectel, 603236) while the sector as a whole was being sold. So the correct statement is: money flowed into semiconductors and consumer electronics and out of communications equipment; it cannot be phrased as "money rotated from hardware into applications". ⚠️ The THS industry classification has no standalone sector corresponding to "AI applications", so this issue did not obtain a sector-level net-inflow figure for that direction and draws no qualitative conclusion about "money on the application side". And among the top 5 limit-up names by turnover value yesterday, 4 did not seal until after 13:00 (浪潮信息 (Inspur, 000977) 13:51, 大族激光 (Han's Laser, 002008) 14:14, 星网锐捷 (002396) 13:46, 中文在线 (Chinese Online, 300364) 13:00), which is late-session chasing, not position-locking.

Driver type: policy (MIIT AI applications, micro-drama measures taking effect, battery consumption tax taking effect) + industry event (AI long-form drama ratings) + company announcement (Innolight buyback) running in parallel on three lines; earnings-driven catalysts are essentially absent today.

Pre-market read: leaning bullish but not a place to chase, and there are two clear headwinds today. Headwind one is offshore — all three major US indices closed lower overnight, the Nasdaq Golden Dragon China Index fell 2.19%, the 10-year Treasury yield was 4.75% (an intraday reading at 8/31 23:58, the highest since January 2025), and COMEX gold fell 0.72%; headwind two is that the battery consumption tax is formally levied from today (2% on lithium-ion storage batteries), and the lithium-battery chain already reacted in advance yesterday (CATL −1.34%, EVE Energy −2.87%, Farasis Energy −8.07%). The index closed yesterday at its high of the day, only 0.34% away from 4,000, but the failed-seal rate deteriorated from 16.33% to 25.42% and first-boards made up 79.5% — this is the "good index, poor relay" combination, and the correct posture today is to pick stocks, not the index.

---

1. News Overview

Impact level: S = changes an industry trend or creates direct order elasticity; A = clearly positive for one branch with multiple beneficiaries; B = positive but with a long chain that needs verification; C = sentiment stimulus with weak persistence. ⚠️ Items tagged "old news · effective today" have an announcement date earlier than this scan window; today is merely the effective date. Per iron rule 4, these must not be treated as brand-new positives, but the effective date itself may trigger a second round of pricing.

⚠️ On the basis conflict between items 6 and 9 (must read): item 6 comes from a closing report published at 9/1 06:05, item 9 from a piece published at 8/31 23:58. At 23:58 Beijing time the US market had not yet closed (04:00 close), so item 9's "S&P −0.23%, Dow −0.39%, Nasdaq −0.18%" is an intraday reading; differing from item 6's closing numbers is normal, not a data conflict. Everywhere this issue refers to US index levels it uses item 6's closing basis; item 9 is used only for the Treasury-yield and China-ADR information unique to it.

⚠️ On the nature of items 4, 5 and 18: all three merely take effect today, with announcement dates of 7/31, 7/16 and 5/29 respectively, all earlier than this scan window. Per iron rule 4 they are tagged "old news · effective today". Item 5, the battery consumption tax, is especially notable: the announcement has been out for six and a half weeks and the lithium chain was already falling yesterday (CATL −1.34%, EVE Energy −2.87%, Gotion High-tech −1.95%, Farasis Energy −8.07%), showing the market has long known. If there is a sharp sell-off today "because the consumption tax starts", that is more likely sentiment than increment.

---

2. Strongest Positive Branches, Descending

2.1 Branch One: AIGC Film & TV / Micro-Drama — Strongest Catalyst, Worst Fundamentals

Why it is positive and where it is hard:

Three catalysts are tightly clustered in time, two of which occurred after the 8/31 close, so yesterday's prices do not contain them:

1. The premiere of the AI long-form drama 《后西游记》 (8/31 18:00 on Mango TV / 20:00 prime time on Hunan TV). This is the first AI long-form drama in Chinese television history to air in a satellite channel's prime slot, with all 30 episodes AIGC-generated and no human actors. Hunan TV's official Weibo said at 21:06 that it "ranked first among provincial satellite channels in real-time ratings in its time slot". ⚠️ After verification this issue confirms: the official wording gives only the ranking "first among provincial satellite channels in its time slot" and publishes no specific ratings figure. The claim circulating in search summaries — "second among provincial satellite channels as of 22:00, market share above 1.5%" — could not be confirmed on the original page and is not cited in this issue.
2. 《微短剧发展管理办法》 formally takes effect today, 1 September. This is the country's first regulation dedicated to micro-dramas, and its core effect is to turn the industry from a traffic business into a licence business: producers of class I / class II micro-dramas must hold the 《广播电视节目制作经营许可证》, have registered capital above RMB 1 million, and have no foreign investment; class I requires NRTA approval of the distribution licence, class II is delegated to provincial bureaus, and class III is platform self-review with per-title coding. For listed companies that already hold licences this raises the entry barrier; for unlicensed small studios it is a clear-out.
3. MIIT's special action to cultivate AI application service providers (8/31 21:20), see branch two.

⚠️ But the fundamentals of this branch have to be laid out name by name, because they do not match the strength of the catalyst at all. This issue pulled the 2026 interim reports of 22 media / film & TV / publishing names one by one:

This is the most important table in this issue. 12 of the 22 posted interim losses; only 4 were both profitable and growing net profit attributable (华策影视 300133, 捷成股份 300182, 天娱数科 002354, 华智数媒 300426), and only the first three had revenue above RMB 500 million (华智数媒's +160.73% comes on revenue of only RMB 54 million, too small to be representative). The 13 media names that hit limit-up yesterday have each had their interim report checked, and break down as follows: ① 8 with net losses: 吉视传媒 (Jishi Media, 601929) −RMB 220 million, 博纳影业 (001330) −RMB 169 million, 欢瑞世纪 (000892) −RMB 144 million, 掌阅科技 (603533) −RMB 50 million, 中文在线 (300364) −RMB 43 million, 大晟文化 (Dasheng Culture, 600892) −RMB 39 million, 荣信文化 (301231) −RMB 16 million, 中广天择 (603721) −RMB 13 million; ② 3 profitable but sharply down: 芒果超媒 (300413) −73.58%, 上海电影 (601595) −71.62%, 龙版传媒 (605577) −34.46%; ③ 1 close to break-even: 时代出版 (600551) −3.37%; ④ 1 profitable with growing net profit attributable: 流金科技 (Liujin Technology, 920021, BSE ±30%) +464.61% — but net profit is only RMB 7 million, ROE 1.01%, and revenue is still −13.64%: an accounting jump off a low base, not an operating improvement.

In other words: not one of the 13 limit-up names rose on earnings. Conclusion: this branch can be traded today, but the reason for participating must be "event + money", not "earnings delivery". Any framing of it as fundamentals-driven is wrong.

Whether it is already fully expected / one-day-wonder risk:

• Day 1 of the theme is complete; today is day 2. The first-seal times of yesterday's 13 limit-up media names ran all the way from 09:34 to 13:22, a complete full-day diffusion path with ample spread.
• But 12 of the 13 were first-boards, and only 时代出版 (600551) was on its second board. The ladder is extremely flat. Per the verification point set in yesterday's post-close issue: "no second consecutive board tomorrow = one-day sentiment; only a second board establishes it as a main theme" — this criterion can be verified on the spot today.
• Limit-up quality varies enormously (seal order / turnover ratio; the larger the number, the harder the lock): 芒果超媒 0.357 (first sealed 09:35, zero failed seals), 欢瑞世纪 0.704, 中广天择 0.620, 上海电影 0.287, 龙版传媒 0.246, 博纳影业 0.243, 荣信文化 0.144, 掌阅科技 0.139, 中文在线 0.101 (not sealed until 13:00), 流金科技 0.034. 芒果超媒 is the only one that simultaneously satisfies "early-session seal + zero failed seals + low turnover (6.41%)".

Board distribution: ChiNext ±20% accounts for 5 (芒果超媒, 中文在线, 荣信文化, 华策影视, 华智数媒), SSE/SZSE main boards ±10% for 7, and BSE ±30% for 1 (流金科技 920021). Note that those ChiNext names already ran 20cm yesterday, and today's drawdown room is likewise 20cm.

2.2 Branch Two: AI Applications / Agents — Demand-Side Policy, Harder Than a Concept

The key part of the MIIT document is not the headcount target of "cultivating 2,000 service providers" but this sentence in the original text:

"Explore models such as first-purchase and first-use, and risk compensation, increase the intensity of procurement of large models, agents, Tokens and other services, and improve the quality and efficiency of model and intelligence use across industries."

This is demand-side government-procurement language, not supply-side capacity encouragement. "First-purchase and first-use" means fiscal funds backstop the first purchase, and "risk compensation" means backstopping a failed purchase — combined, these two phrases target exactly the biggest obstacle to AI deployment: the buyer does not dare go first. The document also encourages service providers to build "forward-deployed engineer (FDE) teams rooted at customer sites", pointing to delivery-type IT service providers rather than pure model companies.

⚠️ But three limits must be stated honestly:

1. The document publishes no procurement budget figure at all; "2,000 / 3,000" are service-provider headcount targets and cannot be converted into revenue elasticity.
2. The beneficiary set is extremely diffuse — any company doing AI integration and delivery can call itself a service provider, making it hard to pin down a single name.
3. The market already began pricing this direction intraday yesterday (万兴科技 300624 +9.91%, 昆仑万维 300418 +8.89%, 中科创达 300496 +5.25%), yet the document was only released at 21:20, which means yesterday's gains came from something else (the AI long-form drama + sentiment rotation) and the document is today's increment.

2.3 Branch Three: Optical Modules — A Line at Odds With Its Fund Flows

中际旭创 (300308)'s buyback announcement is hard in magnitude: RMB 4 billion–8 billion, 0.57% of total share capital, price cap RMB 1,200/share (8/31 close RMB 851.90, so the cap is 40.9% above the current price — this is standard practice and does not mean management sees RMB 1,200). The company's interim report is genuinely very hard too: revenue RMB 41.778 billion +182.49%, net profit attributable RMB 13.651 billion +241.70%, Q2 +38.05% QoQ, ROE 37.62%, gross margin 46.25%, operating cash flow per share +RMB 1.614.

⚠️ But there are two counterpoints that must be written alongside:

1. The buyback is for "an employee stock ownership plan or equity incentives", not cancellation. That means these shares will later be reissued to employees: total share capital does not shrink and EPS is not directly accreted. It is a signal (management thinks the price is low) and an incentive arrangement, not a shareholder return. Reading it as "RMB 4 billion of hard cash returned to shareholders" is a misreading. At the same time RMB 4 billion against a market cap of RMB 1,003.461 billion is just 0.40% (even at the RMB 8 billion cap it is only 0.80%).
2. Yesterday the communications-equipment sector rose 1.56% yet saw net outflows of RMB 3.426 billion, the second-largest net outflow market-wide, while its turnover of RMB 138.369 billion was the second-largest market-wide. High turnover + a rise + large net outflows is the standard shape of profit-taking. 中际旭创 itself actually fell 0.75% yesterday.

How to read it: the buyback is today's incremental positive, but it lands on a sector that was being sold yesterday. That is exactly why this branch gets A− rather than A.

2.4 ⚠️ A Mapping Rule Falsified Yesterday — the Ban Stays in Force Today

Yesterday's pre-market issue used overnight US equities to make a falsifiable inference: "MU/WDC holding up while NVDA/AMAT fell hard → A-share memory will be stronger than compute". The live A-share test yesterday ran exactly the opposite way: the compute group +2.69%, the design/equipment group +3.61%, while the memory group was only +0.78%, with 长鑫科技 (ChangXin Memory, 688825) at −1.01% the weakest in the memory group.

Today the overnight US tape offers the same shape of temptation again: SanDisk +5%, Micron +2%, SK Hynix +2% closing higher against the tape. Per the rule written into yesterday's post-close issue:

Stop transposing "overnight US sub-industry divergence" onto A-share sub-industries of the same name. The information content of the overnight US tape for A-share semiconductors should be revised down as a whole, not sliced more finely.

This issue enforces that ban: branch five (semiconductors) in §2 relies only on "yesterday's own A-share volume, money and price move", and does not use the rise in US memory names as a reason to be bullish on A-share memory names. 长鑫科技 (688825) retains in this issue the downgrade given in yesterday's post-close issue (see §6).

2.5 ⚠️ Tankers: Right Direction, but the Popular Causal Chain Is Wrong

Almost every Chinese write-up today will frame it as "oil price surges → tankers benefit". Mechanically that chain is backwards, and it has to be taken apart here.

First, in a tanker's revenue formula the oil price is a cost that gets deducted, not revenue. The industry-standard definition: TCE = (voyage revenue − voyage costs, including bunkers and port charges) ÷ voyage days. With the freight-rate index (WS) unchanged, a rising oil price mechanically depresses TCE. The Baltic Exchange's 8/21 weekly report (also republished by thedcn.com.au) lists high bunker prices alongside the strait situation as reasons WS was pushed up — i.e. a high oil price is a cost that has to be compensated for by the freight rate, not a source of profit.

Second, the four VLCC route readings from 8/28 falsify "oil-price-driven" outright. If the driver were the oil price, all four routes should move the same way; in fact they diverge (the Baltic Exchange 8/28 weekly report as republished by thedcn.com.au, link¹⁸; this issue could not obtain the balticexchange.com original page):

On the same day, under the same oil price, TD3C which transits the strait surged, while TD15 and TD22 which do not were falling. The oil price cannot explain that divergence; Hormuz can. Against TD34, which is also in the Persian Gulf but outside the strait, TD3C carries a premium of about USD 428,000/day (2.95x) — that is the market's price tag on Hormuz risk.

Third, and most important: TD3C's USD 647,000/day is not money these three companies can actually earn. Hormuz has been effectively closed to commercial shipping for about half a year: on 8/23 only 3 vessels transited in a single day (versus a norm of about 85/day) (USNI News 8/28¹⁹), and trackable vessels fell to 5 last weekend (Epoch Times 8/31²⁰); war-risk premiums are roughly 40x pre-crisis levels (shipping-transit tracker straits.live²¹; that figure is on a third-party tracker's basis and no primary confirmation from an insurer was obtained). A quote printed on a route almost no ship is using is a screen rate, not a transacted rate.

Both companies have already admitted this in their own announcements:

• 中远海能's 7/11 preannouncement, verbatim: "On 28 February 2026 the navigability of the Strait of Hormuz changed materially, and some of the Group's vessels located inside the strait chose to wait for a period… The Group has carefully considered potential factors such as lost operating days, contract disputes and freight recovery and their impact on second-quarter results."
• 招商轮船's 7/6 preannouncement, verbatim: "In the second quarter we overcame the phase of tanker oversupply that followed the closure of the strait."

⚠️ That sentence deserves a second reading: in Q2 the strait closure created a phase of tanker oversupply; it is not a one-way positive. Neither "fully open" nor "fully closed" is optimal for tanker companies; "partly blocked but with a viable detour" is. This risk almost never appears in the write-ups circulating in the market.

Fourth, how to compute honest earnings elasticity. In its preannouncement 中远海能 benchmarks itself against TD15 and TD22, not TD3C, and discloses H1 average TCE: TD15 at USD 117,773/day (+180% YoY) and TD22 at USD 111,083/day (+170% YoY). Working back from the company's own benchmarks:

• TD15: 8/28 spot 179,600 vs H1 average 117,773 = +52.5%
• TD22: 8/28 spot ~171,000 vs H1 average 111,083 = +54.0%

If the 8/28 spot level held for the whole of Q3, the earnings centre of the tanker business would step up about 50% from H1 — not the 5.5x implied by TD3C. This is an assumption (it presumes vessel positions can take non-strait cargoes and that no further stoppage losses occur), not a forecast.

Fifth, pricing progress: this is day 2, not day 1. The US strike occurred in the early hours of 8/31 Beijing time and was already in print at Sina 8/31 07:14, before the A-share open. Yesterday's A-share reaction was: 中远海能 +2.18%, 招商南油 +1.93%, while 招商轮船 was −1.21%; all three saw synchronised volume-driven pushes between 14:30 and 15:00. The same day ports and general shipping fell broadly (宁波港 (Ningbo Port) −0.58%, 厦门港务 (Xiamen Port) −2.23%, 海峡股份 (Strait Shipping) −2.79%), showing this was a targeted tanker move, not a broad "inflation beneficiary / cyclical" move.

There are parts that remain unpriced today: ① Brent's ICE settlement is at 02:30 Beijing time on 9/1, and when A-shares closed at 15:00 yesterday that leg had not played out; ② WTI continued to rise about +0.76% in the Asian morning session on 9/1 (reading 9/1 07:00 CST); ③ the geopolitical situation kept developing overnight.

⚠️ One circulating claim also needs correcting: "US equities fell on inflation fears from rising oil" is an over-reading — the S&P was only −0.33% on 8/31 (noise level) while the energy sector XLE rose 2.04%, a textbook "energy up, everything else slightly down" rotation under an oil shock.

---

3. Master Ranking of Single-Stock Catalyst Strength

Sorted by the total score from the §4 scoring model, descending by name. All price moves are on the 8/31 close basis; "seal ratio" = seal-order funds ÷ that day's turnover value, a column that exists only for limit-up names. ⚠️ The board column is also a risk scale: ChiNext / STAR Market ±20%, BSE ±30%, SSE & SZSE main boards ±10%. On ChiNext +12% is not a limit-up, it is just up 12%.

---

4. Single-Stock Scoring Model (out of 100)

Component breakdown for four representative names (to make the scoring logic checkable):

| Dimension | 昆仑万维 (300418) | 中际旭创 (300308) | 中远海能 (600026) | 芒果超媒 (300413) | 中文在线 (300364) | |---|---|---|---|---| | Source authority (0–15) | 13 (ministry document + company interim report) | 15 (company announcement) | 14 (military statement + Baltic Exchange weekly) | 12 (official Weibo + NRTA regulation) | 11 (NRTA regulation) | | Directness of the catalyst (0–20) | 17 (AI short dramas are its core business) | 18 (its own buyback) | 12 (industry trend, not a company event) | 19 (the company is the producer, the highest on the list) | 14 (short-drama content producer) | | Earnings elasticity (0–15) | 14 (AI short-drama monthly gross billings already quantified) | 13 (earnings already delivered) | 13 (about +50% working back from the company's own TD15/TD22 benchmarks) | 4 (no quantified path from ratings to revenue; the company itself states the AI project "does not directly generate economic benefits") | 3 (continuing losses) | | Industry position and fundamentals (0–15) | 11 (net profit +227%, negative cash flow, 3 points deducted per the model) | 14 (ROE 37.62%, cash flow +1.614) | 14 (ex-non-recurring +150.9%, operating cash flow +128.5%, profit-to-cash 1.53x) | 5 (operating profit of only RMB 30.2 million, Q2 ex-non-recurring loss; but zero goodwill + RMB 4.1 billion net cash + 29.41% leverage are genuine positives) | 2 (ROE −15.38%, PB 75.44) | | Expectation gap (0–10) | 9 (no limit-up yesterday) | 8 (announced post-close at 20:12) | 7 (the interim was already preannounced on 7/11, not an increment; but the overnight settlement is unpriced) | 6 (the event is an increment but the earnings were priced on 8/24) | 3 (already limit-up, information priced) | | Theme persistence (0–10) | 8 | 5 (sector money is flowing out) | 7 | 7 | 6 | | A-share trading characteristics (0–10) | 9 (unsealed, room left) | 6 (−0.75% yesterday) | 8 (+2.18%, −26.3% from the year-to-date high) | 10 (first sealed 09:35, 0 failed seals, turnover rate 6.41%, the best market-wide) | 5 (sealed at 13:00, seal ratio 0.101) | | Risk deduction (0 to −15) | −2 (negative cash flow, PE 161) | −4 (buyback is not cancellation, sector net outflow) | −5 (pricing day 2, screen-rate trap, reopening risk) | −10 (net selling of RMB 134 million by foreign capital + institutions absorbed by hot money −5, forward PE 78.7x −3, 20cm drawdown −2) | −10 (PB 75.44, losses, hot money sealing late in the session) | | Total | 79 | 75 | 70 | 53 | 34 |

⚠️ Three notes on the basis of this model — read them together, they determine how these scores should be used:

① The seven positive dimensions cap out at 95 points in total, not 100 (15+20+15+15+10+10+10 = 95), with the risk item then deducting 0 to −15. So the scores in this table should be read as being out of 95; a name not reaching 100 does not mean something is missing. The components of the five names in the table have each been recomputed and tie strictly to the totals (79 / 75 / 70 / 53 / 34), and the §3 master ranking is sorted on the same basis.

② "Industry position and fundamentals" and "Risk deduction" would double-count the same bad news, and this table avoids that. How: fundamental problems are reflected only in "Industry position and fundamentals" (which is why 芒果超媒 gets 5 there), while "Risk deduction" counts only institutional distribution, forward valuation and the daily price limit, with no repeat deduction for fundamentals. If both were deducted, 芒果超媒 would score 43 — that would be the model double-counting, not its true relative position.

③ 芒果超媒 (300413) is downgraded in this issue to "watch only, 53 pts" and removed from the recommendation slots. The basis is three mutually independent, mutually consistent pieces of evidence, detailed in §5.11: operating profit of only RMB 30.2 million (−97.6% cumulatively over two years), a Q2 ex-non-recurring loss of RMB 5.51 million, and RMB 168 million of the RMB 202 million net profit attributable coming from the recognition of deferred tax assets; core membership revenue −22.6%; and on the limit-up day the Dragon-Tiger List showed Shenzhen Connect net selling RMB 117.7 million and the institutional-only seats 1 buy 2 sells for net selling of RMB 16.38 million, with three hot-money seats absorbing.

---

5. Detailed Analysis of the Top 10 Names

51 昆仑万维300418ChiNext ±20% | Priority deep-dive | 79 pts · Kunlun Tech

• Related news: ① MIIT's special action to cultivate AI application service providers (China News Service 8/31 21:20³); ② 《微短剧发展管理办法》 taking effect today (Xinhuanet 7/31⁴); ③ the AIGC content re-rating driven by the AI long-form drama event.
• Positive logic (direct, affects revenue): this is the only company in this issue whose AI short-drama operating scale has already been quantified in disclosure (⚠️ the "gross billings" below are platform transaction value and do not equal recognised revenue; the difference after channel splits is material). Per relevant disclosure in the company's 2026 interim report: as of end-June 2026, monthly gross billings of the AI short-drama platform business exceeded USD 65 million; SensorTower data show the company's AI short-drama platforms combined ranked first in MAU in the overseas short-drama market in H1, FreeReels ranked first in downloads among overseas short-drama apps and was the only product in the overseas short-drama track with estimated downloads above 200 million in H1 2026; DramaWave has completed its AI-native transition, and in July 2026 more than 90% of new titles added on the platform each month were AI-generated (Xinhua Daily 8/19²²). "More than 90% of new titles AI-generated" is the hardest already-realised number this issue could find on AIGC cost reduction.
• Stage of the industry branch: day 2 of the theme, and it has not run its own course — up 8.89% yesterday with no limit-up, while 芒果超媒, 中文在线 and 荣信文化 in the same branch all hit 20cm limit-ups.
• Fundamental verification: interim revenue RMB 5.359 billion, +43.55%; net profit attributable RMB 1.088 billion, +227.17%; Q2 +322.66% QoQ (profit is heavily concentrated in Q2, matching the timeline of the AI short-drama ramp — a positive cross-confirmation); gross margin 65.58%; ROE 7.92%. Overseas revenue RMB 5.203 billion, 97.1% of the total. ⚠️ Operating cash flow per share is −RMB 0.364, i.e. negative, diverging from the income statement, most likely because user-acquisition spending is front-loaded in selling expenses (earlier quarterly reports showed selling expenses up more than 80% YoY); this issue deducts 3 points for it under the scoring model.
• Industry position: first in MAU in the overseas AI short-drama track, a niche leader.
• Technical sentiment: ChiNext ±20%; 8/31 +8.89%, turnover RMB 4.753 billion, turnover rate 9.50%, unsealed. No recent record of consecutive boards, so it is not a high-level accelerating name.
• Risks: ① negative operating cash flow; ② PE (TTM) 161.13 — ⚠️ to be clear: that TTM figure is inflated by losses in H2 2025; simply annualising the RMB 1.088 billion of H1 2026 would give about 26x, but half-year annualisation is a rough illustration, not a forecast, and cannot be used as a valuation conclusion; ③ 97% overseas revenue brings FX and local regulatory risk; ④ the user-acquisition model is sensitive to customer-acquisition cost.
• Final judgement: priority deep-dive. It is the only name in this issue that simultaneously satisfies "direct catalyst + earnings already delivered + AI revenue already quantified + has not run its course yesterday". That is the entire reason it ranks ahead of 芒果超媒 — not a stronger catalyst, but fundamentals that hold up.

52 中际旭创300308ChiNext ±20% | Watch closely | 75 pts · Innolight

• Related news: at 20:12 on 8/31 it announced a planned buyback of RMB 4 billion–8 billion at a price cap of RMB 1,200/share, roughly 6.6667 million shares (0.57% of total share capital), for an employee stock ownership plan or equity incentives, with a 12-month term from board approval (Sina Finance¹).
• Positive logic: a post-close announcement = today's increment. The RMB 1,200 buyback price cap is 40.9% above the 8/31 close of RMB 851.90 (this is a standard setting and does not represent management's price expectation).
• Fundamental verification (very hard): interim revenue RMB 41.778 billion, +182.49%; net profit attributable RMB 13.651 billion, +241.70%; Q2 +38.05% QoQ (growth is accelerating, not carried by Q1); gross margin 46.25%; ROE 37.62%; operating cash flow per share +RMB 1.614, so profits are cash-backed. These are the most solid fundamentals of any name in this issue.
• Industry position: the global optical-module leader, market cap RMB 1,003.461 billion.
• ⚠️ Three counterpoints that must be written alongside:
  1. The buyback is for an ESOP / equity incentives, not cancellation — the shares will later be reissued to employees, so total share capital does not shrink and EPS is not directly accreted. This is a signal and an incentive, not a shareholder return.
  2. The magnitude has to be seen against market cap: RMB 4 billion ÷ RMB 1,003.4 billion = 0.40%, and even at the RMB 8 billion cap only 0.80%.
  3. The sector it sits in was being sold yesterday: communications equipment rose 1.56% on 8/31 yet saw net outflows of RMB 3.426 billion (the second-largest market-wide), with turnover of RMB 138.369 billion, also second-largest — the standard shape of profit-taking. 中际旭创 itself actually fell 0.75% yesterday.
• Technical sentiment: 8/31 −0.75%, turnover RMB 18.926 billion, turnover rate 2.02%. PE 49.06, PB 25.18.
• Final judgement: watch closely. The news is hard and the earnings are hard, but it lands in a sector that is seeing outflows, and the nature of the buyback is widely misread. Today's verification point is in §8.

53 恺英网络002517SZSE main board ±10% | Priority deep-dive | 73 pts · Kaiying Network

• Related news: 《微短剧发展管理办法》 taking effect today + MIIT's AI application special action. ⚠️ This issue found no company-level announcement from 恺英网络 within the scan window; its inclusion rests on "best fundamentals in the branch + smallest gain yesterday", not on a company-specific catalyst. That is stated honestly, without exaggeration.
• Fundamental verification (best in the branch): interim revenue RMB 4.797 billion, +86.06%; net profit attributable RMB 1.428 billion, +50.34%; ROE 13.74%; gross margin 76.87%; operating cash flow per share +RMB 0.685 — the only company in the "AIGC content" branch with clearly positive operating cash flow. ⚠️ Q2 was −17.01% QoQ, so note that the growth slope is flattening.
• Valuation: PE 15.40, PB 3.50, market cap RMB 36.683 billion. The lowest PE in the whole strong branch. For comparison: 中文在线's PE is negative, 芒果超媒 47.74, 昆仑万维 161.13, 华策影视 62.37.
• Technical sentiment: 8/31 +3.43%, turnover RMB 1.316 billion, turnover rate 4.17%. No limit-up, no consecutive boards, positioning is not elevated.
• Risks: ① its linkage to "micro-dramas" is weaker than that of pure short-drama names, so this is an industry-trend positive rather than a direct benefit; ② game-approval and gross-billings volatility; ③ Q2 turned negative QoQ.
• Final judgement: priority deep-dive. The logic is: if the AIGC content line does become a main theme, money will sooner or later need a name that "delivers earnings, has an explicable valuation and is not elevated" to serve as the core holding — and it is the only one in this branch that qualifies. Conversely, if the line is a one-day wonder, its downside is also the smallest in the branch — an asymmetric payoff position.

54 瑞芯微603893SSE main board ±10% | Watch closely | 71 pts · Rockchip

• Related news: no new stock-specific news. It is included because of yesterday's semiconductor sectornet inflow of RMB 8.376 billion and turnover of RMB 260.856 billion, both first market-wide**(⚠️ the sector's +2.36% gain ranked only 9th) andits own seal time.
• Technical sentiment (key): limit-up on 8/31, first sealed at 09:37 — the only one of yesterday's top-5 limit-up names by turnover to seal in the morning session. Comparison group: 浪潮信息 13:51, 星网锐捷 13:46, 大族激光 14:14, 中文在线 13:00, 红板科技 14:16. Sealing in the morning and sealing late in the session imply completely different levels of next-day certainty. Turnover RMB 3.436 billion, turnover rate 4.21%, 1 failed seal, seal ratio 0.052.
• Fundamentals: PE 60.20, PB 16.73, market cap RMB 82.340 billion. The domestic leader in edge AI SoCs.
• Risks: ① 148 semiconductor names rose yesterday but only 2 hit limit-up (瑞芯微 603893, 博通集成 (Bestechnic, 603068)) — a broad rise, not a ladder; if there is still no ladder today, it is index-weight-driven rather than theme-driven and persistence should be discounted; ② seal ratio 0.052 is low; ③ PB 16.73 is not cheap.
• Final judgement: watch closely. It has the cleanest technical setup in the semiconductor branch.

55 中远海能600026SSE main board ±10% | Watch closely | 70 pts · COSCO Shipping Energy

• Related news: Hormuz escalation — on 8/30 the US military struck two Revolutionary Guard launch installations on Iran's Larak Island (the US side said it had detected preparations to "fire mine-carrying rockets into the Strait of Hormuz"), and Iran subsequently fired missiles at US bases; trackable vessels through the strait fell to 5 over the weekend; the Brent November contract settled +2.71% at USD 90.49.
• ⚠️ The positive logic must be read via the causal chain in §2.5, not as "oil up → tankers up". The real driver is constrained effective capacity through the strait + the war-risk premium, not the oil price itself; in the TCE formula the oil price is a cost that gets deducted. The 8/28 route data are hard evidence: TD3C, which must transit the strait, was +54 points week-on-week to WS623 (TCE > USD 647,000/day), while TD15, which does not, was −17 points and TD22 was −USD 1 million.
• Fundamental verification (the highest quality of the three): interim revenue RMB 15.146 billion +30.03%; net profit attributable RMB 4.545 billion +143.21%; ex-non-recurring RMB 4.449 billion +150.9% (non-recurring items are only 2.1% of net profit attributable, so the earnings quality is clean); Q2 net profit attributable RMB 2.372 billion, +9.1% QoQ (still growing, not peaking); gross margin 41.39%; operating cash flow RMB 6.962 billion, +128.5%, profit-to-cash conversion 1.53x; leverage 49.04%; ROE 9.26%.
• ⚠️ The interim report is not an increment, and this must be stated clearly: the company issued an earnings preannouncement on 2026-07-11 (announcement no. 2026-048) guiding to net profit attributable of about RMB 4.50 billion (roughly +141%), versus the actual RMB 4.545 billion — beating the preannouncement by only about 1.0%. If a double narrative of "interim report far above expectations + rising oil prices" appears today, at least one of the two is a false catalyst.
• Valuation and positioning: PE (TTM) 15.66, PB 2.19, market cap RMB 105.157 billion. Still 26.3% below its 2026 year-to-date high — the sector already went through one explosive re-rating and pullback in the early phase of the strait events in February–May, so this is not starting from zero, but it is not at a high level either.
• Yesterday's tape: 8/31 +2.18%, spiked and faded in the morning to 18.39, then two volume-driven pushes between 14:30 and 15:00, closing at 19.22 (day high 19.35, day low 18.38); turnover RMB 1.414 billion, turnover rate 1.96%. The same day ports and general shipping fell broadly (宁波港 −0.58%, 厦门港务 −2.23%, 海峡股份 −2.79%), showing this was a targeted tanker move, not a broad inflation/cyclical move.
• Why this one rather than 招商轮船 (601872): ① 招商轮船 is not a pure tanker play — its 7/6 preannouncement itself said dry bulk was +170% QoQ in Q2 and tankers +50% QoQ, so a substantial part of its earnings elasticity comes from the BDI, and treating it as "the name with the greatest oil elasticity" is a mismatch; ② 招商轮船's PE (TTM) of 13.96 looks cheapest, but its PB of 3.23 is the highest of the three and its 800-day closing-price percentile is 97.5% — the classic cyclical combination of "low PE + high PB + high percentile", where E has already been pushed to its maximum by peak-cycle earnings; ③ it fell −1.21% yesterday, the only one of the three to close lower, and its interim was fully preannounced on 7/6, zero increment.
• Why not 招商南油 (601975): it is the only one of the three whose interim was a genuine increment (+42.4% did not trigger the SSE's 50% mandatory preannouncement threshold, so the interim was first disclosed on 8/28), but the incremental content is negative: Q2 net profit attributable −12.0% QoQ (revenue +10.9% QoQ, so more revenue but less profit, with the quarterly net margin falling from 27.5% to 21.8%), operating cash flow −6.1% YoY, and non-recurring items jumping from 0.7% to 9.9% of net profit attributable — three quality metrics deteriorating together, while it has the largest gain since the start of August.
• Risks: ① ⚠️ TD3C's USD 647,000/day is a screen rate, not a transacted rate — the strait has been effectively closed to commercial shipping for about half a year, with only 3 transits on 8/23 (versus a norm of about 85/day), so extrapolating earnings linearly from it would overstate them by more than 3x; ② ⚠️ both "the strait reopening" and "it staying fully closed" are risks — in its own 7/11 preannouncement the company wrote that the change in the strait caused "lost operating days, contract disputes and freight recovery" effects, and "partly blocked but with a viable detour" is the optimal state; ③ war-risk premiums are about 40x pre-crisis levels, lowering the conversion of gross freight into net profit; ④ this issue did not obtain a primary reading of the Baltic indices for 8/31 (Monday), so the table above is on the 8/28 basis and must not be written up as "the latest"; ⑤ OPEC+ output cadence and newbuild deliveries, for which this issue obtained no primary data.
• Final judgement: watch closely. The event is real, the direction is positive and the company is the highest quality of the three, but this is pricing day 2 rather than day 1, and the real elasticity is far smaller than the magnitude implied by the screen rate. The reason to participate is "the overnight settlement and the Asian morning move were not priced by A-shares yesterday", not "oil prices surged".

56 华策影视300133ChiNext ±20% | Watch closely | 69 pts · Huace Film & TV

• Related news: 《微短剧发展管理办法》 takes effect today — producers of class I / II micro-dramas must hold the 《广播电视节目制作经营许可证》, have registered capital above RMB 1 million and no foreign investment, and Huace, as a leading licensed producer, benefits from the higher licence barrier.
• ⚠️ One contamination this issue actively excluded: the search surfaced a Cailianpress item, "Huace Group's Beijing short-drama team: two AI-created micro-dramas expected in September–October", highly consistent with today's main theme. Fetching the original page confirmed a publication time of 2024-06-25, two-year-old news, and it was discarded. This issue attributes no same-day AI-related catalyst to Huace.
• Fundamental verification (the only one in the branch that holds up): interim net profit attributable RMB 167 million, +41.98%; gross margin 40.80%; Q2 +5.33% QoQ; ROE 2.27%. Among the 22 media names surveyed in this issue, it is the only film & TV company that is "revenue above RMB 500 million + profitable + growing net profit attributable".
• ⚠️ But the quality of that growth has to be spelled out: revenue RMB 685 million, −13.22% YoY — profit growth comes from the cost side, not the revenue side; and operating cash flow per share is −RMB 0.112, i.e. negative. This is not a high-quality +41.98%.
• Technical sentiment: 8/31 +15.77% but no limit-up (ChiNext ±20%, so +15.77% is still 4.23pct short of the limit); it opened only +0.58%, so open-to-close was +15.11% and virtually the entire gain was made intraday; turnover RMB 1.908 billion, turnover rate 15.03%. ⚠️ A 15.03% turnover rate is high, indicating that chips changed hands thoroughly — this is not a locked-position shape.
• Final judgement: watch closely. It embodies the contradiction of "the most explicable fundamentals" and "the most indirect catalyst" in this branch.

57 伟测科技688372STAR Market ±20% | Watch closely | 68 pts · Wtest Technology

• Related news: no new stock-specific news; this is continuation verification of yesterday's recommendation slot.
• Yesterday's performance: +6.25%, opening −4.25%, so open-to-close +10.97%, the strongest of yesterday's 5 recommendation slots; turnover rate 7.10%. Yesterday's self-set verification point, "can it recover the −3.75% of 8/28", was recovered and exceeded.
• Fundamentals: ex-non-recurring growth is 3x that of net profit attributable, and its valuation is the lowest in the group (conclusion from yesterday's issue); PE 61.12, PB 5.20, market cap RMB 22.091 billion.
• Risks: it has risen for several sessions, so today is the second verification; if volume shrinks and it closes down, it should be downgraded.
• Final judgement: watch closely, verification point in §8.

58 国科微300672ChiNext ±20% | Watch closely | 67 pts · Nationalchip

• Related news: no new stock-specific news; continuation of yesterday's recommendation slot.
• Yesterday's performance: +5.11%, gapped down −1.03% and then climbed all day, open-to-close +6.21%; turnover rate 5.27%, meeting yesterday's self-set verification point of "break above Friday's 3.18% turnover rate on volume".
• Fundamentals: ex-non-recurring +1,917.55%, operating cash flow 3.6x (verified in yesterday's issue). ⚠️ PE (TTM) is −665.64, distorted by historical losses, and that basis cannot be used for valuation judgement; PB 9.57.
• Final judgement: watch closely, today is the second consecutive day of verification.

59 齐鲁银行601665SSE main board ±10% | Watch closely | 65 pts · Qilu Bank

• Related news: the provisions of the 36 property-credit measures affecting bank asset sides (Article 12 lead-bank system, Article 13 extending development-loan tenor to a maximum of 7 years with first principal repayment pushed to after completion filing, Article 20 mortgages of up to 40 years, Article 22 mortgages to be disbursed only after completion filing).
• Yesterday's performance (the single piece of shape evidence this issue values most): +6.68%, and it opened at +3.19% — in a morning where all three major indices gapped down (SSE Composite −0.65%, ChiNext Index −1.67%), banks were one of very few sectors to gap up. Being priced in the auction phase means this was a consensus formed overnight, not intraday chasing. At sector level: banks +1.38%, net inflow RMB 6.729 billion, second market-wide, 38 up 4 down.
• Fundamentals: interim revenue RMB 7.497 billion +10.55%, net profit attributable RMB 3.174 billion +16.06%, operating cash flow per share +3.003; PE 7.03, PB 0.83.
• Risks: already up 6.68% yesterday; city commercial banks have high elasticity but also high volatility; if it gaps down and merely follows today, the first day's pricing is complete.
• Final judgement: watch closely, verification point in §8.

510 中科创达300496ChiNext ±20% | Watch closely | 63 pts · Thundersoft

• Related news: MIIT's special action to cultivate AI application service providers (8/31 21:20). The document's language encouraging service providers to build "forward-deployed engineer (FDE) teams rooted at customer sites" points precisely at delivery-type AI / smart-terminal service providers such as Thundersoft.
• Fundamentals: interim revenue RMB 4.305 billion, +30.46%; net profit attributable RMB 246 million, +54.97%; Q2 +39.88% QoQ (accelerating); gross margin 29.08%. ⚠️ Operating cash flow per share is −RMB 0.138, negative. PE 51.13, PB 2.60, market cap RMB 27.468 billion.
• Technical sentiment: 8/31 +5.25%, turnover RMB 995 million, turnover rate 4.64%, no limit-up, positioning not elevated.
• Risks: ① the policy gives no monetary guidance, so the benefit cannot be quantified; ② negative cash flow; ③ the AI-application-service-provider concept is extremely diffuse with a poor competitive structure.
• Final judgement: watch closely.

511 [Feature] 芒果超媒300413ChiNext ±20% | Watch only (downgraded this issue from "watch closely") | 53 pts · Mango Excellent Media

It ranks 17th by total score and is no longer in the Top 10, yet this issue still gives it its own section, because the judgement itself — "the most direct catalyst market-wide + the best seal quality market-wide + and still no recommendation slot" — is more worth writing out than any single name on the list.

⚠️ It ranks 17th by total score and is no longer among the recommendation slots. This issue still gives it its own section because the judgement itself — "the most direct catalyst market-wide + the best seal quality market-wide + and still no recommendation slot" — is more worth writing out than any name on the list. The three pieces of evidence below are mutually independent and mutually consistent.

• Related news: at 21:06 on 8/31 Hunan TV's official Weibo said the premiere of 《后西游记》 "ranked first among provincial satellite channels in real-time ratings in its time slot" (Sina²). The drama premiered on Mango TV at 18:00 on 8/31 and in Hunan TV's 20:00 prime slot, with all 30 episodes AIGC-generated and no human actors.
• Positive logic: 芒果超媒 is the only listed company that is a party to this news (produced by Mango TV, aired by Hunan TV), which is why "directness of the catalyst" scores 19, and that has not changed.

First piece of evidence for the downgrade: the income statement is far worse than "net profit attributable −73.58%".

Item (RMB mn)	2026H1	2025H1	2024H1
Operating profit	30.2	569	1,233
Pre-tax profit	36.1	580	1,237
Income-tax expense	−168.3 (benefit)	−182.1	+194
Net profit attributable	201.7	763.4	1,043

Table scrolls horizontally

Over two years operating profit went from RMB 1,233 million → RMB 569 million → RMB 30.2 million, a cumulative −97.6%. The RMB 202 million of net profit attributable is built like this: pre-tax profit of RMB 36 million + an income-tax benefit of RMB 168 million. The balance sheet shows deferred tax assets rising from RMB 1.5286 billion at the start of the year to RMB 1.6970 billion, an increase of RMB 168.4 million, corresponding almost exactly to the RMB 168.3 million tax benefit — this profit comes from recognising deferred tax assets, a book item that generates no cash, and it embeds the assumption that "the company will be profitable in future and able to use these deductions". More importantly, here is what the company itself writes in the "analysis of non-core business" section of the interim report: investment income of RMB 119.66 million, equal to 331.65% of pre-tax profit, mainly interest income from time deposits, large-denomination certificates of deposit and structured deposits, "sustainable: no". Pre-tax profit is RMB 36 million while wealth-management interest of RMB 120 million equals 331.65% of it — strip out the wealth-management interest and the core business is loss-making before tax. That is the company's own framing, not this issue's computation. Q2 alone: net profit attributable of RMB 2.32 million (−98.8% QoQ) and ex-non-recurring of −RMB 5.51 million, already a loss. Almost all of the RMB 202 million in the interim was earned in Q1. Operating cash flow was −RMB 341.7 million (versus +RMB 460.1 million a year earlier), with profit-to-cash conversion of −169%.

Second piece of evidence for the downgrade: the quality behind revenue growth of +3.86% is negative (the company does not disclose this breakdown; it is computed in this issue from the two interim reports' original text).

Business	2026H1	2025H1	YoY	Gross margin (26H1)
├ Membership	1,933	2,496	−22.6% (RMB 563 million less)	—
├ Advertising	1,684	1,587	+6.11%	—
└ Operators	845	800	+5.63%	—
New-media interactive entertainment content production	614	606	+1.24%	20.03% (−6.03pp)
Content e-commerce	1,100	446	+146.35%	6.16% (−12.42pp)
Other	18	28	−36.53%	3.85%
Total	6,194	5,964	+3.86%	20.05% (−6.51pp)

Table scrolls horizontally

① Membership revenue is −22.6%, RMB 563 million lower in absolute terms. This is the company's most core, highest-margin engine, and the decline in this single line exceeds the entire fall in operating profit. In the interim report the company uses only the words "under pressure" and "declined", and never gives this −22.6%. ② Revenue growth of +3.86% is entirely carried by content e-commerce at a gross margin of just 6.16% — RMB 654 million of extra revenue brought only RMB 68 million of gross profit, actually less than the RMB 83 million of the prior year. Replacing internet-video revenue at a 23.54% gross margin (the segment membership belongs to; the company does not disclose membership gross margin separately) with revenue at a 6.16% gross margin is a downgrade of the revenue mix, not a transformation. ⚠️ But it must be clear how much this explains: the structural effect computed as above (membership −RMB 563 million × 23.54% + e-commerce +RMB 654 million × 6.16%) is about −RMB 92 million, whereas company-wide gross profit actually fell from RMB 1.584 billion to RMB 1.242 billion, a drop of RMB 342 million. In other words the structural downgrade explains only about a quarter to a third of it, not the whole thing; the rest comes from rigid content costs (operating costs +13.06%, a growth rate 3.4x the +3.86% of revenue). ③ "Content e-commerce +146%" is a low-base recovery, not new growth — that business was −67.09% YoY in 2025H1, implying RMB 1.356 billion in 2024H1, so this year's RMB 1.100 billion is still below the level of two years ago.

Third piece of evidence for the downgrade: on the limit-up day, foreign capital and institutions were selling while hot money was buying.

The 8/31 Dragon-Tiger List (triggered by "top 5 by daily gain of 15% or more") shows the listed seats accounting for 47.69% of the day's turnover and net selling of RMB 30.82 million in total:

Seat	Net amount
Shenzhen Connect dedicated	−117.7 million
Institutional-only (1 buy, 2 sells combined)	−16.38 million
Three hot-money seats incl. Guotai Haitong Shanghai Jing'an Xinzha Road (buy side)	+144 million
Subtotal of the seven seats above	+9.92 million
All other listed seats combined (incl. Huatai Securities Nanjing Changjiang Road −27.75 million)	−40.74 million
Dragon-Tiger List total	−30.82 million

Table scrolls horizontally

Basis note (the table is completed so it closes, for line-by-line checking): the Dragon-Tiger List discloses only the top five buy and sell seats; the first three rows above are the key seats this issue focuses on, summing on their own to net buying of RMB 9.92 million; adding the −RMB 40.74 million of the remaining listed seats gives the total net selling of RMB 30.82 million. When citing, please distinguish the two figures: "foreign capital + institutions net sold about RMB 134 million" and "the whole list net sold RMB 30.82 million" are two different bases — the former is this issue's argument, the latter is the list's total.

This issue's argument rests on structure, not on the total: Shenzhen Connect's single-seat net selling of RMB 117.7 million is the largest sell on the whole list, while the institutional-only seats were in fact 1 buy and 2 sells for combined net selling of RMB 16.38 million — the data source's automatic reading of "1 institution bought" is misleading. The absorbing side was three hot-money seats. ⚠️ A note on the basis of the turnover rate is also needed: turnover value of RMB 1.087 billion ÷ free-float market cap of RMB 17.348 billion = 6.26%, matching the 6.41% given by the quote data (computing on total market cap of RMB 31.765 billion would give 3.42%, which does not match), showing that this turnover rate is already on a free-float basis and should not be discounted a second time by the free-float share. The "low turnover" point stands, and A-share trading characteristics score 10. The real deduction is the Dragon-Tiger List structure, not the turnover rate.

Timeline (this is what caps "expectation gap" at 6 points):

Date	Event	Price
8/21	Last trading day before the interim	close 15.03
8/22	First disclosure of the 2026 interim report (triple-checked: no preannouncement among 26 CNINFO announcements from June to September, no such name among 4,885 records in Eastmoney's preannouncement database, and the 8/24 price-volume reaction)	—
8/24	The market priced the interim	−6.98%, volume up 3.3x
8/31	AI long-form drama premiere	+20.00% limit-up, close 16.98

Table scrolls horizontally

The 8/31 close of 16.98 is already 12.97% above the 15.03 that preceded the interim disclosure. One limit-up not only erased the entire fall from the earnings blow-up, it went 13% further.

• Positives that still stand (one cannot write only the bear side): leverage of 29.41% (the lowest range among peers), cash of RMB 2.775 billion + trading financial assets of RMB 1.386 billion against short-term borrowings of only RMB 80 million, i.e. net cash of about RMB 4.1 billion, zero goodwill on the consolidated balance sheet (no goodwill-impairment risk), and non-recurring items of only RMB 20.66 million, so no one-off gain dressed up profits in H1. PB of 1.38 and PS (TTM) of 2.26 are both the lowest among peers. This company faces no near-term solvency or blow-up risk and can afford the tuition for a transformation.
• ⚠️ But the PB of 1.38 has to be looked through: of RMB 23.084 billion of net assets, intangible assets of RMB 8.157 billion + receivables of RMB 4.053 billion + contract assets of RMB 1.231 billion = RMB 13.44 billion, 58% of net assets. The intangibles are mainly film and TV copyrights, with net amortisation of RMB 598 million in the period; receivables are being impaired at an accelerating pace (credit impairment of RMB 66.31 million, up about 2.1x YoY). The low PB and the realisability of these assets are two sides of the same question.
• ⚠️ The company itself has answered the question of AI's financial significance: in its special report on raised-funds usage, its description of the "Mango TV audio-video industry large-model matrix project" is that it "does not directly generate economic benefits, and therefore the project's benefits cannot be separately accounted for" (project size RMB 183.75 million). Every reference to AIGC in the interim report is a process metric (Mango Lingchuang has cumulatively served more than 40,000 professional users, generated content 1.71 million times and supported over 3,900 projects), with no monetary or percentage figure at all. AIGC revenue share: no reliable data available.
• Technical sentiment (its only strength): limit-up on 8/31, first sealed 09:35, zero failed seals all day, turnover rate 6.41%, seal orders of RMB 388 million ÷ turnover of RMB 1.087 billion = 0.357, the best seal quality among yesterday's 13 limit-up media names.
• Final judgement: watch only. The direction may still be right — yesterday's lesson from 我爱我家 (5i5j) is precisely that "a name passed on grounds of positioning or quality does not necessarily fall", so this issue keeps it on the list rather than removing it. But it simultaneously carries five things — collapsing fundamentals + profit resting on a book item and wealth-management interest + core business −22.6% + net selling of RMB 134 million by foreign capital and institutions on the limit-up day + a forward PE of 78.7x — which is not enough to occupy a recommendation slot.
• The real falsification/confirmation window is the Q3 2026 report (around late October), where two numbers matter: whether the YoY decline in membership revenue narrows, and whether the internet-video gross margin (23.54%) stops falling. Until those two numbers appear, "AI restructures the cost base" is a hypothesis, not a fact.

6. Pass List

Per the requirement of yesterday's post-close issue, Pass reasons are split into three columns. Yesterday's live test: the Pass group averaged −0.70% with 35.7% of names up (versus 58.9% market-wide), the most effective output in the whole issue; but all 4 names judged wrongly were passed for "poor financial quality". This issue therefore distinguishes explicitly: category one (narrative refuted by primary evidence) = removed; category two (positioning too high / already priced) = weighted down only, not removed; category three (poor financial quality) = weighted down only, not removed.

6.1 Category One: Narrative Refuted by Primary Evidence (removed from the list)

6.2 Category Two: Positioning Too High / Already Priced (weighted down only, not removed)

⚠️ Yesterday's direct lesson: 我爱我家 (5i5j, 000560) was passed for "positioning too high" and then locked limit-up from the open at +9.85%, the best performer on the entire list. The conclusion is that chip and positioning metrics determine size, not direction — the correct response is to reduce size, not to remove the name from the list. This column follows that strictly.

6.3 Category Three: Poor Financial Quality (weighted down only, not removed)

6.4 Directions to Explicitly Avoid (Not Single Names)

---

7. Within-Branch Rankings

7.1 AIGC Film & TV / Micro-Drama

Hardest name: 恺英网络 (002517) — its fundamentals are the only ones in the branch that are simultaneously profitable, growing, cash-flow positive and cheaply valued. Most recognisable name in trading terms: 芒果超媒 (300413) — first sealed 09:35, 0 failed seals, turnover rate 6.41%, the best seal quality in the branch. ⚠️ But "most recognisable in trading terms" and "worth buying" are two orthogonal conclusions: its Dragon-Tiger List shows foreign capital + institutions net selling about RMB 134 million on the limit-up day, absorbed by three hot-money seats, so this issue gives it "watch only" rather than a recommendation slot (details in §5.11). Who is following: 掌阅科技, 时代出版, 龙版传媒, 中广天择 — all sealed after 10:25 and none supported by earnings. Who is Passed (matching §6.1 category one name by name): 欢瑞世纪 (000892), 荣信文化 (301231), 果麦文化 (301052), 光线传媒 (300251), 慈文传媒 (002343), 横店影视 (603103), 儒意电影 (002739) — refuted by the companies' own statements. Who is only weighted down, not removed: 掌阅科技 (603533), 时代出版 (600551), 博纳影业 (001330), 中广天择 (603721), 龙版传媒 (605577), 中文在线 (300364).

7.2 AI Applications / Agents

7.3 Optical Modules / Compute Hardware

7.4 Domestic Semiconductor Design / Testing

Basis for downgrading 长鑫科技 (executing yesterday's post-close verification point): yesterday's verification point was "turnover rate expands above 5.02% and it closes above the opening price"; measured, turnover rate 7.94% met the bar but the close of 58.01 was below the open of 58.80, so only one of the two conditions was met and by the self-set criterion it falls on the "good news, distribution" side. Yesterday's post-close issue further specified: "if it again trades on volume without rising tomorrow, it should be downgraded from 'watch closely' to 'watch only'". This issue executes that downgrade.

7.5 Tankers (Hormuz)

Hardest name: 中远海能 (600026) — the only one of the three that simultaneously satisfies "pure tanker + clean ex-non-recurring quality + Q2 still growing QoQ + cash conversion of 1.53x + still 26% below the year-to-date high". Who is Passed: 招商轮船 (601872) — not because the company is bad; on the contrary, its ROE of 15.61% and ex-non-recurring purity are the highest of the three. The Pass reason is cycle positioning: a PE (TTM) of 13.96 looks cheapest, but that is because E has already been pushed to its maximum by peak-cycle earnings, while PB of 3.23 is the highest of the three and the price percentile is 97.5%. "Low PE + high PB + high percentile" is the textbook shape of a cyclical top, and a cheap PE carries low information content here. ⚠️ The biggest risk shared by the whole branch: TD3C's USD 647,000/day is a screen rate. With the strait effectively closed, only 3 vessels transited on 8/23, while the benchmarks the companies themselves use in their preannouncements are TD15/TD22 (H1 averages of USD 118,000 / USD 111,000 per day), implying spot elasticity of about +50%, not 5.5x.

---

8. Opening Verification Signals for Today

⚠️ A-shares have no continuous pre-market auction, and at the 07:00 writing time of this issue no same-day pre-market price data exists. Everything below is a criterion that can be judged on the spot during today's session, not a forecast. Method requirement (executing item 1 of yesterday's post-close "must fix" list): the execution-price basis for every recommendation slot in this issue is the opening price produced by today's 09:25 call auction, not the 8/31 close. On a gap-down day executing as listed costs less than the figure converted from the prior close; on a gap-up day the odds worsen accordingly.

8.1 Auction Signals (09:15–09:25)

8.2 Sector Signals (09:30–10:30)

8.3 Single-Stock Signals (Full Day)

8.4 Risk Signals (Downgrade on Sight)

1. Gap up then dive: media names collectively gap up >5% and turn negative within half an hour → yesterday's 13 first-boards are being distributed collectively, and the entire branch is void for the day.
2. A lone limit-up: in AIGC film & TV only 芒果超媒 seals while everything else turns negative → the leader is strong on its own with no relay from followers, the same shape as yesterday's second-hand housing agency (我爱我家).
3. The core holdings do not follow: both core-holding candidates 昆仑万维 and 恺英网络 trade without volume → money is only playing pure sentiment names, do not participate.
4. Yesterday's late-session sealers break at the open: 浪潮信息 (000977, SZSE main board ±10%, first sealed 13:51), 星网锐捷 (002396, SZSE main board ±10%, 13:46), 大族激光 (002008, SZSE main board ±10%, 14:14), 红板科技 (Redboard Technology, 603459, SSE main board ±10%, 14:16) — if these four break their boards within half an hour of the open, that is the fastest leading signal of market-wide sentiment weakening, because they were the entire support of yesterday's afternoon push.
5. A weak index: if the SSE Composite opens below 3,950 (−0.91% from yesterday's close of 3,986.30) and fails to recover it during the morning, then every theme judgement for the day is downgraded — yesterday the index closed at its high of the day, only 0.34% from 4,000, and a pullback from that level would trigger profit-taking.
6. Yesterday's strong themes fading: if media and semiconductors both weaken during the morning while the number of limit-ups is below 40 (88 yesterday), and if the failed-seal rate deteriorates further from 25.42%, then none of the day's theme judgements hold.
7. Offshore transmission: if Hong Kong stocks and the A50 both weaken sharply in the morning (the Nasdaq Golden Dragon China Index was −2.19% overnight), that means foreign sentiment is transmitting and index heavyweights will drag the index down.

---

9. Final Recommendations

9.1 The 5 Names Most Worth Watching Today

⚠️ Two necessary notes: ① No. 1 (昆仑万维 300418) and No. 3 (恺英网络 002517) belong to the same branch (AIGC content), so the branch-level risk is shared. If the "≥2 second consecutive boards" criterion in §8.2 fails during the morning, these two should be downgraded together rather than judged one by one. ② 芒果超媒 (300413) has been removed from the recommendation slots in this final draft and downgraded to "watch only". Its seal quality yesterday was the best market-wide and its catalyst is the most direct market-wide, but the fundamental check produced three mutually independent, mutually consistent pieces of evidence (operating profit of only RMB 30.2 million and −97.6% over two years, RMB 168 million of net profit attributable coming from recognition of deferred tax assets, and foreign capital + institutions net selling about RMB 134 million on the limit-up day while three hot-money seats absorbed), which is not enough to occupy a recommendation slot. It stays on the list rather than being Passed, because yesterday's lesson from 我爱我家 is that a name excluded on grounds of quality or positioning is not necessarily wrong on direction. The full re-rating process is in §5.11.

9.2 The 3 Strongest Branches Today

The fourth and fifth branches that follow closely were also given recommendation slots in this issue, and are listed here so that "only 3" is not misread: ④ Tankers (Hormuz escalation) — the geopolitical situation kept developing overnight + Brent settled at 02:30 Beijing time on 9/1 (after the A-share close yesterday). ⚠️ But this is pricing day 2, and the causal chain must be read per §2.5: the driver is constrained capacity through the strait, not the oil price. Representative: 中远海能 (600026). ⑤ Domestic semiconductor design/testing — yesterday the sector ranked first market-wide on both net inflow (RMB 8.376 billion) and turnover (RMB 260.856 billion) (the +2.36% gain ranked 9th), the only one of yesterday's three main themes to fully deliver; but today at least 3 limit-ups must appear for it to count as theme-driven rather than index-weight-driven. Representatives: 瑞芯微 (603893), 伟测科技 (688372), 国科微 (300672).

9.3 Directions Not Worth Chasing Today

9.4 The Final One-Sentence Judgement

Today is a day when two contradictions sit on the table at once — "the branch with the strongest catalyst has the worst fundamentals" and "the company with the strongest fundamentals sits in a sector that is being sold" — which is why the five recommendation slots in this issue are split across three different types of evidence: 昆仑万维 and 恺英网络 rest on "if this line holds, money will sooner or later come back looking for earnings"; 中际旭创 and 中远海能 rest on "the 20:12 announcement and the 02:30 settlement are both absent from yesterday's prices"; and 瑞芯微 rests on "sealing in the morning and sealing late in the session are not the same thing". And the single action in this issue most worth remembering is taking 芒果超媒 — best seal quality market-wide, most direct catalyst market-wide — off the recommendation slots, because the check found that RMB 168 million of its RMB 202 million net profit attributable came from recognising deferred tax assets, that operating profit is down to RMB 30.2 million, and that on the limit-up day foreign capital and institutions net sold RMB 134 million while three hot-money seats took the other side. There is only one criterion that really matters, and it will be visible during the morning: whether the AIGC film & TV line can produce two or more second consecutive boards today — 12 of yesterday's 13 limit-ups were first-boards, and if there is still no ladder today, then it was a sentiment move with a one-day lifespan, lit by one television drama and two policy documents, and not a main theme.

---

⚠️ Risk warning: this list is a pre-market information review and observation only and does not constitute investment advice. A-share volatility is extremely high, and automatically generated content may contain timeliness gaps or errors in industry-chain mapping; it cannot be used directly as a basis for trading. The board and daily price limit of each name are marked in the text, and the STAR Market and BSE additionally have investor-suitability thresholds (RMB 500,000 / 2 years, RMB 500,000); readers should decide according to their own permissions and risk tolerance. Data are subject to the formal disclosures of the exchanges and the companies.