A-Share Market Recap | 2026-08-20 Thursday

0. One-Sentence Recap

Sentiment rebounded sharply from the freezing point (118 limit-downs yesterday → 12 today; limit-ups 36 → 79), but the strongest theme of the day did not get a single word in the pre-market list. Of the 79 limit-ups market-wide, 37 (46.8%) came from the pharmaceutical chain (biologics 14, chemical pharma 12, medical services 4, medical devices 4, TCM 2, pharmaceutical distribution 1); the vaccine sector led at +13.64%, biologics +8.64%, and the innovative-drug concept took main-force net inflow of +RMB 6.261 billion. The catalyst was Merck and Moderna announcing on 8/19 US Eastern time that the Phase III INTerpath-001 trial met its primary endpoint (the world's first Phase III success for an mRNA therapeutic cancer vaccine); Moderna closed +176.97% that day on 163 million shares (28x the 10-day average volume of 5.81 million shares), with market cap rising to $69.619 billion; XBI +5.90%, IBB +6.58%, MRK +12.60%, BNTX +21.96%.

This catalyst fell entirely inside the scan window the pre-market piece declared for itself (8/19 15:00 → 8/20 07:30) — the 8/19 US close corresponds to 04:00 Beijing time on 8/20, and Chinese media had already reported it on the evening of 8/19. Not only did the pre-market fail to catch it, in §2 it rated "innovative drugs / commercial insurance" B− and stated "no recommendation slot in this piece," on the grounds that "innovative drugs were −2.99% on 8/19 with main-force −RMB 5.363 billion." Today that concept was +4.85% with main-force +RMB 6.261 billion, 184 up and 13 down.

The parts the pre-market got right must also be recorded honestly: the direction of the precious-metals branch played out completely (the precious-metals sector +5.56%, 12 up 0 down, gold main-force +RMB 1.977 billion), 赤峰黄金 (Chifeng Gold, 600988) +7.91% with main-force +RMB 536 million; the hardest calls on the avoid list also all played out — 宇树科技 (Unitree Robotics, 688836) −18.70%, 埃斯顿 (Estun Automation, 002747) −9.07%, 恒瑞医药 (Hengrui Pharma, 600276) −6.00%, 罗普斯金 (Luopusijin, 002333) −9.94%, 赛意信息 (SIE Consulting, 300687) gapped up +6.37% and closed −0.46%.

Tomorrow's tone: the pharmaceutical theme is at "day 1, no ladder (highest is only 2 consecutive limit-ups), sharply polarized seal quality" — that is the crux; precious metals already delivered today and the commodity leg gave back ground during the A-share session (COMEX gold fell from 4,579.50 at 07:01 Beijing time to 4,542.50 at 15:26, −0.81%), so tomorrow it moves into disagreement; the housing-fund chain and the coke chain are confirmed to be receding.

---

1. Market Overview

• Combined turnover of the two exchanges: RMB 2,079.36 billion; share volume contracted versus 8/19 (SSE −11.5%, SZSE Component −14.4%), i.e. this is a repair on shrinking volume, not a counterattack on expanding volume.
• The divergence of weak indices and strong individual stocks is extremely pronounced: the SSE Composite was only +0.24% and the SSE 50 −0.49%, while market-wide 4,096 stocks rose / 1,347 fell / 98 were flat (full sample of 5,900, obtained by cross-taking both ascending and descending pagination, so there is no tail gap). The CSI 500 at +0.86% beat the SSE 50 by 1.35pp — heavyweights are dragging, small- and mid-caps are rising.
• Limit-ups / limit-downs (on the official limit-up and limit-down pool basis, with len(pool)==tc verified):

Sector price-limit basis has been judged separately per iron rule 3: of today's 79 limit-ups, ±20% ChiNext/STAR names total 17 and the ±30% BSE name is 1 (三元基因 Sanyuan Gene, 920344, +30.0%), the rest being main-board ±10%. Had a uniform 10% test been applied, all 18 of these would have been misrecorded.

• Sentiment temperature: a sharp rebound from the freezing point to "single-track euphoria + overall caution." Limit-downs collapsing from 118 to 12 is the hardest evidence of repair; but three pieces of evidence show this is not broad euphoria: ① the seal-break rate actually rose from 25.0% to 36.8% (46 names touched the limit without sealing — money swarming into one track degraded relay quality); ② board height reached only 4 and that was a lone agricultural-processing name, while the pharmaceutical theme topped out at just 2 boards (汉森制药 Hansen Pharmaceutical, 002412), with no ladder at all; ③ volume contracted versus yesterday and the heavyweight indices fell. This was a track-level siphon driven by a single external event, not a broad rally.

---

2. Pre-Market List Reconciliation

This table gives both a "close change %" column and a "relative to today's open (executable price) %" column. The genuinely executable price for an A-share pre-market list is today's opening price; reconciling on close change alone systematically overstates the hit rate — today's precious-metals line is the textbook case: the closing gains look good, but almost all of them came from the call-auction gap, and buying after the open is a different matter entirely.

2.1 The 5 recommendation slots (pre-market §9①)

Hit rate (the 5 recommendation slots):

• On close change: 3 up 2 down = 60%
• On price relative to today's open (executable): 1 positive 4 negative = 20%
• The split between the two bases (60% → 20%) is itself the single most important conclusion of the day: pre-market §8-1 had already flagged that "a gap up of ≥ +6% is chase territory" and judged "today is the high-probability case"; in the event Chifeng gapped +5.34%, Shengda +6.49%, Shanjin +5.60% — the criterion itself was right, but the recommendation slots did not step back in line with it, and as a result all the return in the three precious-metals names was locked into that one call-auction jump.

2.2 "Watch only" list check

2.3 Pass list check

Pass / avoid list hit rate: 17 right / 10 wrong = 63.0% (counting the 9 entries in §2.2 "watch only" and the 18 in §2.3; 湖南黄金 (Hunan Gold, 002155) being suspended is a factual determination rather than a prediction, and excluding it gives 16 / 26 = 61.5%). The memory / CPO / optical-module chain is excluded from both numerator and denominator as "direction right but classification too coarse."

2.4 One-Sentence Self-Critique

What was called right was "order-book quality" and "single-stock falsification"; what was called wrong was "using yesterday's fund flows to negate today's catalyst" and "using fundamentals to rank short-term trades."

Three specifics:

1. The biggest failure was not picking the wrong stock, it was missing an entire theme — and the reason for missing it can be pinpointed exactly. Pre-market §1 item 16 cited a CNBC quote source and listed .SPX / .IXIC / IGV / SMH / the SOX and 30Y/10Y Treasuries one by one; same endpoint, same trading day — pulling just one more symbol, XBI or IBB, would have shown +5.90% / +6.58%; Chinese media had also already reported on the evening of 8/19 that "the mRNA cancer vaccine Phase III succeeded, a re-rating of the whole industry chain." What was scanned was "the places I expected something to happen," not "everything" — this is another recurrence of the "breadth before depth" failure already logged in this repo, except this time it happened at the cross-market sector level rather than the single-stock level.
2. "Main-force net outflow of −RMB 5.363 billion yesterday" was treated as a reason to negate today's catalyst. The entire basis for the pre-market's B− rating on innovative drugs was A-share price and volume on 8/19. But the catalyst was an external event that occurred overnight; yesterday's fund flow carries no information about it. Using T−1 pricing to negate information that only appeared on T will necessarily point the wrong way.
3. "The cleaner the fundamentals the more it should rise" was again falsified for short-term trading. Passed-over 志邦家居 (Zbom Home, 603801, H1 pre-announced loss) produced 2 consecutive limit-ups, passed-over 金健米业 (Jinjian Cereals & Oils, 600127, dragon-tiger list net buy of only +RMB 3 million) produced 4 boards, while 华帝股份 (Vatti, 002035) and 老板电器 (Robam Appliances, 002508) — highest policy exposure and lowest valuation — both fell. This is exactly the same shape as the "financial quality is not a short-term ranker" already logged in this repo: first look at which set the money went to, then rank within that set using fundamentals — rather than using fundamentals to pick the set first.

2.5 §8 Verification Points, Item by Item (18 primary criteria)

Pre-market §8 wrote every item in falsifiable form, and this section performed markedly better than the recommendation slots — the second consecutive day the same structural difference has appeared.

Net verification-point hit rate: 12 clearly correct / 18 primary criteria = 66.7% (a further 4 questionable, 2 wrong; the 8-6 auxiliary criterion was also correct but is not counted in the denominator). Of these, 8-3b and 8-10 belong to the class of "pre-set my own failure case and it actually happened" — criteria of this kind are worth more than the conclusion itself.

⚠️ But one thing must be said plainly: not one of the 18 verification points pointed at pharmaceuticals. Verification points can only test hypotheses already written into the report; they cannot discover branches that are not in the report at all. Today's lesson is therefore: verification points solve "was my judgment right," they cannot solve "did I even see it" — the latter can only be addressed by widening the scan surface.

---

3. Today's Theme Verification

3.1 Did the pre-market call the strongest theme: no

The "3 strongest branches today" given in pre-market §9② were ① gold/silver ② the housing-fund chain ③ semiconductor equipment. The actual ranking:

1. The pharmaceutical chain (rated B− pre-market, no recommendation slot) — 37/79 limit-ups, four sub-sectors in the top 7 of the gainers list
2. Precious metals (rated A pre-market, first branch) — direction called entirely correctly, ranked 8th/9th/11th on the gainers list
3. Optical fiber (not listed separately pre-market)
4. The housing-fund chain (2nd branch pre-market) — receding
5. Semiconductor equipment (3rd branch pre-market) — +0.60%, main-force only +RMB 25 million, essentially went nowhere

Put precisely: the pre-market called the second-strongest branch right, and got the strongest and the second-weakest wrong.

3.2 Unexpected theme #1: the pharmaceutical chain — catalyst chain and tiers of benefit

Primary source (release timing verified): Merck and Moderna announced on 2026-08-19 US Eastern time that in the Phase III INTerpath-001 trial, intismeran autogene (mRNA-4157/V940) in combination with KEYTRUDA in completely resected stage IIB–IV melanoma met the primary endpoint of recurrence-free survival (RFS) and the key secondary endpoint of distant metastasis-free survival (DMFS), with both statistical and clinical significance, and no new safety signals observed; 1,137 patients enrolled. This is the world's first Phase III success for an mRNA therapeutic cancer vaccine, and also the first successful individualized neoantigen therapy.

Overseas pricing (CNBC, actually measured; read moment marked as the 8/19 US Eastern close):

⚠️ A-share tiers of benefit — among today's 37 limit-ups, only a minority have an actual business link to mRNA cancer vaccines (concept ≠ beneficiary):

• Upstream raw materials (hardest link): 近岸蛋白 (Novoprotein, 688137, STAR Market ±20%) — mRNA vaccine raw-material enzymes, capacity corresponding to 5 billion doses/year, built an mRNA raw-material enzyme system for tumor cell therapy as early as 2018 and has supplied domestic mRNA platforms such as 沃森生物 (Walvax Biotechnology, 300142); 诺唯赞 (Vazyme Biotech, 688105, STAR Market ±20%) — a full enzyme solution covering template synthesis, transcription, modification and purification; 键凯科技 (JenKem Technology, 688356, STAR Market ±20%) — one of the few worldwide able to produce GMP-grade polyethylene glycol derivatives at scale, a key partner for novel mRNA LNPs.
• Platform players: 沃森生物 (Walvax Biotechnology, 300142, ChiNext ±20%) — one of the domestic mRNA platform builders, with a raw-material enzyme partnership with 近岸蛋白 (Novoprotein, 688137).
• CXO/CDMO: 博腾股份 (Porton Pharma, 300363, ChiNext ±20%), 凯莱英 (Asymchem, 002821, +6.92%, main-force +RMB 782 million), 药明康德 (WuXi AppTec, 603259, +2.38%, main-force +RMB 705 million).
• Most of the rest are track β in traditional vaccines, chemical pharma and IVD — 智飞生物 (Zhifei Biological, 300122), 康泰生物 (Kangtai Biological, 300601), 安科生物 (Anke Biotechnology, 300009), 康希诺 (CanSino Biologics, 688185) and others have no direct product correspondence with "individualized mRNA neoantigen cancer vaccines." Per the iron rule "concept ≠ beneficiary, tier it," this layer is sentiment diffusion, not logical benefit.

This stratification already showed up in the order book today — see the seal-quality split in §4.

3.3 Unexpected theme #2: optical fiber ≠ optical modules; the pre-market's classification granularity failed

The pre-market put the "memory / CPO / optical-module chain" wholesale on the do-not-participate list and enumerated 兆易创新 (GigaDevice, 603986), 中际旭创 (Innolight, 300308), 新易盛 (Eoptolink, 300502), 通富微电 (Tongfu Microelectronics, 002156), 长鑫科技 (ChangXin Memory, 688825). Those five indeed went essentially nowhere today (+0.96%, +0.61%, −0.05%, −0.69%, +0.03%), so the direction was right. But another sub-chain under the same word "optical" was very strong:

The sector-level evidence is unambiguous: the optical-fiber concept (BK1660) took main-force +RMB 3.661 billion and communication cable & accessories (BK1592) +RMB 2.032 billion, while optical communication modules (BK1136) took main-force −RMB 338 million. Same day, one sub-chain taking money in, the other paying money out. This line has been fermenting for days since the hollow-core cable / G.654.E centralized procurement landed in mid-August; today is not day 1.

Lesson: packaging "memory / CPO / optical modules / optical fiber / optical cable" into a single "chain" and avoiding it wholesale is far too coarse; the granularity of an avoid call must be pushed down to the smallest sector that fund flows can distinguish.

---

4. Limit-Up Ladder / Board Height

4.1 Height and promotion

• Highest board: 4 — 金健米业 (Jinjian Cereals & Oils, 600127, SSE main board ±10%), closed at 8.60 (+9.97%), turnover RMB 2.357 billion, turnover rate 43.41%, 9 broken seals, seal order RMB 55 million (seal order / turnover only 0.023). The pre-market judged from a dragon-tiger list net buy of +RMB 3 million that "relay money is exhausted" and expected it could not reach 4 boards — the direction was called wrong; but the order-book judgment was right: 9 broken seals, a seal ratio of 0.023 and a 43.41% turnover rate make it the least stable 4-board name on the whole market.
• 3 names on 2 boards: 速达股份 (Suda, 001277, SZSE main board) seal order RMB 223 million / turnover RMB 222 million = 1.00 (the strongest of all the 2-board names), 志邦家居 (Zbom Home, 603801, SSE main board), 汉森制药 (Hansen Pharmaceutical, 002412, SZSE main board) seal order RMB 234 million / turnover RMB 40 million = 5.82.
• 75 first boards — first boards are 94.9% of the total.
• ⚠️ Structural read: today was an "extremely wide, extremely low" board. Of the 79 limit-ups only 4 were consecutive boards, and the pharmaceutical theme topped out at just 2 boards (汉森制药 Hansen Pharmaceutical, 002412, and it is a TCM name, not part of the mRNA chain). The theme only launched today, so a ladder can only appear tomorrow at the earliest — which means both room to run and the fact that everything bought today is a first board, with not a single name having survived a second-day test.

4.2 Core names (large turnover + high recognition)

4.3 Seal-quality tiers (pharmaceutical chain) — this layer already split today

Hardest seals (seal order / turnover ≥ 1.0, mostly sealed in the 09:25 call auction with 0 broken seals):

Weakest seals (seal order / turnover < 0.1, sealed only in the afternoon or repeatedly broken):

⚠️ A consistency worth noting: the best-sealed tier (键凯科技 JenKem Technology, 688356; 近岸蛋白 Novoprotein, 688137) happens to be exactly the upstream raw-material companies with a real business link to mRNA, and the worst-sealed tier (安科生物 Anke Biotechnology, 300009; 康泰生物 Kangtai Biological, 300601; 永安药业 Yongan Pharmaceutical, 002365) is traditional vaccine / chemical pharma track-following. The money was already doing this stratification today; it was just masked by the 20% limit width.

4.4 Broken boards

• Of the 36 limit-ups on 8/19, only 14 rose today (38.9%) and 22 fell — an extremely low consecutive-board survival rate.
• Among yesterday's 3-board names: 金健米业 (Jinjian Cereals & Oils, 600127) advanced to 4 boards, 红四方 (Hongsifang, 603395) +5.66% broke its board; none of yesterday's high-position boards were able to continue.
• The housing-fund chain's 6 limit-ups from 8/19, today: 志邦家居 (Zbom Home, 603801) +10.05% (advanced to 2 boards), 万和电气 (Vanward Electric, 002543) +0.40%, 日出东方 (Sunrise East, 603366) −2.88%, 好太太 (Hotata, 603848) −6.82%, 亿田智能 (Yitian Intelligent, 300911) −7.50%, 罗普斯金 (Luopusijin, 002333) −9.94% — 1 advanced, 5 retreated.
• None of the coke chain's 4 limit-ups from 8/19 sealed today: 宝泰隆 (Baotailong New Materials, 601011) +5.26%, 大有能源 (Dayou Energy, 600403) +4.25%, 陕西黑猫 (Shaanxi Black Cat, 601015) −2.96%, 美锦能源 (Meijin Energy, 000723) −7.20%.

---

5. Fund Flows

5.1 Northbound flows: no longer disclosed, no figure given in this piece

Daily net buy turnover for the Shanghai and Shenzhen Stock Connect has not been published since 2024-08-19, and today's reading is null. This is "not disclosed," not "net inflow of zero" — any reading of it as "foreign investors are neutral" is wrong, and this piece makes no such inference. (One accompanying reminder: the "420" that some quote sources display in that field is the daily quota ceiling for Stock Connect, not the day's inflow.)

Southbound flows (available): Stock Connect (Shanghai) net sold RMB 4.792 billion, Stock Connect (Shenzhen) net sold RMB 1.016 billion, for a combined net sale of about RMB 5.807 billion.

Basis note: this reading is taken from an endpoint snapshot after the A-share close (15:30 Beijing time), while the Stock Connect trading session had not fully ended, so it is an intraday reading, not a final closing reading; the Hang Seng Index at the same moment was +0.96%, likewise an intraday reading.

5.2 Dragon-tiger list: not obtained in this piece

The exchanges publish dragon-tiger list data at around 18:00 each trading day, whereas this piece was generated at 15:30, so today's data was not yet released. This piece therefore contains no dragon-tiger seat analysis, which is a known gap — it directly affects the judgment of "is the pharmaceutical chain institutional accumulation or hot-money relay," and it has been listed in §6④ as an item to be fetched tomorrow morning.

5.3 Main-force net inflow — sectors

Top net inflows (industries):

On a concept basis: innovative drugs +RMB 6.261 billion, pathogen control +RMB 5.173 billion, pharma/medical style +RMB 4.489 billion, optical-fiber concept +RMB 3.661 billion, CRO +RMB 3.334 billion, antiviral +RMB 3.065 billion, CAR-T cell therapy +RMB 2.841 billion, AI pharma +RMB 2.503 billion, biological vaccines +RMB 2.171 billion, gold concept +RMB 5.919 billion.

Top net outflows (industries): semiconductors −RMB 7.481 billion (8/19 was −RMB 36.358 billion, outflow scale narrowed 79.4%), semiconductor materials −RMB 2.585 billion, batteries −RMB 1.632 billion, components −RMB 1.545 billion, minor metals −RMB 1.483 billion, IC manufacturing −RMB 1.291 billion, consumer electronics −RMB 1.006 billion, wind turbines −RMB 914 million, aerospace equipment −RMB 854 million, fluorochemicals −RMB 829 million.

⚠️ One basis issue that must be called out: today the "semiconductor equipment" sector showed +0.60% with main-force +RMB 25 million, which looks like it has stabilized. But over the same period "semiconductor materials −RMB 2.585 billion," "IC manufacturing −RMB 1.291 billion" and "components −RMB 1.545 billion" were still seeing large outflows. Using "semiconductor equipment turned positive" to argue that the semiconductor chain is repairing does not hold; that sector has only 25 constituents and is the smallest by size.

5.4 Main-force net inflow — individual stocks

Top 10: 紫金矿业 (Zijin Mining, 601899) +RMB 2.081 billion, 通鼎互联 (Tongding Interconnection, 002491) +RMB 1.335 billion, 太辰光 (T&S Communications, 300570) +RMB 1.088 billion, 凯莱英 (Asymchem, 002821) +RMB 782 million, 药明康德 (WuXi AppTec, 603259) +RMB 705 million, 仕佳光子 (Shijia Photons, 688313) +RMB 698 million, 普冉股份 (PUYA Semiconductor, 688766) +RMB 655 million, 亨通光电 (Hengtong Optic-Electric, 600487) +RMB 652 million, 盛达资源 (Shengda Resources, 000603) +RMB 605 million, 兴业银锡 (Xingye Silver & Tin, 000426) +RMB 574 million.

Top 6 net outflows: 中际旭创 (Innolight, 300308) −RMB 1.563 billion, 风华高科 (Fenghua Advanced Technology, 000636) −RMB 1.272 billion, C 宇树-W (C-Unitree-W, 688836) −RMB 1.096 billion, 海光信息 (Hygon Information Technology, 688041) −RMB 911 million, 金风科技 (Goldwind, 002202) −RMB 887 million, 长鑫科技 (ChangXin Memory, 688825) −RMB 849 million.

⚠️ An important correction to the pre-market: the real destination of money in the precious-metals chain today was 紫金矿业 (Zijin Mining, 601899), a single name at +RMB 2.081 billion, more than the entire gold sector's (BK1617) net inflow of +RMB 1.977 billion. The pre-market list's precious-metals branch lined up 7 names (Chifeng, Shengda, Shanjin, China National Gold, Xingye, Shandong, Western Region), and 紫金矿业 (Zijin Mining, 601899) alone was missing — because the pre-market screened for purity using "mined gold as a share of revenue," and Zijin is copper-and-gold balanced, so it failed the "structurally pure" standard. This has the same shape as the "blind spot of narrative-first screening" already logged in this repo: the screening standard is itself a silent discard mechanism, and a discarded name never appears in any table, so it can never be caught by self-review either.

---

6. Tomorrow's Outlook

① Judgment on theme continuation tomorrow

Pharmaceutical chain — high probability of continuation, but its character will shift from "everything up together" to "stratified." Three supports: the catalyst is a Phase III positive clinical result verifiable at primary source (Merck's website / company announcement), not a rumor; it is the world's first, so there is no "old news recycled" problem; and sector main-force net inflow totals over RMB 15 billion while not a single name reached a second board today, so the money has no realization pressure yet. But stratification signals already appeared today — upstream raw materials (键凯科技 JenKem Technology, 688356, seal ratio 14.75; 近岸蛋白 Novoprotein, 688137, 1.34, both sealed in the 09:25 call auction with 0 broken seals) were markedly stronger than the track-followers (安科生物 Anke Biotechnology, 300009, sealed only at 13:00 with 3 broken seals; 永安药业 Yongan Pharmaceutical, 002365, 5 broken seals). The risk tomorrow is not that the theme disappears, it is that "of the 37 that hit the limit yesterday, only a minority will still be standing tomorrow."

Precious metals — moving into disagreement. Today the sector was 12 up 0 down with main-force +RMB 3.38 billion, so the direction is fine; but two pieces of counter-evidence: ① the commodity leg gave back ground during the A-share session — COMEX gold fell from 4,579.50 (+0.75%) at 07:01 Beijing time on 8/20 to 4,542.50 (−0.06%) at 15:26, an intraday retreat of about 0.81%; COMEX silver 66.73 (+1.37%), still stronger than gold; ② the A-share gap has been completed in one go — the intraday (open→close) moves of Chifeng, Shengda and Shanjin were only +2.44%, −0.12% and −1.41% respectively, so almost all the return is locked into that one call-auction jump, and that gap will not repeat tomorrow.

Housing-fund chain / coke chain — ebb confirmed, continuation no longer discussed.

② Key focus for tomorrow (branches + stocks + falsifiable verification points, all marked with sector and read moment)

The drafting follows today's lesson: every item is of the form "is X > threshold," and each also sets an absolute-direction gate (to avoid repeating §8-6's "outperformed but all fell").

Branch A: upstream raw materials of the pharmaceutical chain (the hardest layer of the link)

1. [Does the upstream-vs-follower stratification hold] Closing read: do 近岸蛋白 (Novoprotein, 688137, STAR Market ±20%) and 键凯科技 (JenKem Technology, 688356, STAR Market ±20%) both close up, and both outperform 安科生物 (Anke Biotechnology, 300009, ChiNext ±20%) and 康泰生物 (Kangtai Biological, 300601, ChiNext ±20%)?

• Absolute gate: both must post a positive change; merely "outperforming" does not count (today's §8-6 failure mode).
• Yes → money is pricing by mRNA business link, stratification holds, the theme can continue.
• No → it is pure track β; treat it as a sentiment line and shorten the window.

2. [Does the theme have a ladder] Close: among today's 37 pharmaceutical limit-ups, is the limit-up count ≥ 8 tomorrow (i.e. the 2-board count)?

• ≥ 8 → a ladder has formed, the theme enters its fermentation phase.
• 0–2 → today was a one-day event-driven siphon; treat subsequent action as an ebb.

3. [Can the core name take the baton] Close: is 石药创新 (CSPC Innovation Pharmaceutical, 300765, ChiNext ±20%) seal-order / turnover ratio > 0.3? (today only 0.073, sealed at 09:52, 1 broken seal)

• No → the largest-turnover name cannot hold its seal, and the ladder lacks a core.

4. [Is sector money continuous] Close: is biologics (BK1044) main-force net positive for a 2nd straight day? (today +RMB 3.234 billion)

Branch B: optical fiber / optical cable / optical components (the second line the pre-market missed today)

5. [Does the sub-chain split continue] Close: is the optical-fiber concept (BK1660) main-force net > 0 while optical communication modules (BK1136) remains negative? (today +RMB 3.661 billion and −RMB 338 million respectively)

• Both turn positive together → it is a broad rally across the whole optical-communication chain and the classification can be re-merged.
• Split persists → tomorrow the sub-chains must continue to be treated separately.

6. [Leader absorption] Close: does 太辰光 (T&S Communications, 300570, ChiNext ±20%) close up with main-force net > 0? (today +18.38%, +RMB 1.088 billion, never sealed — a large gain without a seal is the easiest to give back the next day)

Branch C: precious metals (downgraded to observation, not placed first in the watch sequence)

7. [Can the commodity leg stop falling] Next-day 09:00 Beijing read: is COMEX gold (@GC.1) > 4,542.50? (today's 15:26 read)

• No → A-share gold names lack any new driver, and today's gap was the entire room for this round.

8. [Is only leader inertia left] Close: does 赤峰黄金 (Chifeng Gold, 600988, SSE main board ±10%) beat 山东黄金 (Shandong Gold, 600547, SSE main board ±10%)?

• Today 山东黄金 (Shandong Gold, 600547) +8.59% beat Chifeng's +7.91%, so the pre-market's "pick the structurally cleanest" logic has been rejected once; if the leader outperforms again tomorrow, that stock-selection logic will no longer be used in this round.

9. [SGE catch-up] Next-day read: is the SGE Au99.99 closing price > RMB 945.22/gram? (the 8/19 close; the 8/20 close had not yet been updated when this piece was generated)

Risk-type verification points

10. [Survival rate of today's limit-ups] Close: among today's 79 limit-ups, is the number rising tomorrow ≥ 40 (50%)? (today's corresponding reading: of the 36 from 8/19, only 14 rose = 38.9%)

• < 30% → the money-making effect has not genuinely repaired, and all relay strategies fail.

11. [High board] Close: does 金健米业 (Jinjian Cereals & Oils, 600127, SSE main board ±10%) reach 5 boards? (4 boards today, but 9 broken seals, seal ratio 0.023, turnover rate 43.41% — the least stable high board on the whole market)
12. [Style] Close: does the CSI 500 beat the SSE 50 for a 2nd straight day? (today's gap 1.35pp)
13. [Has semiconductor bleeding really stopped] Close: is semiconductors (BK1036) main-force net > −RMB 4 billion? (today −RMB 7.481 billion, 8/19 −RMB 36.358 billion)

③ Avoid tomorrow

1. The entire housing-fund chain — kitchen appliances −1.93%, real-estate services −0.19%, its 6 limit-ups showing 1 advance and 5 retreats, leaving only 志邦家居 (Zbom Home, 603801) as a lone 2-board survivor, and its H1 pre-announced loss is RMB 126–189 million. Lone survivor + pre-announced loss + sector ebb is the most textbook avoid pattern.
2. All coke / coal — coke −3.20%, 1 up 6 down, main-force −RMB 154 million, limit-up count from 4 to zero; all four names are still loss-making.
3. 金健米业 (Jinjian Cereals & Oils, 600127, SSE main board ±10%) — 4 boards but with a 43.41% turnover rate, 9 broken seals, and seal order / turnover of 0.023. Judging that it could not reach 4 boards was wrong pre-market, but every order-book basis for judging "relay money exhausted" was borne out today — borne out in quality rather than in outcome.
4. Pharmaceutical names that only sealed in the afternoon / repeatedly broke their seal today — 安科生物 (Anke Biotechnology, 300009, sealed 13:00, 3 broken seals, seal ratio 0.031), 永安药业 (Yongan Pharmaceutical, 002365, 5 broken seals, 0.024), 康泰生物 (Kangtai Biological, 300601, sealed 10:00, 2 broken seals, 0.054), 西藏药业 (Tibet Rhodiola Pharmaceutical, 600211, sealed 11:29, 0.071). In a track with 37 limit-ups, the tail-end sealers are the first batch to break the next day.
5. 通鼎互联 (Tongding Interconnection, 002491, SZSE main board ±10%) — despite main-force +RMB 1.335 billion, 13 broken seals is the most of any name market-wide and its turnover rate was 19.10%, making the sealing process extremely unstable.
6. Large gainers that never sealed — 太辰光 (T&S Communications, 300570) +18.38%, 仕佳光子 (Shijia Photons, 688313) +8.32%, 山东黄金 (Shandong Gold, 600547) +8.59%: the gains are large enough but there is no sealed-board absorption, so the probability of giving it back the next day is higher than for sealed names.
7. The robotics chain — day 2 of the ebb, 埃斯顿 (Estun Automation, 002747) −9.07%.
8. 中际旭创 (Innolight, 300308), 长鑫科技 (ChangXin Memory, 688825), 海光信息 (Hygon Information Technology, 688041), 风华高科 (Fenghua Advanced Technology, 000636) — top of the market-wide main-force net outflow list at −RMB 1.563 / 0.849 / 0.911 / 1.272 billion respectively.

④ Input hints for tomorrow morning's pre-market list

1. [Highest priority] The scan surface must be completed before depth is discussed. The direct reason the strongest theme was missed today: the pre-market pulled .SPX/.IXIC/IGV/SMH/the SOX from the same quote endpoint but did not pull XBI/IBB. Tomorrow morning, please turn the overnight US sector scan into a fixed checklist and read every item, covering at minimum: XBI, IBB (biopharma), SMH (semiconductors), IGV (software), XLE (energy), XLU (utilities), XLV (healthcare), GDX (gold miners), KWEB/FXI (China ADRs) — read all sectors first, then decide which one to dig into.
2. Do not use T−1 A-share fund flows to negate an overnight catalyst that only appeared on T. Today "innovative drugs, 8/19 main-force −RMB 5.363 billion" was used as the core basis for a B− rating, while the catalyst occurred after the 8/19 US session. The chronology of those two data points is inverted.
3. Verification points must add an "absolute-direction gate." §8-6 held today on "outperformed" while all four names fell, voiding the conclusion. Any criterion of the form A outperforms B must add "and A's change > 0."
4. Verification points must cover "branches this piece did not select." None of today's 18 verification points pointed at pharmaceuticals, so no amount of accurate verification could have found the missed theme. Suggestion: permanently add one item — "do the three industry sectors with the most limit-ups today appear in the branch list in §2 of this piece?" — a criterion capable of discovering "what I did not see."
5. The granularity of avoid calls must be pushed down to the smallest sector. "Avoid the memory / CPO / optical-module chain wholesale" today caught the optical-fiber concept (main-force +RMB 3.661 billion) in the crossfire; tomorrow morning please list optical modules (BK1136) and optical fiber (BK1660) separately.
6. The screening standard itself needs a back-sweep. The precious-metals branch lined up 7 names yet missed 紫金矿业 (Zijin Mining, 601899, +RMB 2.081 billion), the No. 1 main-force net inflow market-wide that day, because the "mined gold share" purity standard silently discarded it. Suggestion: once a branch is finalized, check back whether the top 3 main-force net inflows in that branch's sector that day are all in the table; for any that are not, state why they were excluded.
7. Tomorrow, today's dragon-tiger list must be fetched (published after 18:00), focusing on the seat structure of the pharmaceutical-chain limit-ups: institutional entry or hot-money relay, which determines whether tomorrow should be treated as a "new theme" or a "one-day wonder."
8. Already delivered today, no need to repeat tomorrow: the judgment that 恒瑞医药 (Hengrui Pharma, 600276) earnings have stalled was fully confirmed by −6.00%; 赛意信息 (SIE Consulting, 300687) gapping up then fading is confirmed; 宇树科技 (Unitree Robotics, 688836) at −18.70% is confirmed. These three do not need to be re-argued tomorrow morning.

---

⚠️ Risk disclaimer: this recap is post-close information organization and observation only and does not constitute investment advice. Data may differ in timeliness or basis; please defer to exchange and company disclosures, and do not use this directly as a trading basis.

Data-Fetch and Run Log (not sent to clients)

1. Data-fetch failures

• push2.eastmoney.com / push2his.eastmoney.com are still unavailable across the board (same as the 8/20 pre-market): akshare's stock_zh_index_spot_em and stock_zh_index_daily_em reported RemoteDisconnected on 4 consecutive retries. All quote / sector / fund-flow data was switched to a direct REST connection to push2delay.eastmoney.com (200 throughout).
• push2delay's K-line endpoint /api/qt/stock/kline/get returns klines: [] for sectors (secid=90.BKxxxx), and could not fetch individual stocks either; therefore the 8/19 historical sector change baseline could not be obtained, and all 8/19 comparison values in §3 were switched to readings already recorded in the 8/20 pre-market report (innovative drugs −2.99% / −RMB 5.363 billion, coke main-force +RMB 681 million, semiconductors −RMB 36.358 billion, gold BK1617 +RMB 432 million, precious-metals sector −0.89%), and are used in the body as "as recorded pre-market." Index historical volume was switched to Tencent's web.ifzq.gtimg.cn (successful).
• CNBC quotes: the restQuote path returns non-JSON and requires the FormattedQuoteResult.FormattedQuote structure inside the response body; the multi-symbol separator is a bare | (URL-encoded %7C also works, but a User-Agent is required). Re-verified that the previous_day_closing field returns 174.38 for MRNA = last, a known dirty field; the actual previous close is back-calculated as last − change = 174.38 − 111.42 = 62.96.
• Eastmoney dragon-tiger list RPT_DAILYBILLBOARD_DETAILSNEW: passing sortColumns=NET_BUY_AMT returns success:false (even with a valid date); with the sort parameter removed, 8/19 returned 83 records normally. With sorting removed, 8/20 still returned "data is empty," confirming this is the 18:00 publication-timing issue, not an endpoint failure.
• The regularMarketPrice regex scrape of stockanalysis.com returned empty (the page structure has changed), so cross-verification of MRNA/MRK was switched to CNBC as a single source plus cross-symbol consistency (MRK +12.60%, BNTX +21.96%, XBI +5.90%, IBB +6.58% corroborate each other on direction and magnitude).

2. Full-sample basis verification

⚠️ On avoiding stock_zt_pool_previous_em: this repo has already logged the trap that this endpoint "returns the same day, not the next day." This piece did not use it at all; instead it took the 36 codes from the 8/19 official limit-up pool directly and computed the survival rate from today's ulist quotes itself (14 up / 22 down), sidestepping the trap at the source.

3. Freshness verification

• A-share quote read moment: 2026-08-20 15:30 (after the close); both indices and individual stocks are on a closing basis (the SSE Composite at 3903.72 matches the kline close).
• The two COMEX read moments are marked: @GC.1 was 4,579.50 at 07:01 Beijing time on 8/20 (+0.75%, the pre-market report's reading), and 4,542.50 at 15:26 Beijing time (03:26 EDT US Eastern) (−0.06%). The 0.81% intraday give-back is the key argument in this piece's outlook for tomorrow, and both read moments are marked in the body.
• US data last_time is 2026-08-19 throughout, confirming an 8/19 closing basis and not stale data. MRNA's yrhidate = 08/19/26 is consistent with last, cross-consistent.
• The latest SGE Au99.99 entry is still 2026-08-19 (945.22); the 8/20 close had not been updated at this piece's generation moment, and the body marks it "not obtained" rather than substituting 8/19 (guarding against "silent one-day lag").
• Southbound flows / Hang Seng Index are on an intraday basis (the Stock Connect had not closed when data was taken at 15:30 A-share time); this is stated explicitly in the body, guarding against "cross-market data with no read moment marked."

4. One factual correction to the 8/20 pre-market report

The pre-market wrote "崧盛股份 (Sosen Electronics, 300899)" in several places; the code is wrong. Checked against the Eastmoney code table: 崧盛股份 (Sosen Electronics) = 301002; 300899 = 上海凯鑫 (Shanghai Kaixin) (+3.74% today). Pre-market §1 item 13, §6 and §9③ all cited the wrong code. This entry did not enter the scoring of the recommendation slots or the Pass table, so the impact is limited to citation accuracy, but it violates the substantive requirement of the "every stock must carry its code" iron rule (a code was carried, but it points to a different company). Takeaway: a code is not decoration on a name; before writing it, run one lookup against the code table — a citation with the right name and the wrong code is more dangerous than no code at all, because it looks as if it has already been verified.

5. Not done / not obtained in this piece

• Dragon-tiger list (8/20) not obtained — published after 18:00, this piece generated at 15:30. It is therefore impossible to judge whether the pharmaceutical-chain limit-ups are institutional accumulation or hot-money relay, which is this piece's largest blind spot for the "continuation tomorrow" judgment, stated explicitly in §5.2 and §6④ item 7 of the body.
• No northbound data — disclosure stopped from 2024-08-19; the body explains that "not disclosed ≠ net inflow of zero."
• The 8/20 SGE Au99.99 close was not obtained, so §6 verification point 9 uses 8/19's 945.22 as the threshold baseline.
• No stock-by-stock announcement back-sweep was done on today's 79 limit-ups — this piece's attribution of the pharmaceutical chain rests on three layers of evidence (sector level + overseas catalyst primary source + business link), but it does not rule out that some of these individual names had their own announcement drivers (lesson 0 of the 8/20 pre-market report is precisely that "any attribution must include a single-stock announcement back-sweep"). A stock-by-stock back-sweep of 37 names exceeds this piece's time window and is explicitly marked as not done; sector-level attribution should not be treated as attribution for every single name.
• medRxiv/NEJM-level detailed data on INTerpath-001 not obtained — only the Merck/Moderna topline statement is public (RFS and DMFS endpoints met), with the specific HR, confidence intervals and absolute benefit magnitude undisclosed; foreign media have already pointed this out under the headline "Landmark Result, Incomplete Data." This piece therefore only states that "the primary endpoint was met," makes no assessment of the magnitude of efficacy, and does not extrapolate revenue elasticity for domestic companies from it.
• Among today's 37 pharmaceutical limit-ups, mRNA business exposure was not verified name by name apart from 近岸蛋白 (Novoprotein) / 诺唯赞 (Vazyme Biotech) / 键凯科技 (JenKem Technology) / 沃森生物 (Walvax Biotechnology) / 博腾股份 (Porton Pharma); the body labels the rest uniformly as "track β, not a logical beneficiary," and no benefit quantification is given for any of them.

6. Methodology takeaways from this piece (for direct use in tomorrow's pre-market)

1. Verification points solve "was my judgment right," they cannot solve "did I even see it." Today 12 of 18 verification points were correct (66.7%), yet the recommendation slots were only 20% (on the relative-to-open basis) — a structural difference appearing for the second consecutive day. But none of the 18 pointed at pharmaceuticals. A new class of criterion is needed: "do the three sectors with the most limit-ups that day all appear in this piece's branch table" — it tests coverage rather than judgment.
2. A relative criterion must be paired with an absolute gate. §8-6's "Vatti & Robam both outperform Yitian & Sunrise East" still held with all four falling, and produced the inverted conclusion that "the policy chain can continue." Any A outperforms B must add "and A > 0."
3. A screening standard is a silent discard mechanism. 紫金矿业 (Zijin Mining) (+RMB 2.081 billion, No. 1 main-force net inflow market-wide) never entered any table because it failed the "high mined-gold share" purity standard, and therefore could never have been caught in any round of self-review either. This has the same shape as the already-logged "blind spot of narrative-first screening." Fix: once a branch is finalized, check back the top 3 main-force net inflows in that sector that day; for any not in the table, the reason for exclusion must be stated.
4. The granularity of a "chain" must be pushed down to the smallest unit that fund flows can distinguish. Optical fiber (+RMB 3.661 billion) and optical modules (−RMB 338 million) moved in opposite directions on the same day, yet were packaged into one "chain" and avoided wholesale.
5. T−1 fund flows cannot negate a T-day catalyst. This is the second reason the theme was missed today (the first being that it was never scanned).