A-Share Market Recap | 2026-08-13 Thursday

Data basis: A-share quotes / turnover / turnover rate / net main-force flow / sector and concept fund flows are all on the East Money 2026-08-13 close basis; the limit-up / limit-up-break / limit-down pools are the corresponding East Money pools (ST included, new listings excluded). Limit thresholds by board: SSE/SZSE main board ±10%, ChiNext / STAR Market ±20%, BSE ±30%.

---

0. Recap in One Line

Sentiment temperature: gap up, then ebb. All three major indices gapped up (ChiNext Index opened +1.52%) then fell all day and all closed lower: SSE Composite 3926.96 −0.50%, SZSE Component 14289.44 −0.87%, ChiNext Index 3586.04 −0.45%, STAR 50 1717.75 −1.11%, BSE 50 1097.80 −1.62%. Combined turnover of the two exchanges was RMB 2,550.3 billion, but advancers/decliners came in at 1,142 up / 4,222 down / 81 flat — only 21% of stocks rose, a textbook "index down a little, individual stocks down a lot" gap-up-and-fade day.

Today's strongest theme was not optical communications, the pre-market's top call, but pharmaceuticals — which the pre-market had explicitly listed as "not recommended to chase." Pharma & biotech +1.51%, main-force +RMB 4.524 billion (first across all industries); CRO +4.49%, pharma R&D outsourcing +5.39%, medical services +3.90%, diagnostic services +5.66%. Meanwhile the pre-market's first branch, optical communications, turned down across the board: optical modules −1.45%, optical fibre concept −1.35%, PCB −1.84%, memory chips −1.94%, semiconductors −1.96%.

Pre-market list scorecard: direction right, ranking wrong, execution more wrong still. All 5 recommended names gapped up and all closed below their open; on the executable "buy at the open → close" basis the average was −3.95%, the worst of the four groups; the 15 names given a Pass averaged −0.85% with 6 of them closing higher — the Pass group beat the recommendation group. The three risk warnings the pre-market wrote for itself (gap up then fade; leaders stalling while second-tier names hit limit-up means divergence; late-stage diffusion) all came true, but none of them were written into the recommended actions.

Tomorrow's tone: rebalancing after divergence. The highest consecutive-limit-up count fell from yesterday's 7 boards to 5 boards, the limit-up break rate was 37.9%, and sentiment has turned from euphoric to ebbing; intraday, money completed a defensive reallocation out of "electronics / non-ferrous" and into "pharma / power / banks." Tomorrow is not a day to chase the main theme — it is a day to wait for the divergence to resolve.

---

1. Market Overview

Turnover has been cross-checked: summing every stock across the whole market gives RMB 2,550.3 billion, consistent with the sum of the two exchanges' index turnover (1,164.2+1,386.7 = RMB 2,550.9 billion); the difference comes from a handful of suspended / delisting-arrangement stocks.

Advance-decline and sentiment indicators

Sentiment call: the turning day from euphoria to ebb. Three things hold simultaneously: ① the highest board fell from 7 to 5, and yesterday's market-wide highest name 百花医药 Baihua Pharmaceutical (600721, SSE main board ±10%) closed +3.35% without sealing the limit, breaking its 7-board streak, while 开开实业 Kaikai Industry (600272, SSE main board ±10%) fell −7.35%, breaking its 5-board streak; ② limit-up break rate 37.9%; ③ the median of yesterday's limit-up names today was −1.47%, so the follow-through money-making effect has turned negative. But this is not a collapse: only 4 limit-downs with no sector pattern, still 59 limit-ups, and the pharma ladder is intact — this is "high-flyers ebbing + money rotating," not a systemic risk release.

---

2. Pre-Market List Reconciliation

This section adds an "Open→Close %" column, the single most important basis correction in this recap. The pre-market list was published at 07:00 today, so the earliest executable moment for a reader was today's open. Looking only at "closing change %" systematically overstates the hit rate — of today's 45 tracked names, 41 gapped up and 33 of them (80%) closed below their open, with a whole-sample open→close median of −3.35%. "Still green on the close" and "profitable if you executed the list" were two completely different things today.

2.1 The 5 Recommended Names (pre-market §9.1)

Recommendation list totals: on the close basis 1 up / 4 down, average −1.14%; on the open→close basis 0 of 5 positive, average −3.95%, median −4.06%. All 5 gapped up and all 5 closed below their open.

2.2 Watch / Watch-Only / Pass Group Reconciliation (summary)

One thing that must be spelled out: mean and median give opposite conclusions, and both have to be written down.

• On the median, the pre-market ranking worked: recommendation list median −1.04% beat watch −1.87%, watch only −1.35%, Pass −1.98%. At the "typical stock" level, the grading was right.
• On the mean, the ranking is completely inverted: the Pass group at +0.46% was the best and the recommendation list at −1.14% the worst. The reason is that the Pass group hid the day's two biggest tails — 万邦医药 Wanbang Pharmaceutical (301520, ChiNext ±20%, +20.00%), the top gainer of the whole sample, and 中鼎股份 Zhongding Sealing Parts (000887, SZSE main board ±10%, +9.98%, limit-up).
• Conclusion: the pre-market list earned the "median" but excluded all of the day's tail returns — and on a day when 79% of stocks fell, the returns were exactly in the tail. This is not a ranking-precision problem; it is a problem with the distribution shape of the list.

2.3 Name-by-Name Reconciliation

Called wrong (bullish pre-market, weak today)

Called right (bearish / Pass pre-market, duly weak today)

Passes called in the wrong direction (pre-market said don't touch, big gains today) — the category most worth recording this round

Listed pre-market as "watch only / watch" yet strongest today

2.4 Hit Rate and Post-Mortem

Hit rate (all three bases given; any one alone is misleading)

Post-mortem in one line: the pre-market got the "risks" right but never wrote them into the "actions."

Pre-market §8.4 said verbatim "a gap up followed by a fade is the pattern most in need of defending against today," §8.2 said "if today repeats the pattern of leaders +3% and second-tier names at 20cm, judge it as divergence rather than continuation," and §9.4 said "a gap up is highly likely today, but this is a divergence day, not an acceleration day." All three came true today. Yet the recommendation list still put forward 5 names whose implied action was "buy today," all concentrated in the optical communications / AI hardware chain that had already risen for two straight days. The disconnect between judgement and action is the primary failure of this round — not a lack of information.

Three specific errors, ranked by severity:

① Most serious: using "single-day main-force flows" to exclude a branch, when single-day flow extremes are a contrarian indicator. Item 7 of pre-market §9.3 passed on the entire pharma chain on the grounds that "on 8/12 pharma & biotech had a main-force net outflow of RMB 3.440 billion, the largest of any industry — the ebb is under way." Today pharma & biotech main-force flow was +RMB 4.524 billion, rebounding to first across all industries, with CRO +4.49% and pharma R&D outsourcing +5.39% — the strongest theme in the market. One day of fund flow was treated as a trend criterion. The same error produced a correct result on MLCC (风华高科 Fenghua −6.75%), which is precisely the point: it is a coin flip, not a method. Correction: single-day net flow may only be used to "confirm" a pre-existing multi-day trend, never to "exclude" a branch; the pre-market set a "two consecutive days in the same direction" gate for optical communications yet failed to apply the same yardstick to pharma.

② Ranking misalignment: the news the list itself graded S-tier belonged to a stock that never made the recommendation list. Pre-market §1 graded the 8/12 evening share-incentive draft from 天孚通信 T&S Communications (300394) as S-tier, and §5.4 spelled out in detail that "the grant price of RMB 120 is only half the closing price," "the 2029 net profit hurdle is +700%," "the CPO-supporting FAU is already in order-delivery status," and "if today's market leadership is CPO rather than 1.6T optical modules, T&S is the most tightly linked name." That analysis was fully validated today — CPO was the only sub-segment within optical communications to hold up (−0.02%, main-force +RMB 2.633 billion), while optical modules were −1.45% and optical fibre −1.35%; T&S Communications rose +7.13% with main-force +RMB 1.052 billion, the strongest in the whole sample. But it was ranked "watch closely" and left off the recommendation list, on the grounds of "the only one of the three with clearly decelerating Q2 single-quarter growth + PE/PB percentiles both at 96.6%." This is using "financial quality" as a short-term ranker. In a single-day move driven by a company event (a half-price share incentive), single-quarter growth and valuation percentiles are not the day's pricing variables. Correction: when a name simultaneously has an "S-tier new company-level event + membership in the day's leading sub-segment," its same-day ranking weight should outrank valuation percentile and single-quarter growth; the latter two determine holding-period risk, not same-day direction.

③ Basis error: the list is benchmarked on the closing price, while readers execute at the open. The 5 recommended names gapped up an average of +2.4% and closed down an average of −1.14%, an open→close of −3.95%. If tomorrow's list does not flag the gap between "pre-market price vs executable price," the hit rate will be systematically overstated by roughly 3 percentage points. Correction: from tomorrow, every name on the pre-market list must carry a quantified gate of the form "if it gaps up more than X%, today is not a buy point"; this recap's reconciliation tables now permanently add the "Open→Close %" column.

One thing done right and worth keeping: the Pass on memory modules and MLCC. The pre-market rejected the four memory-module names on the grounds of an "unresolved conflict between overseas positives and domestic money flows," and today all four fell (江波龙 Longsys −4.18%, 佰维 Biwin −4.03%, 德明利 Demingli −2.71%, 深科技 Shenzhen Kaifa −1.98%), with the memory chip concept −1.94% and main-force −RMB 10.331 billion. The discipline of "do not participate until the conflict is resolved" is the highest-quality part of this list and should be kept.

---

3. Verification of Today's Leading Themes

3.1 Did the pre-market call the strongest theme? No — and it called it backwards

The pre-market put optical communications as the S-tier first branch and put pharma into item 7 of "not recommended to chase today." Today's result: the optical communications branch turned down across the board and pharma was the strongest in the market. This is not a precision problem; it is a direction problem.

But one qualifier must be written alongside it: the pre-market's read on the "internal structure" of optical communications was right. The pre-market repeatedly stressed three points — ① the move had already spread to 20cm catch-up names, i.e. mid-to-late stage; ② leaders stalling while second-tier names hit limit-up means divergence; ③ "if the main theme is CPO rather than 1.6T optical modules, 天孚通信 T&S Communications is the most tightly linked name." All three landed today: leaders at 0% gain, second-tier 星网锐捷 Star-Net Ruijie and 共进股份 Gongjin Electronics still limit-up, CPO the only sub-segment holding up, and T&S Communications leading at +7.13%. The quality of the pre-market's "structural understanding" of this chain was high; its judgement on the chain's "direction and position implications" was poor.

3.2 Unexpected Theme One: Why Pharma Rose

This must be stated honestly: this recap could not identify a single new catalyst for today's pharma rally. The verified facts are —

• Pharma/CRO did not start today. On 2026-08-07 there was already a round of "a wave of limit-ups across the whole CRO sector" (21st Century Business Herald¹), and on 8/08 there were reports of a "pharma + technology dual main-theme" market. Today is an already roughly week-old theme regaining strength after a one-day pullback, not day one of a new theme.
• Industry fundamentals background (not today's news; must not be treated as today's driver): WuXi AppTec's 2026H1 net profit attributable to parent was RMB 11.08 billion, +29.43% year on year, with a backlog of RMB 66.43 billion, +25.2% year on year, and full-year guidance raised to RMB 58.5–60.5 billion — that interim report was released on the evening of 2026-08-03, ten days ago (Sina Finance²).
• The real problem with the pre-market line "the pharma ebb is under way" is therefore this: it used a single day's RMB 3.440 billion net outflow on 8/12 to negate a theme that had already been running for about a week. This is the same error as item ① in section 2.4.

A risk fact that must be added about 万邦医药 Wanbang Pharmaceutical (301520): the company issued a clarification announcement on 2026-07-16 explicitly stating that media and online reports about it advancing a core-pipeline innovative drug (huperzine A controlled-release tablets) and signing cooperation agreements for new drugs in ALS and other diseases were false information, and that the market had confused it with another company with a similar name; it also issued an abnormal share-trading fluctuation announcement on 2026-08-11. Today the stock rose +20.00% to limit-up on a 49.28% turnover rate with 5 limit-up breaks, and net main-force flow of only +RMB 26 million — enormous churn accompanied by almost zero net inflow, a purely sentiment-driven speculative float. The pre-market Pass still holds on risk grounds, and today's limit-up does not change that.

3.3 The Three Verification Points Set Pre-Market — All Answers Now In

Pre-market §8 set three falsifiable observation points; all produced results today, and this is the highest-quality design in the list:

3.4 The Answer to the "Semiconductor Verdict Day"

Pre-market §8.2 said: "Today is a verdict day for semiconductors: the +RMB 5.450 billion on 8/12 is the first sample; if there is net inflow again today that constitutes two consecutive days in the same direction and it can be formally ranked; if there is outflow, then 8/12 was a single-day pulse."

Answer: today semiconductors had a main-force net outflow of RMB 9.800 billion with the sector −1.96%. The +RMB 5.450 billion on 8/12 is confirmed as a single-day pulse. Under the pre-market's own rule, the semiconductor equipment direction must not enter tomorrow's ranking. 中微公司 AMEC (688012) −2.76%, 北方华创 Naura Technology (002371) −2.82%, semiconductor equipment sector −2.20% (main-force −RMB 1.990 billion). This gate was correctly designed, took effect today, and should continue to be used.

3.5 Unexpected Theme Two: Power — Plus a Failed Verification That Must Be Stated

Power is a direction the pre-market did not cover at all today, yet both money and limit-ups confirmed it: thermal power +1.56%, main-force +RMB 3.144 billion, main-force share 10.55% (among the highest of any industry); power +RMB 3.610 billion, utilities +RMB 3.539 billion; 大唐发电 Datang International Power (601991, SSE main board ±10%) hit limit-up on turnover of RMB 5.97 billion, first in today's limit-up pool by turnover, with a limit queue of RMB 377 million and zero limit-up breaks — the best tier of seal quality in the limit-up pool; 华电能源 Huadian Energy (600726, SSE main board ±10%) also hit limit-up.

But this recap could not obtain reliable evidence for the specific catalyst behind today's power rally, so no inference is made. The reports found — "Datang International Power at 6 consecutive limit-ups," "power sector surges, 10 stocks limit-up" — were all published in May 2026, are historical market moves, and are unrelated to today: Datang International Power's consecutive-limit-up count today is 1 (first board), it is not a multi-board name. Using May reports as today's attribution would constitute a factual error, so they are not used. At the industry level there is a longer-term backdrop of "computing-power/electricity coordination" policy and peak summer electricity demand, but these are not new information today and cannot explain today's single-day move.

---

4. Limit-Up Ladder / Board Height

Ladder structure (59 limit-ups)

Core names (large turnover + rising + recognizable)

紫光股份 Unisplendour and 浪潮信息 Inspur are the two names most worth remembering today: turnover of RMB 16.46 billion and RMB 11.95 billion respectively, among the two largest money flows in the whole market, both touched limit-up intraday then broke it, and still closed +7.03% and +5.92%. Their computing power / server direction saw concept main-force net inflow of RMB 4.602 billion, first among concepts, yet the sector index fell (computing power concept −0.44%) — big caps up, small caps down, the textbook early form of money crowding into a few names, though it may equally be nothing more than a same-day flight to safety. One day is not enough to characterize it.

Broken streaks (direct evidence of the step down in board height)

The name in the limit-up-break pool that says the most about sentiment: 风语筑 Fengyuzhu (603466, SSE main board ±10%) — 2 consecutive limit-ups, but 28 limit-up breaks over the day, turnover rate 24.01%, turnover RMB 1.92 billion, and a limit queue of only RMB 10 million. 28 breaks means the money sealing the limit is extremely irresolute — a textbook reading of ebbing sentiment.

---

5. Fund Flows

5.1 Northbound Flows: No Longer Disclosed; This Section Cannot Give a Net Inflow Direction

The basis has to be explained first, or you will read a wrong "0." Daily net buying via Shanghai/Shenzhen Stock Connect has not been disclosed since 2024-08-19; the "net turnover bought 0.00, net fund inflow 0.0" returned by the data interface today is a silent return value after the field stopped updating, not "northbound net inflow was zero today" — and it certainly cannot be read as "foreign investors are neutral." The "420.0" that appears in the same interface is the daily quota cap, not an inflow amount.

This report therefore offers no judgement whatsoever on the direction of northbound flows. (Southbound Stock Connect net turnover bought today was RMB 4.029 billion via Shanghai and RMB 709 million via Shenzhen, but the read was taken at 15:40 Beijing time, before the Hong Kong market closed, so these are intraday rather than closing figures — for reference only, no conclusion drawn.)

5.2 Dragon-Tiger List: Today's Data Not Yet Published; Not Obtained for This Report

The Dragon-Tiger List is published by the exchanges after 18:00 each trading day, while this recap is generated at 15:30–15:45, so the day's seat-level data is structurally unobtainable. This section is therefore left blank, and yesterday's Dragon-Tiger List is not substituted for it. Today's Dragon-Tiger data will be covered in full in tomorrow morning's pre-market list.

5.3 Main-Force Money: A Clear Style Reallocation Happened Today

Inflow side (industries)

Outflow side (industries)

How to read it: today was a wholesale switch "from high-β cyclicals (electronics + non-ferrous) to low-β defensives (pharma + power + banks)." Electronics and non-ferrous together saw net outflows of RMB 33.8 billion, while pharma, power, utilities and banks together saw net inflows of about RMB 13.6 billion. Note that the outflow scale far exceeds the inflow scale — this is not an equal-sized relocation, it is net position reduction plus partial rotation into defensives, consistent with 79% of stocks falling market-wide on turnover that did not expand.

Two concept-level items worth recording separately

• Computing power concept main-force +RMB 4.602 billion (first among concepts), liquid cooling concept +RMB 2.736 billion, East Data West Computing +RMB 3.594 billion, CPO concept +RMB 2.633 billion — yet the indices of all four concepts were flat or down. "Money in, price not up" usually means large money is absorbing selling pressure and the direction is undecided.
• Margin-trading concept main-force net outflow of RMB 42.112 billion, FTSE Russell −RMB 24.587 billion, MSCI China −RMB 23.243 billion, 2026 interim pre-announced earnings growth −RMB 24.509 billion. These are broad constituent baskets, so their large outflows show that today's selling pressure was general rather than an isolated event in any one sector. The RMB 24.5 billion net outflow from "2026 interim pre-announced earnings growth" is especially notable — stocks with pre-announced earnings growth were sold down as a group today, which is the other side of the same "defensive switch" coin.

⚠️ A methodology reminder, written for tomorrow's self: the single-day fund-flow figures above must not be used directly to rank branches. Section 2.4 of this recap has just criticized the error of "using a single day's RMB 3.440 billion net outflow on 8/12 to negate the entire pharma chain"; turning around and using today's "computing power +RMB 4.602 billion" to argue for being bullish computing power tomorrow would be the same error committed again in the other direction. The computing power / liquid cooling direction currently has only 1 trading day of fund-flow sample and, under the pre-market's own "two consecutive days in the same direction" gate, is not eligible for ranking before tomorrow.

---

6. Outlook for Tomorrow (2026-08-14 Friday)

6.1 Theme Continuation Calls

6.2 Priority Watchlist for Tomorrow

Two scheduled event times:

• Applied Materials (AMAT) reports FY26Q3 results after the US close on 8/13 (early morning Beijing time on 8/14). The pre-market had already listed this as an event risk for semiconductor equipment; today's session was "before the event" and tomorrow's session is "after the event," so the semiconductor equipment direction will have an external pricing input tomorrow.
• The interim reports of 生益科技 Shengyi Technology (600183) and 联瑞新材 Lianrui New Materials (688300) are scheduled for disclosure on August 15. (Note: 8/15 is a Saturday; the "this Friday" in the pre-market list was wrong — take the date 8/15 as authoritative.) Both were flagged pre-market as "pre-announcement range already capped at the top / no pre-announcement," and tomorrow is the last trading day before their interim reports.

6.3 What to Avoid Tomorrow

1. High-level names with broken streaks — 百花医药 Baihua Pharmaceutical (600721, SSE main board ±10%, 7-board streak broken, turnover rate 48.10% today, main-force −RMB 347 million), 开开实业 Kaikai Industry (600272, SSE main board ±10%, −7.35%). The first trading day after a broken streak is the moment with the highest probability of catch-down selling.
2. Consecutive-limit-up names with terrible seal quality — 风语筑 Fengyuzhu (603466, SSE main board ±10%, 28 limit-up breaks today, limit queue of only RMB 10 million), 一鸣食品 Yiming Food (605179, SSE main board ±10%, 9 limit-ups in 13 days, 9 limit-up breaks today). The money sealing the limit is irresolute and next-day absorption is weak.
3. The optical communications catch-up ladder — 通鼎互联 Tongding Interconnection (002491, SZSE main board ±10%, open→close −8.41% today, turnover rate 28.08%), 永鼎股份 Yongding Co. (600105, SSE main board ±10%, main-force −RMB 878 million). The ladder has broken; no second attempt.
4. The entire memory-module and semiconductor-equipment chain — 江波龙 Longsys (301308), 佰维存储 Biwin Storage (688525), 德明利 Demingli (001309), 深科技 Shenzhen Kaifa (000021), 中微公司 AMEC (688012), 北方华创 Naura Technology (002371). Semiconductors have been confirmed as a single-day pulse under the rule; the four module names have had two straight days of outflows.
5. Non-ferrous / precious metals / minor metals — the extreme outflows, with main-force shares of −10% to −14%, are not over.
6. Pure concept plays that the companies themselves have denied — 中鼎股份 Zhongding Sealing Parts (000887), 万邦医药 Wanbang Pharmaceutical (301520), 城地香江 Chengdi Xiangjiang (603887). All three rose today; that does not change their risk profile. What they share is "company announcement pointing the opposite way to the share price." This recap does not withdraw the Pass because of today's gains, but it must be acknowledged: in a sentiment-driven phase, a clarification announcement is not effective bearish news (see 2.3).
7. Orphan limit-ups with no theme membership — a large share of today's 37 first boards were scattered across decoration & renovation, beverages & dairy, real estate, gas and other directions with no industrial logic. In an environment where 79% of stocks fell, this type has the lowest next-day win rate.

6.4 Input Notes for Tomorrow Morning's Pre-Market List

1. A "gap-up gate" must be added. All 5 of today's recommended names gapped up and all closed below their open, open→close −3.95%. Every name on tomorrow's list must carry a quantified threshold of the form "if it gaps up more than X%, today is not a buy point," and the reconciliation tables must permanently retain the "Open→Close %" column.
2. Single-day fund flows must never again be used to exclude any branch. Today's reversal in pharma has proven this is a coin flip. Single-day flows may only be used to "confirm" a multi-day trend, and the "two consecutive days in the same direction" gate must be applied to all branches alike — today it correctly blocked semiconductors, yet yesterday it wrongly waved through the exclusion of pharma.
3. Same-day ranking weights need correcting: an S-tier new company-level event > valuation percentile and single-quarter growth. The lesson of 天孚通信 T&S Communications (300394): the stock behind the share-incentive news the list itself graded S-tier was kept off the recommendation list because of "Q2 deceleration + PE/PB percentiles at 96.6%," and it turned out to be the strongest in the whole sample. Valuation and growth determine holding-period risk, not same-day direction; the two must not be mixed into the same ranking.
4. The "reverse probe" design needs a precondition. The premise of the 光迅科技 Accelink (002281) probe is "the leaders are rising"; today the leaders themselves had zero gain, so the probe failed. Tomorrow it should be changed to: first judge the leaders' direction, then use the laggards to judge discrimination.
5. Complete today's two unresolved verification tasks: ① the true catalyst behind today's move in the power sector; ② whether there was any new company- or policy-level news for the pharma theme today (this report can only confirm that it is a theme that has been running for about a week, and could not locate a new driver for today).
6. The Dragon-Tiger List is published after 18:00 today, so tomorrow morning's pre-market list can fully obtain the 8/13 seat data, which should be used to verify the nature of the money behind today's pharma limit-up ladder and behind 大唐发电 Datang International Power (601991) (hot-money relay vs institutional accumulation).
7. Tonight's Applied Materials (AMAT) results are the external input for the semiconductor equipment direction tomorrow and must be covered in §1 of the pre-market list.

---

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

Data Retrieval and Run Log (not sent to clients)

1. akshare's default UA was disconnected by East Money; bypassed with a self-built fetcher stock_zh_index_spot_em and stock_board_industry_name_em threw ConnectionError(ProtocolError('Connection aborted.', RemoteDisconnected)) three times in a row. Index and sector data were switched to work/em.py (browser UA + three-host rotation over push2 / push2delay / 82.push2). The four interfaces for the limit-up / limit-up-break / limit-down / previous-day limit-up pools work fine in akshare itself and were unaffected. The suggestion from the previous pre-market run (move the fetcher out of /tmp and into the repo) has been partly implemented this time — it now lives at work/em.py, but work/ is inside .gitignore, so it is still not in the repo's tooling layer and will be lost on restart; recommend hardening it into bin/ next time.

2. There is a whole-market pagination trap that causes statistical bias; fixed When clist is paginated in descending order by fid=f3, a single direction can only retrieve about 3900 rows (total=5895). The first retrieval stopped at 3900 rows, and the advance-decline count computed from it was "1142 up / 2422 down," while the 1995 missing names were all at the tail of the decliners board — i.e. the bias is one-directional and makes the market read far stronger than it actually is. Fix: take one pass with po=0 (ascending) and one with po=1 (descending), then deduplicate and merge by code, giving 5800 rows in the end, 5445 with valid quotes, and a true count of 1142 up / 4222 down. Lesson: any interface with "sort by some field + pagination cap" will produce a systematic bias aligned with the sort field if retrieved in one direction only; you must retrieve in both directions or switch to an unsorted full-set interface. Without this fix, the whole report's sentiment call would have been written as "differentiation" instead of "ebbing."

3. Neither per-stock single-point queries nor the ulist batch interface were usable

• Looping /api/qt/stock/get over 45 names → 2-minute timeout (consistent with the previous pre-market log).
• Switched to /api/qt/ulist.np/get in batch → field misalignment in the return (f60 returned numbers on the order of turnover), so all 45 names came back as dirty data. The whole batch was discarded and never entered the report.
• Final approach: filter locally from the whole-market snapshot all.json, getting OHLC + net main-force flow for all 45 names in a single network round trip. This matches the previous run's conclusion: standardize on "pull the whole market + filter locally."

4. Data not obtained (explicitly disclosed to clients in the body)

• Dragon-Tiger List: published after 18:00, while this report is generated at 15:30, so stock_lhb_detail_em returns NoneType. This is a scheduling problem, not an interface failure, same cause as yesterday's recap. Section 5.2 of the body honestly leaves it blank.
• Northbound flows: the interface returns 0.0. Verified to be the silent return value after daily net buying stopped being disclosed on 2024-08-19, not a real 0. Section 5.1 of the body was rewritten as a basis explanation, and explicitly warns that the "420.0" in the same interface is the quota cap, not an inflow amount — this came within a hair of being written as "northbound net inflow of zero, neutral stance."
• Today's catalyst for the power sector: not obtained. The search hits "Datang International Power at 6 consecutive limit-ups" and "power sector, 10 stocks limit-up" were all reports from May 2026; Datang International Power's consecutive-limit-up count today is 1 (first board), so accepting them would have constituted a serious stale-news mismatch. Section 3.5 of the body explicitly states the verification failure and makes no inference.
• A new catalyst for pharma today: not located. Confirmed that CRO / innovative drugs is a pre-existing theme dating from 8/07, and that WuXi AppTec's interim report was released on the evening of 8/03 (not today). Section 3.2 of the body honestly labels it "could not locate a single new catalyst" and explicitly marks the stale news as stale.

5. Three near-misses on stale news / basis mismatch (same class as the previous run; worth hardening into a checklist item)

• A search summary listed WuXi AppTec's H1 (released 8/03) alongside Biocytogen's pre-announcement as "the reason for today's rally"; accepting it directly would have turned a ten-day-old earnings report into today's driver. Caught by going back and checking the release date.
• A search summary gave "Datang International Power at 6 consecutive limit-ups" (May), contradicting today's first board. Cross-refuted using our own limit-up pool data (consecutive-limit-up count = 1).
• The media basis of "2 limit-ups in 3 days" for 万邦医药 Wanbang Pharmaceutical is inconsistent with the limit-up pool's "consecutive-limit-up count = 1"; the body uses no consecutive-limit-up phrasing and states only the measured +20.00%, turnover rate 49.28%, 5 limit-up breaks, and main-force +RMB 26 million. Common thread: all three times a search summary supplied a self-consistent but stale/misaligned fact, and all three times it was caught by re-checking the release date or by local market data.

6. Things not done this time

• Did not run the fundamentals-analyst / risk-auditor sub-agents (the recap slot is primarily about market facts, and the 15:30 generation deadline is tight). The three critiques in section 2.4 and the verification-point checks in 3.3 are all self-checks by the main body, with no third-party quality review, so confidence is lower than in the pre-market slot.
• Did not verify the original text of Wanbang Pharmaceutical's 7/16 clarification announcement (only secondary confirmation via search results); the wording in the body has been softened accordingly to "issued a clarification announcement explicitly stating that … was false information."