A-Share Market Recap | 2026-08-14 Friday

0. One-Sentence Recap

Sentiment temperature: shrinking volume, changing direction. The index barely moved (SSE Composite 3927.18 +0.01%), but a clean capital migration happened inside it: the optical-communications / optical-fiber-cable chain took in over RMB 10 bn of main-force net inflow (Telecom Equipment +RMB 11.873 bn, No. 1 market-wide), while the "yesterday's consecutive limit-ups" index fell −2.23%, "yesterday's 2nd-board-and-above attempts" −4.91%, and "theme stocks" −2.22%. Combined turnover across both exchanges was RMB 2,156.7 bn, down RMB 393.6 bn from yesterday (−15.4%); the failed-limit-up rate fell from yesterday's 37.9% to 23.2%; advancers/decliners repaired from yesterday's 1142/4222 to 2400/2970.

There was no A-share pre-market list today, so this piece reconciles instead against the falsifiable judgments made for today in the "Tomorrow's Outlook" section of yesterday's recap — that is a set of explicit branch directions + named stocks + verification points, better suited to closing the loop than a missing list. The result is split: at the single-stock level the "sealing quality / announcement contradiction" criteria went 6 for 6, while at the sector level the "extrapolate from one day's fund flow" criteria went 1 for 3, the biggest miss being that today's strongest sector market-wide — optical communications — was judged "divided, possibly fading".

Tone for Monday: the main line is optical communications, but it was only lit by policy today, and no genuine high-level board has appeared yet.

---

1. Market Overview

Data basis and validation: quotes taken from the East Money market interface, snapshot timestamp 2026-08-14 15:31–15:32 (checked line by line; not stale data). Turnover has been cross-validated: summing every stock market-wide gives RMB 2,156.6 bn, consistent with Shanghai 990.37 + Shenzhen 1,152.47 + Beijing 13.85 = RMB 2,156.69 bn.

The index structure is itself today's first piece of information: SSE Composite +0.01% and SSE 50 −0.41%, versus ChiNext +1.12% and CSI 1000 +0.70%. Large caps dragged, small/mid caps rose, consistent with the judgment that "money flowed into one specific industrial chain rather than into index heavyweights." BSE 50 at −0.94% was the only broad-based index clearly down.

Breadth and sentiment indicators

Advancer/decliner counts were cross-validated with two-directional pagination (sorting one way by percentage change systematically loses the tail of the decliners list, so one pass was taken in each direction and merged by ticker). Independently checked against the Stock Connect eligible pool: Shanghai 667 up / 924 down, Shenzhen 835 up / 1000 down, totaling 1502/1924 = advancer share 43.8%, consistent with the 43.3% market-wide figure — direction and magnitude corroborate each other.

Sentiment read: this is not a warming-up, it is a change in what is being traded.

Three indicators look contradictory — advancer/decliner counts repaired sharply and the failed-limit-up rate fell hard, yet turnover shrank 15%, limit-downs more than doubled, and the median of yesterday's limit-up stocks slid further from −1.47% to −2.21%. Only one explanation holds them together:

The money playing theme relays is leaving (shrinking volume + negative relay effect + "yesterday's consecutive limit-ups" index −2.23%), while the money playing the industrial main line is arriving (Telecom Equipment main force +RMB 11.873 bn). The failed-limit-up rate did not fall because sentiment improved; it fell because a good share of today's limit-ups were names inside the main line with industrial logic, and those seal more solidly than pure sentiment plays.

A detail easily read backwards: the market-wide average gap-open was only +0.14% with a median of +0.00%; the average open→close was −0.02% with a median of −0.30%, and only 40.7% closed above their open. In other words, the ordinary stock still "walked down after the open" today; today's positive returns were highly concentrated inside the main line. This is a different shape from yesterday's "all three indices gapped up and then slid all day": yesterday was a gap-up being cashed in; today was a gap-down chop that the main line then lifted.

---

2. Pre-Market List Reconciliation

2.1 Note: no pre-market list today, so the reconciliation target is yesterday's recap "Tomorrow's Outlook"

No A-share pre-market list was produced today, for reasons on our side, unrelated to the market.

Per the ironclad rule against self-scoring with no object, this section instead reconciles the judgments made for today in Section 6 "Tomorrow's Outlook" of yesterday's recap (2026-08-13-recap). That section contains 6 branch-direction judgments, 13 named stocks with verification points, and 7 avoid items — all of them written and fixed before today's open, and falsifiable, which makes a valid reconciliation object.

Per the self-imposed requirement in Section 6.4 of yesterday's recap, this table adds and permanently retains the "Gap-open %" and "Open→Close %" columns.

2.2 Branch-Direction Reconciliation (6 items)

2.3 Named-Stock Reconciliation (13 stocks)

2.4 Avoid-List Reconciliation (7 items)

2.5 Hit Rate and Post-Mortem

Hit rate by category

Post-mortem: the line between what was right and what was wrong is extremely clean.

1. Everything that was right rests on "single-stock microstructure" criteria. Number of failed limit-ups (风语筑 28 times → −8.38% today), turnover-rate pressure levels (誉衡药业 22.90% → rose to 44.27% today with main force −RMB 0.475 bn), contradictions between company announcements and share price (中鼎, 万邦, 城地香江 — all three right), and sealing quality. These criteria went 9/11 today, the most reliable part of the whole piece; they should be kept and weighted up.

2. Everything that was wrong rests on "extrapolating tomorrow's direction from one day of sector fund flow." Yesterday used "semiconductors main force −RMB 9.800 bn" to call memory fading, and "non-ferrous −RMB 16.765 bn" to call non-ferrous fading; today both were slapped down (the memory chain closed up across the board, Rare Earths +5.82% topped the market). This is the second time this month of falling in exactly the same place — single-day fund-flow extremes mean-revert the next day with high probability, so they are a contrarian indicator, not a trend-following one. Section 6.4 of yesterday's recap wrote for itself that "single-day fund flow must never again be used to exclude any branch," yet Sections 6.1 and 6.3 of that very same piece used it to exclude memory and non-ferrous. The rule was written but did not bind the body text — this is the single most important thing to fix.

3. The biggest misjudgment came paired with a correct verification point. Yesterday judged optical communications "divided, possibly even fading"; today it was the strongest thing in the market. But the verification point set yesterday for 天孚通信 (300394) was "whether main-force net inflow runs for a 2nd straight day," and today's answer is +RMB 1.180 bn, a 2nd straight day — the verification point held completely. In other words: the mechanism design was right, the headline conclusion was written backwards. The lesson: when a "branch judgment" conflicts with a "verification point," the verification point governs, rather than letting an already-written conclusion override it. Next pre-market must write the verification point first and derive the conclusion from it, not state a conclusion and then bolt on a verification point.

4. The gap-open gate would not have caused collateral damage this time. Section 6.4 yesterday required a new gate: "a gap-open above X% does not constitute an entry." Backtesting on today's data: main-line leader 亨通光电 (600487) gapped up only +1.48%, 剑桥科技 (603083) +0.70%, 利通电子 (603629) +2.13% — all below any reasonable threshold, so the gate would not have blocked them; what it would have blocked is precisely 亚泰集团 (600881, gap-open +8.37% → open→close −11.38%) and 博济医药 (300404, gap-open +12.25%). Recommend setting the threshold at a +5% gap-open; on today's backtest that threshold gives 0 false blocks and 1 correct block.

---

3. Main-Line Verification

3.1 The main line is optical communications / optical fiber cable, and the catalyst came from policy, not companies

3.2 Catalyst Sourcing (three items, publication dates all checked)

① CCTV Xinwen Lianbo broadcast on the evening of Aug 13, "China accelerates construction of next-generation communication networks" — this is the direct trigger of the main line. Key content: 10-gigabit optical network pilots run from 2025, with 136 pilot projects built so far; 5G base stations upgrading comprehensively toward 5G-A; MIIT organizing 6G ministry-province coordinated innovation pilots, aiming to produce a set of indigenously innovated 6G technology solutions before 2029; satellite-internet supporting construction accelerating (the state Guowang LEO constellation and the Qianfan commercial constellation continuing launches and networking, with direct handset-to-satellite connection having completed practical validation). Publication timing verified: East Money and Jrj.com both republished around 2026-08-13 21:01, making this an overnight catalyst with today as the first trading day.

② MOFCOM announcement of Aug 13: effective 2026-08-14, anti-dumping duties on imported single-mode optical fiber originating in India will continue to be levied, for a period of 5 years. But this one must be downgraded in interpretation. It is a "continued levy" following an expiry review, not a new levy: first imposed on 2014-08-14 (duty rates 7.4%—30.6%), already extended once on 2020-08-13, and this is the second extension, with neither rates nor scope changed. That is, its marginal economic impact is close to zero — the real change is "not cancelled," not "new protection added." The market traded it as fresh good news; that is a sentiment-level catalyst, not a fundamentals-level increment. Any recommendation premised on "anti-dumping being stepped up" does not stand.

③ On Aug 14 the PBoC conducted RMB 1 trillion of outright reverse repos (6-month tenor, 185 days, maturing 2027-02-15). Media framing needs correcting here too: several headlines wrote "PBoC releases a trillion in liquidity," but this was in fact a like-for-like rollover (the PBoC did not opt to roll over a smaller amount, but an equal one), so net injection is roughly zero. It explains why "the market did not fall on tightening funding conditions"; it cannot be used to explain any sector's rise.

3.3 Did pre-market call the strongest main line correctly? No — and it was avoidable

Wrong. Yesterday's recap listed optical communications as "divided; if the leader stays flat while second-tier names limit up, judge it fading." Today it was the strongest branch in the market.

But the nature of the error must be stated precisely: when yesterday's recap wrote that conclusion (8/13 15:30—16:00), Xinwen Lianbo had not yet aired (19:00 that evening), so that catalyst was genuinely unknowable at the moment of writing. This is therefore not "saw it and read it backwards," but "a conclusion written harder than the information supported" — with money already flowing into that main line for consecutive days (天孚通信 main force +RMB 1.052 bn on 8/13, the largest that day) and only price not moving, a directional "fading" forecast was issued anyway.

The correct formulation should have been: money flowing in for consecutive days while price has not moved = pending, with direction decided by whether the next day's money continues; do not preset a direction. That is exactly what yesterday's own verification point expressed.

3.4 Any unexpected main lines missed pre-market? Two

① IDC / CDN (missed). 网宿科技 (300017) +20.01%, main force +RMB 2.045 bn, 2nd market-wide; 数据港 (603881) limit-up (+RMB 0.709 bn); 奥飞数据 (300738) +7.08%. What the three share is "computing-power carriage" rather than "computing-power whole machines" — on the same day that 紫光股份 (−RMB 1.028 bn) and 浪潮信息 (−RMB 0.501 bn) were being distributed, money moved to the data-center and bandwidth side. This is a clean structural switch, entirely uncovered by yesterday's recap. Checked: 网宿科技's most recent announcement is an Aug 10 update on a subsidiary guarantee; there are no company-level announcements at all on 8/13—8/14, so its limit-up is theme-driven, not event-driven.

② Rare earths (called backwards). Rare Earths +5.82%, No. 1 industry by gain; 中国稀土 (000831) limit-up. Catalyst could not be located. Checked: 中国稀土's most recent announcements are the 8/13 "notice of the 2026 interim results briefing" (a procedural notice, not good news) and the 8/08 interim report; 北方稀土 (600111)'s most recent is an Aug 10 investor-relations activity record. Neither company has any substantive new announcement on 8/13—8/14. The rare-earth export-control documents retrieved were all Announcements No. 61/62 of 2025, i.e. old news, and must not be used as today's attribution. Stated honestly in this piece: the driver of today's rare-earth move is unidentified and no presumption is made. This is a verification task that must be completed pre-market on Monday.

A detail that would mislead any ranking: the leading gainer in the concept sector "Rare Earth Permanent Magnets +2.30%" is 金田股份 (Jintian Copper, 601609, Shanghai main board ±10%), but that stock's industry attribution in the limit-up pool is Industrial Metals, its main business is copper processing, and it is simultaneously counted in the "High-Speed Copper Cable Connectivity" concept (+2.27%). The logic of its limit-up belongs to the optical-communications / copper-connectivity chain, not to rare earths. Using the "Rare Earth Permanent Magnets" index or its leading gainer to measure rare-earth strength would overstate that branch. The true rare-earth reading should be the industry index "Rare Earths +5.82%" plus 中国稀土 (000831) and 北方稀土 (600111).

---

4. Limit-Up Ladder / High-Level Boards

Today: 63 limit-ups, 10 limit-downs, 19 failed limit-ups, a 23.2% failed-limit-up rate, and a highest streak of 5 consecutive limit-ups (flat with yesterday, no new high-water mark).

Ladder distribution: 5 boards ×1, 3 boards ×5, 2 boards ×5, first boards ×52. The structure is extremely "wide at the base, narrow at the top" — first boards make up 82.5%, while the 4-board slot is completely empty.

4.1 The high board: 5-board 蓝盾光电, but its sealing quality is among the worst market-wide

蓝盾光电 (Blue Shield Optoelectronics, 300862, ChiNext ±20%) +19.99%, 5 limit-ups in 5 days, today's turnover RMB 3.495 bn, turnover rate 51.69%, failed the limit 5 times, sealing order only RMB 0.031 bn.

A sealing-order/turnover ratio of just 0.01 — the money sealing the board equals only 1% of the day's turnover, while the turnover rate exceeded 50% and the board broke 5 times intraday. This is the textbook shape of "many people, but nobody willing to commit chips," and it is the least stable high board in the market. Its sector, General Equipment, has nothing to do with today's main line; it is a standalone sentiment play.

4.2 Sealing-quality ranking: the gap between locked boards and sloppy boards is enormous

How to read it: names with seal/turnover > 1 (坤泰股份, 神奇制药, 澳洋健康) are genuinely chip-locked boards, but their turnover is tiny (RMB 0.030—0.087 bn), so they lack capacity; the streak stocks with large turnover (蓝盾光电, 金螳螂, 共进股份, 博济医药) all have seal/turnover below 0.06 with turnover rates of 14%—52%, making them high-intensity churn boards. The latter will be markedly more volatile the next day than the former.

4.3 The core names: the main line's most identifiable stocks are on first boards, not streaks

Today's most identifiable stocks are all first boards, and all concentrated in optical communications:

亨通光电 (600487) is today's only stock that simultaneously ranks No. 1 in turnover, No. 1 in sealing order, and No. 1 in main-force net inflow (+RMB 3.121 bn); it is the unambiguous core of this main line.

4.4 An intraday structure easily missed: 32% of limit-ups only sealed in the afternoon

Distribution of first seal times (63 names in total):

Nearly all the main-line leaders sealed in the afternoon: 亨通光电 13:59, 利通电子 14:19, 网宿科技 14:14, 数据港 13:44, 中瓷电子 (China Zhongci Electronic) 14:26, 海星股份 14:26, 旭光电子 (Xuguang Electronics) 13:39.

What this structure means: today's main line was not the gap-up-and-grab type, but the strengthen-through-the-session, accelerate-in-the-afternoon type. 亨通光电 gapped up only +1.48% for the whole day with open→close +8.40%; 剑桥科技 gapped up +0.70% with open→close +9.23%. In other words, buying the main-line leaders at today's opening price made money — the exact opposite shape from yesterday's "all 5 recommended stocks gapped up, all closed below their open, open→close −3.95%."

Direct implication for Monday: an afternoon seal means the absorption came from that day's new money rather than overnight chips; such limit-ups are more likely to gap up the next day, and also more likely to fade after gapping up. Monday must be verified with the first 30 minutes after the open, not by chasing the gap directly.

4.5 Ten limit-downs: the count doubled, but still no sector character

What they share is not an industry but an identity: 秦安股份 (5 boards yesterday), 京投发展 (4 boards yesterday) and 德龙汇能 (2 boards yesterday) were all streak stocks yesterday and went straight to limit-down today. Add 同力天启 (Tongli Tianqi, 605286, 4 boards yesterday) at −10.00%, 皇氏集团 (Huangshi Group, 002329, 4 boards yesterday) at −7.58% and 北京文化 (Beijing Culture, 000802, 4 boards yesterday) at −5.43%, and not one of yesterday's 4-board-and-above high-level stocks escaped. This matches exactly the concept index "yesterday's 2nd-board-and-above performance" at −4.91%.

---

5. Fund Flows

5.1 Northbound flows: disclosure has ceased; the 0 returned by the interface is not a real 0

Today's Stock Connect aggregate readings for the Shanghai and Shenzhen northbound channels are net buy turnover 0.00 and net capital inflow 0.0.

This is not "zero northbound net inflow"; it is the silent return that follows the exchanges' cessation of daily northbound net-buy disclosure from 2024-08-19 onward. This piece makes no judgment whatsoever about the direction of northbound flows.

⚠️ The "420.0" shown for southbound (Hong Kong Stock Connect) in the same interface is the daily quota ceiling, not an inflow amount, and must not be quoted as a flow reading. Today's southbound net buy turnover was +RMB 1.009 bn on the Shanghai leg and −RMB 0.988 bn on the Shenzhen leg.

5.2 Dragon-Tiger List: today's data not yet published

The Dragon-Tiger List is published after 18:00 on a trading day, and this piece was generated at 15:30, so this section is left empty for today. This is a publication-timing constraint, not missing data.

The Aug 14 Dragon-Tiger List will be available after 18:00 tonight and should be used in Monday's pre-market list to verify: ① the split between institutional seats and hot-money seats within 亨通光电's (600487) RMB 19.197 bn of turnover — this determines whether the optical-communications main line is institutional accumulation or a hot-money relay; ② the buy-side structure behind 网宿科技's (300017) +20%; ③ the character of the money behind 中国稀土's (000831) limit-up.

5.3 Main-force money: direction points, with high consistency, at one single chain

Top 15 main-force net inflows (unit: RMB bn)

8 of the top 15 belong to the same optical-communications/IDC chain (亨通, 网宿, 新易盛, 中天, 剑桥, 天孚, 杭电, 数据港), with combined net inflow of RMB 12.556 bn. Concentration this high has been rare recently.

Top 10 main-force net outflows (unit: RMB bn)

Leading sectors by main-force net inflow (industry basis)

Leading sectors by main-force net outflow

This comparison table is the single most important one today: money flowed out of "theme stocks (−RMB 6.758 bn), yesterday's consecutive limit-ups (−RMB 1.519 bn), innovative drugs (−RMB 3.568 bn) and power (−RMB 8.909 bn)" and into "telecom equipment (+RMB 11.873 bn)." This is a clearly directional switch from hot-money themes to an industrial main line, not a broad rally.

---

6. Tomorrow's Outlook

Note: today is Friday. The next trading day is Monday 2026-08-17, with two market holidays in between, so there is a weekend news window. Every "tomorrow" in this section means Monday 8/17.

6.1 Monday (8/17) main-line continuation calls

6.2 Monday (8/17) watch list

Scheduled events (note the weekend gap):

• 生益科技 (Shengyi Technology, 600183, Shanghai main board ±10%) and 联瑞新材 (Lianrui New Materials, 688300, STAR Market ±20%) will disclose interim reports on Saturday, Aug 15. 8/15 is a non-trading day, so both interim reports will be priced in all at once on Monday 8/17. The two closed today at +1.04% (main force −RMB 0.414 bn) and +1.50% (+RMB 0.100 bn) respectively, neither moving in advance.
• The Aug 14 Dragon-Tiger List publishes after 18:00 tonight and can be retrieved in full pre-market Monday, to determine the character of the money behind the optical-communications main line.
• The weekend contains two full news days; follow-up policy interpretation of the Xinwen Lianbo communication-network report, plus any industry data releases, will all be priced in together on Monday. Every judgment in this piece is bounded by information as of the 8/14 close.

6.3 Monday (8/17) avoid list

1. Yesterday's 4-board-and-above high-level stocks as a group. Today 秦安股份 (Qin'an Electromechanical, 603758, Shanghai main board ±10%, 5 boards yesterday), 京投发展 (Jingtou Development, 600683, Shanghai main board ±10%, 4 boards yesterday) and 同力天启 (Tongli Tianqi, 605286, Shanghai main board ±10%, 4 boards yesterday) all limited down, with 皇氏集团 (Huangshi Group, 002329, Shenzhen main board ±10%) −7.58% and 北京文化 (Beijing Culture, 000802, Shenzhen main board ±10%) −5.43%. The concept index "yesterday's 2nd-board-and-above performance" was −4.91%. This one has been validated today; keep executing it.
2. 蓝盾光电 (Blue Shield Optoelectronics, 300862, ChiNext ±20%). A 5-board height but seal/turnover of only 0.01, turnover rate 51.69%, failed the limit 5 times — the worst sealing quality of any high board in the market, and its theme (General Equipment) has nothing to do with the main line.
3. 华西股份 (Huaxi Co., 000936, Shenzhen main board ±10%). 3 boards but failed the limit 7 times today, seal/turnover 0.05 — the sealing money is extremely irresolute.
4. The whole pharma chain. Innovative Drugs, Chemical Pharmaceuticals and Medical Services show main-force outflow in the same direction on all three measures, totaling −RMB 8.035 bn, and the ladder is broken. Includes 誉衡药业 (Yuheng Pharmaceutical, 002437, Shenzhen main board ±10%, turnover rate up to 44.27%, main force −RMB 0.475 bn).
5. Power (thermal and green). Thermal Power −3.31%, Green Power −RMB 5.390 bn; 大唐发电 (601991), 华电能源 (Huadian Energy, 600726, Shanghai main board ±10%), 华电辽能 (Huadian Liaoning Energy, 600396, Shanghai main board ±10%, −9.29%). Yesterday's rule "if the catalyst is unverified, do not chase" directly avoided a loss today; keep executing it.
6. The computing-power whole-machine end. 紫光股份 (000938), 浪潮信息 (000977) — 2 straight days of main-force net outflow, confirmed through the gate.
7. Pure concept plays that contradict the company's own announcements. 中鼎股份 (Zhongding Sealing Parts, 000887, Shenzhen main board ±10%, −7.89% today), 万邦医药 (Wanbang Medical Technology, 301520, ChiNext ±20%, −3.26%), 城地香江 (Chengdi Xiangjiang, 603887, Shanghai main board ±10%, −4.56%). All three delivered on two consecutive days; this criterion has proven stably effective.
8. "No-money catch-up names" inside the main line. 永鼎股份 (Yongding Co., 600105, Shanghai main board ±10%) gained only 1.40% with open→close −0.53% on a day the main line rose +5.33%; 通鼎互联 (Tongding Interconnection, 002491, Shenzhen main board ±10%) rose +3.75% but main force was −RMB 0.017 bn. Names that cannot outrun the main line on the main line's breakout day should not be chased the next day.

6.4 Input notes for Monday's pre-market list

1. [Highest priority] Write the verification point first, then derive the conclusion from it. Today's biggest misjudgment (optical communications) and its most accurate judgment (天孚通信's consecutive net inflow) came from the same piece. When a branch conclusion conflicts with a verification point, the verification point governs. The writing order for the pre-market list should change to: list the money/price facts → set a falsifiable verification point → let the verification point determine the conclusion, with stating a conclusion first and bolting on a verification point afterwards forbidden.
2. [Hard constraint] One day of sector fund flow may only be used to "confirm," never to "exclude." Yesterday used "semiconductors −RMB 9.8 bn" and "non-ferrous −RMB 16.7 bn" to exclude two branches, and today both were slapped down (the memory chain closed up across the board, rare earths topped the market). This is the second time this month of falling in the same place. In Monday's list, excluding any branch must be backed by same-direction evidence on 2 consecutive days; single-day extremes are for observation only, never for exclusion.
3. Keep the "Gap-open %" and "Open→Close %" columns, with the gap-open gate set at +5%. Today's backtest: that threshold gives 0 false blocks (main-line leaders all gapped up < +2.6%) and 1 correct block (亚泰集团 gapped up +8.37% → open→close −11.38%).
4. Add a "first seal time" column. 32% of today's limit-ups only sealed after 13:00, and the main-line leaders were almost all afternoon seals. An afternoon seal = absorption by that day's new money, with a higher probability of fading after gapping up the next day — information completely invisible if you only look at the closing percentage change.
5. Add a "seal/turnover" column to replace the raw sealing-order amount. 蓝盾光电's sealing order of RMB 0.031 bn looks non-trivial today, but against turnover it is 0.01; 坤泰股份's sealing order was RMB 0.337 bn, a ratio of 11.26. Only the ratio distinguishes a locked board from a sloppy one; the absolute value misleads.
6. Three verifications that must be completed: ① the driver of today's rare-earth move (neither leader has a new announcement, and everything retrieved was old 2025 policy — take care not to write old policy up as today's catalyst); ② the 8/14 Dragon-Tiger List (available after 18:00 tonight), to determine the seat character behind 亨通光电's (600487) RMB 19.197 bn of turnover; ③ weekend news flow, especially follow-up interpretation of the communication-network policy and the interim reports of 生益科技 (600183) and 联瑞新材 (688300) disclosed on Saturday 8/15.
7. When citing a policy catalyst, always label it "new" or "continued." Today's MOFCOM anti-dumping announcement was written up as good news by several media outlets, but it was first levied in 2014 and, after a first extension in 2020, this is the second extension, with neither rates nor scope changed, so the marginal impact is close to zero. By the same token, the PBoC's RMB 1 trillion outright reverse repo was a like-for-like rollover rather than a net injection, and the phrasing "releasing a trillion in liquidity" cannot be accepted. Both of these nearly got written up as reasons for the rise today.
8. The fallback path when the pre-market list is missing has been validated: reconcile against the previous day's recap "Tomorrow's Outlook" instead. That section comes with named stocks and verification points, and is no less falsifiable than a regular list. From now on, handle the same situation this way and do not skip the reconciliation step.

---

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

Data-Sourcing and Run Log (internal, not sent to clients)

1. No A-share pre-market list today — the root cause is cron's single-lock serialization, not a script failure

reports/a-share/2026-08-14.md does not exist. Checking reports/cron.log: today has only 06:00 START us recap and 15:30 START ashare recap; the 07:00 ashare premarket did not even print a START.

Root cause: the four report slots share one lock, and it is non-blocking —

0 7  * * 1-5  flock -n /run/lock/report.lock ... ashare premarket
0 6  * * 2-6  flock -n /run/lock/report.lock ... us recap

Today's us recap started at 06:00 and ran until it was TERMed at 07:30 (ALERT SENT — terminated by TERM (most likely cron's 90-minute timeout ceiling)), holding report.lock throughout. The 07:00 ashare premarket could not get the lock and was silently skipped by flock -n — when flock -n fails to acquire, it exits immediately and prints nothing, so there is no trace in the log at all.

This is fail-silent: one slot timing out eats the next slot with it, leaving no log and sending no alert. Note that us recap did send its own failure alert email, but the ashare premarket it squeezed out sent no alert whatsoever — we only find out when the post-close recap has no reconciliation object.

Suggestions (not applied; left for a human to decide):

• add || echo "==== $(date) SKIP <slot> — lock held ====" to flock -n to at least leave a trace;
• or switch to a per-slot lock (report-ashare-pre.lock, etc.); the slots share no state, and the point of one shared lock was to limit concurrent LLM calls, but the price is cascading skips;
• the 90-minute timeout on us recap deserves a separate look: it ran only 21 minutes on 8/13, but ran the full 90 minutes today and was killed.

2. push2 main host returns 502 on this machine; switched to push2delay

push2.eastmoney.com and 82.push2.eastmoney.com returned 502 Bad Gateway (nginx) to this machine today, and akshare's stock_zh_index_spot_em / stock_board_industry_name_em raised RemoteDisconnected accordingly. Working channels: push2delay.eastmoney.com (market-wide snapshot / sectors / K-line) + push2ex.eastmoney.com (limit-up / failed-limit / limit-down / streak pools). Verified that push2delay is not stale data: checking index snapshots line by line, the f124 timestamps are 15:31—15:32, and turnover corroborates the per-stock sum. The fetch script landed in work/fetch_recap_20260814.py. ⚠️ work/ is inside .gitignore, and this is already the third rewrite of the same fetcher (the previous two are recorded in the internal blocks of 8/13 and 8/12), still not hardened into bin/. Recommend doing a dedicated pass next time.

3. Two-directional pagination came out clean this time, but must be kept

Following the 8/13 lesson, clist was fetched in both directions with po=0 / po=1 and merged. Today push2delay returned the complete 5896 records in a single direction (total matched collected), with none of the 8/13-style truncation where "one direction gives only 3900 records and everything missing is the tail of the decliners." But two-directional fetching is very cheap while its failure mode is silent, so keep it. Cross-validation passed: per-stock turnover sums to RMB 2,156.6 bn vs Shanghai 990.37 + Shenzhen 1,152.47 + Beijing 13.85 = RMB 2,156.69 bn.

4. Limit-up/down determination cannot be screened by self-set thresholds; the official pool is mandatory

The first version screened limit-ups/downs from the market-wide snapshot using "change ≥ board limit −0.6%", which produced 19 limit-downs while the official DT pool has only 10. The difference is entirely in ST stocks: ST main-board stocks are limited to 5%, and the L−0.6 tolerance band is far too wide for a 5% limit, counting every ordinary decline below −4.4% as a limit-down (ST 喜临门 (ST Sleemon) −5.06%, *ST 萃华 (*ST Cuihua) −5.95%, *ST 联翔 (*ST Lianxiang) −7.35%; the latter two cannot possibly be limit-downs under a 5% cap, showing that their actual cap is not 5% and that my board→limit mapping is wrong for ST names). Everything has been switched to the official push2ex ZT/DT/ZB/QS pools, and all figures in the body come from the official pools. Lesson: limit-up/down determination involves overlapping board × ST × new-listing × registration-system rules, so deriving it from your own thresholds is guaranteed to be wrong; the exchange-basis pool is mandatory. Also: getTopicDTPool and getTopicQSPool with sort=fbt:asc return a correct tc but an empty pool; switching to sort=fund:asc / zdp:desc finally returned data. A non-zero tc with an empty pool is a silent failure that raises no error; retries with different sort keys have been added.

5. Three stale-news / basis-mismatch items intercepted

• The vast majority of optical-fiber-cable articles returned by search are from February and June 2026 (中天科技 limit-up at a record high, 亨通 3 boards in 6 days on RMB 17.4 bn, etc.). Taking them at face value would have written a six-month-old move up as today's attribution. All discarded, replaced by going back to the 8/13 21:01 republication timestamp of Xinwen Lianbo and the date of the MOFCOM announcement.
• "PBoC releases a trillion in liquidity" is media-headline framing; the reality is RMB 1 trillion of like-for-like rollover, net injection ≈ 0. Explicitly corrected in body Section 3.2, with the note that it cannot be used to explain any sector's rise.
• The MOFCOM anti-dumping announcement is a "continued levy" (a second expiry-review extension, with rates and scope unchanged), not a new protective measure. The body downgrades it and states that the marginal impact is close to zero. If these three had been copied straight from search summaries, two of today's attributions would have been wrong.

6. Data not obtained / not identified (all disclosed to clients in the body)

• Dragon-Tiger List: published after 18:00, this piece generated at 15:30, stock_lhb_detail_em returns NoneType. A scheduling problem, not an interface failure, for the third consecutive day with the same cause. Body Section 5.2 leaves it blank and explains why.
• Northbound flows: the interface returns 0.0, which is the silent return following the cessation of daily net-buy disclosure on 2024-08-19. Body Section 5.1 has been rewritten as a basis-of-measurement note, and again warns that the 420.0 in the same interface is a quota ceiling.
• Today's rare-earth catalyst: not identified. Checked: 中国稀土's (000831) most recent announcements are the 8/13 results-briefing notice (procedural) and the 8/08 interim report; 北方稀土's (600111) most recent is the 8/10 investor-relations record. Rare-earth export-control hits from search were all Announcements No. 61/62 of 2025 (old news). Body Section 3.4 honestly states "could not be located, no presumption made" and lists it as a must-check for Monday.
• The driver of 网宿科技's (300017) +20%: checked, no company announcements on 8/13—8/14, judged theme-driven. This one was confirmed by actively going to check the announcements, not by "found nothing in search so wrote that there was nothing" — per last month's lesson, "no news found" ≠ "no news," and limit-up stocks must have their announcement dates checked.

7. Methodology choice this time

When the pre-market list is missing, reconciling against the previous day's recap "Tomorrow's Outlook" proved workable this time: that section contains 6 branch judgments + 13 named stocks + 7 avoid items, all fixed before today's open, and its falsifiability is arguably stronger than a regular pre-market list (because it carries its own verification points). Recommend hardening this as the standard fallback path; already written into body Section 6.4, item 8.

Sub-agents were not run (fundamentals-analyst / risk-auditor): the recap slot is driven mainly by market facts, and the 15:30 generation deadline is tight. The four post-mortem points in Section 2.5 and the error-nature analysis in Section 3.3 are all self-review by the main body, with no third-party QC, so their confidence is lower than in the pre-market slot.

One doubt that could not be fully eliminated: today's "falling failed-limit-up rate + repaired advancer/decliner counts" coexists with "15% lower volume + negative relay effect," and the explanation I gave is "theme money leaving, main-line money arriving." That explanation is consistent with every observation, but I have no independent data directly proving it is the same pool of money switching (that would need seat data, i.e. the Dragon-Tiger List, unavailable today). The body's phrasing is limited to "only one explanation holds," and claims no proof.