Hong Kong — End-of-Day Verdict
3 ACQUIRE + 3 DISPOSE. ACQUIRE: 0823 LINK REIT ADD @39.12 buy-limit (adj 78 = 73 + 0 + 5; PullbackBuy_VWAP, Buy_Rank 2, RSI 56.8, the only unextended buy in the book; quality gate N/A on an ADD; also armed-not-fired so the stop rises 37.84 -> 37.86 alongside the add) = HK; 9618 JD-SW BUY 650 QUEUED @123.16 exp 08-12 (adj 78 = 73 + 0 + 5; same family, quality gate PASSES, but 2.2 ATR extended so queued) = HK; 0992 LENOVO BUY 2,000 QUEUED @28.57 exp 08-12 (adj 70 = 73 - 3 + 0; quality rank and news confirms a record quarter with AI at 30% of revenue, but a 52W-high chase family at RSI 69 - sized DOWN one lot per HK-1) = HKDISPOSE: 0939 CCB SELL 15,000/30% @8.855 market (ADR-0012 armed, ONE tick - dead 12wk; tick (b) REVERSED as conviction went, so the tier DE-ESCALATES from 08-07's 50%; still the largest banked trim at HK); 3750 CATL SELL 200/67% @641.00 market (two ticks - dead 12wk + NO TRADE on five consecutive prints; negative on all three horizons; HK); 7709 XL2CSOPHYNIX SELL 900/50% @28.70 market (breakdown-family signal + NEWS OVERRIDE: SK Hynix is >20% off its highs after its largest one-day drop ever, and this is a 2x daily-reset product on it; HK - ADR-0012 is silent, origination is the leveraged-ETF rule). Sell proceeds HKvs HKact-now + HKconditional buying - 1.4:1; goes to ~HK if only the add fills, ~HK if both queues fill. 0981 SMIC RETIRED to a WATCH line (HK-11 session 3; the model withdrew its own trim - STRONG SELL cv73 decayed to AVOID cv60, Portion 0, no Sell_Rank; HK-9: that decay is NOT a re-entry case). 0316 OOIL blocked at adj 94 and 3690 MEITUAN at adj 75, both by the quality gate - for a fourth session this book is candidate-constrained, not capital-constrained, and the binding constraint is a gate nobody has ever graded. SYSTEMIC: PROFIT_ARMED fires on held names for a third session and is ANTI-CORRELATED with actual profit - the one unarmed name (0522) is the most underwater at -14.7%.
Recommendations
3 to acquire · 3 to dispose
| Ticker | Name | Close | Action | Action Price Where to act — a buy-limit on a pullback or a buy-stop above resistance, not the current price. Sanity-check vs Close: on a held name, an action price stranded above the close means the queued level was set when the stock was higher — stale, awaiting reprice. | Stop The risk-defined exit, set below the entry for a long (~1.2–1.8×ATR). A held long whose Stop sits at/above the Close is already below its stop — a stale box awaiting reprice. | 1wk Model estimate of the likely price at this horizon — a forward projection, not a guarantee, and separate from the trade's Action Price or Stop. Where the ML estimator is enabled these are per-ticker calibrated forecasts (~80% coverage band) that are refreshed periodically and can drift between refreshes; otherwise they're interpolated from the signal's swing target. Confirm against the settled close. | 4wk Model estimate of the likely price at this horizon — a forward projection, not a guarantee, and separate from the trade's Action Price or Stop. Where the ML estimator is enabled these are per-ticker calibrated forecasts (~80% coverage band) that are refreshed periodically and can drift between refreshes; otherwise they're interpolated from the signal's swing target. Confirm against the settled close. | 12wk Model estimate of the likely price at this horizon — a forward projection, not a guarantee, and separate from the trade's Action Price or Stop. Where the ML estimator is enabled these are per-ticker calibrated forecasts (~80% coverage band) that are refreshed periodically and can drift between refreshes; otherwise they're interpolated from the signal's swing target. Confirm against the settled close. | Rationale |
|---|---|---|---|---|---|---|---|---|---|
| 0823 | LINK REIT | 39.22 | ADD | 39.12 | 37.86 | 39.40 (+0.7%) | 39.54 (+1.1%) | 40.31 (+3.0%) | Hong Kong's largest REIT has pulled back into its own volume-weighted average price while still trending above its 200-day line, and the model rates that setup a BUY on the one signal family this book has actually made money in - a 75% one-week hit rate over ten graded trades. Valuation is full rather than cheap, so this is a trend-following add on an existing 2,800-share holding, not a value entry: the 12-week view is +3.0% with a tight 3.2% stop, which is a favourable risk line rather than a big return. Adding at HKtakes the position to 4,600 (30% of the sizing base, inside the 60% cap). Adjusted conviction 78 = 73 raw + 0 valuation + 5 scorecard. The name also prints PROFIT_ARMED but no ADR-0012 tick fires, so the required response is a trail-raise (stop 37.84 -> 37.86), executed alongside the add rather than instead of it. Entry range 38.96-39.28 (+-0.25 ATR). Quality gate does not apply to an ADD. 12wk 40.31 exceeds the DCF anchor 37.84 - projection assumes growth beyond priced-in. Board lot 100 verified. · news |
| 9618 | JD-SW | 130.40 | BUY — QUEUED @123.16 | 123.16 | 119.49 | 130.85 (+6.2%) | 131.58 (+6.8%) | 135.48 (+10.0%) | China's second-largest e-commerce platform triggered a fresh buy signal on the pullback family this book grades best, and unusually for a Hong Kong name it also clears the quality screen - positive free cash flow of HK6bn, quality rank - with the shares priced at roughly fair value rather than at a premium. The 12-week view is +10.0% against a 3.0% stop. The catch is timing: the stock closed 2.2 ATR above the model's entry after a +2.5% day at RSI 69, which is a chase, so this is queued at HK (-5.6%) rather than bought. Adjusted conviction 78 = 73 raw + 0 valuation + 5 scorecard; ranked behind LINK REIT only on Buy_Rank (2 vs 4). Sizing 70.59% x HKx 20% = HKrounded DOWN to (HK) to stay inside the cap. Entry range 122.35-123.97 (+-0.25 ATR). expires_on 2026-08-12 is deliberate: Q2 results land 13-Aug after the HK close per hk_horizon.json (Yahoo estimates 11-Aug; the divergence is unresolved), so the order is off the book before the later date, and L6 forbids chasing the gap afterwards. Board lot 50 assumed - verify. · news |
| 0992 | Lenovo Group Limited | 29.00 | BUY — QUEUED @28.57 | 28.57 | 24.00 | 29.53 (+3.4%) | 29.61 (+3.6%) | 29.01 (+1.5%) | Lenovo broke to a fresh 52-week high on heavy volume and is the highest-quality name in this book by a distance - F-Score 7, ROE 27.9%, positive free cash flow, quality rank - with the news confirming why: a record quarter at US5bn revenue (+14.6%), adjusted net income +25%, and AI-related sales now 30% of the group. The caution is price and timing, not the business. The shares are up 114% in 30 days, the signal family is one this book has never made money in, and the stock now trades at 1.63x its estimated intrinsic value, so the entry is queued at the model level HK (-1.5%) and sized DOWN one board lot rather than up. Adjusted conviction 70 = 73 raw - 3 valuation + 0 scorecard (BreakoutUp_52W|buy is n=2, report-only; the adjacent BreakoutUp_20D|buy cell grades -3 and HK-1 treats all momentum-chase families as chase-grade). MoS_FV recomputed at the live 29.00 close (-0.632) rather than the stale 24.74 fairvalue price (-0.392) - the name moved +17% since the valuation fetch and the conservative direction is taken for a buy. RECONCILE: the 08-07 GTC @27.32 TRIGGERED today (low 27.28) and was NOT executed - an execution miss worth -HKat the close, not an unfilled queue; repriced to the live entry per the provenance rule, never rolled. expires_on 2026-08-12 puts the order off the book one session before Q1 FY2026/27 results on 13-Aug (L6). It printed BUY - ANTICIPATE at 15:00 (cv75) and 15:30 (cv83) before graduating to a full BUY at the anchor, so it takes an ACQUIRE slot and the separate ANTICIPATE slot is left unused rather than double-counting the name. Entry range 28.12-29.02. Board lot 2,000 assumed - verify. · news |
| 0939 | CCB - H SHS | 8.85 | SELL 30% | 8.86 | 8.43 | 8.84 (-0.2%) | 8.80 (-0.6%) | 8.71 (-1.6%) | China Construction Bank has done its work and the model now projects nothing further from it - the 12-week view sits below today's price and all three horizons are mildly negative (-0.2% / -0.6% / -1.6%) - so this books a third of the book's second-largest position, HKwhile the shares are firm rather than falling. It is a redeployment of dead capital, not a call that the bank is in trouble: the signal is a stable HOLD, the price rose 0.7% today and sits above its 200-day line, and stay invested. ADR-0012 profit-take: armed (T12_Progress 101.66), ONE tick fired - (c) dead forward view, 8.71 < 1.02 x 8.855. Tick (b) has REVERSED (conviction 39 on 08-07 -> 47 today), so the tier DE-ESCALATES from the 50% proposed on 08-07 to profit_take_1 = 30%. Valuation NA_CCY, blind. Scorecard cell FailedBreakdown_Up_20D|sell is n=3, report-only, no credit. HK-11 count: this is re-proposal # (card #1 was 08-07 @8.79, unexecuted) - if unexecuted again it must retire to a WATCH line tomorrow and never be re-issued. Residual stop RAISED 8.34 -> 8.43 (2.0x ATR under the close). Board lot 1,000 assumed - verify. · news |
| 3750 | CATL | 641.00 | SELL 50% | 641.00 | 625.50 | 638.55 (-0.4%) | 625.19 (-2.5%) | 604.48 (-5.7%) | The world's largest EV battery maker has stopped generating a tradeable signal in this book - the model has printed NO TRADE on five consecutive snapshots across two sessions, refuses to quote an entry, a stop or a target, and projects the price lower on every horizon (-0.4% / -2.5% / -5.7%). The position is 14.7% underwater and roughly 2.4% above the line where it would be cut anyway, so this takes HKoff the table into a firm tape and keeps against a possible recovery. ADR-0012 profit-take: armed (T12_Progress 106.04), TWO ticks fired - (c) dead forward view (604.48 < 1.02 x 641.00) and (d) repeated NO TRADE - so profit_take_2 = 50% tier. is not a lot multiple; rounded UP to 200 because a dead forward view on a name whose model will not quote a stop makes the overshoot the safer error. Valuation FAIR_VALUE, MoS -1.08: neither <= -3 (no tier bump) nor UNDERVALUED (no de-amplification), so valuation is silent. NoTrade|sell grades +3 actionable but is NOT applied - the origination is the arm, not a label, and crediting a bias to the absence of a label would double-count. HK-11 count: re-proposal # (card #1 was 08-07 @632.50, unexecuted; today's exit is HKbetter). Residual stop held at 625.50, HKunder today's 626.50 low - raising it further would put the line inside today's range. Board lot 100 assumed - verify. · news |
| 7709 | XL2CSOPHYNIX | 28.70 | SELL 50% | 28.70 | 28.00 | 30.20 (+5.2%) | 30.08 (+4.8%) | 30.40 (+5.9%) | This is a 2x leveraged bet on SK Hynix, and SK Hynix has just fallen more than 20% from its highs after the largest single-day drop in the company's history, dragging Micron and Samsung into a bear market on profit-taking, AI-financing worries and a brokerage note flagging a current-quarter profit miss. A daily-reset double-leverage product does not ride out a drawdown like that - it decays through it. Halving the position takes HKoff a holding already down 75% from cost and leaves against a bounce. The model's own estimates point the other way (+5.2% / +4.8% / +5.9%) and are deliberately discounted: HK-6 (calib_1w -0.249), HK-9 (an HK sell read is a one-week instrument), and because a 12-week point estimate on a daily-reset 2x ETF is not a meaningful object. Technical basis: the label flipped back into the breakdown family (BreakdownDown_20D, AVOID cv54 flat across all four prints, RSI 36.2) and the shares probed the 28.00 cap intraday for a second consecutive session (low 27.50 today, 28.04 on 08-07) without a decisive close below it. Origination is the standing leveraged-ETF rule (trim 50% on a deteriorating read; full exit reserved for conviction >=70 + 3 prints + a close below the stop, OR a breakdown-family signal) - a breakdown signal is present and would permit a full exit, but the 50% tier is taken deliberately because the 07-17 full-exit-then-withdraw episode (HK-12) is this name's failure mode at both ends. ADR-0012 is SILENT here: armed but no tick fires. Scorecard BreakdownDown_20D|sell +1 (n=15, actionable). Adjusted 55 is below the 65 signal floor and does not need to clear it - the floor governs the signal path, this card originates on the instrument rule. Residual: the 28.00 cap stands; a decisive CLOSE below it is a full exit. Model Stop_Price 42.81 (49% ABOVE the close) and Target_Price -27.94 (negative) are rejected as HK-12 degenerate. · news |
| 0981 | SMIC | 66.20 | HOLD | 66.20 | 58.86 | 66.55 (+0.5%) | 67.63 (+2.2%) | 69.22 (+4.6%) | China's largest chip foundry is retired from the sell queue rather than sold. The model has withdrawn its own trim - the label decayed from STRONG SELL - REDUCE at conviction 73 on 08-07 to AVOID at 60 across all four prints today, with 0.00 and no Sell_Rank - so there is no longer a signal to act on, and the position stays whole at behind a 58.86 stop. This is the HK-11 terminal step: cards ran 08-03 @62.20, 08-04 @65.30, retire+re-arm 08-05, re-fired 08-06 @65.35 and 08-07 @66.90; today is session 3 since the re-arm, so the exit retires to a WATCH line and must never be re-issued without a fresh trigger. RE-ARM: a fresh SELL-REDUCE print, or a decisive close below 58.86. HK-9 governs what the decay does NOT mean - it is not a re-entry case; a buy here needs its own fresh BUY-side signal clearing the normal floor. Structure is still unfavourable: close 66.20 below EMA200 68.48, valuation GROWTH_PRICED_IN at MoS -4.39. Model Stop_Price 73.80 prints ABOVE the close and is rejected as HK-12 degenerate; state stop stays 58.86. The 06-Aug Q2 print flagged in hk_horizon.json still could not be confirmed on a third session of looking. · news |
| 0316 | OOIL | 150.00 | WATCH | 146.94 | 141.95 | 151.36 (+3.0%) | 154.87 (+5.4%) | 172.25 (+17.2%) | The container shipper is the strongest technical and valuation combination in the book - STRONG BUY at conviction 86, Buy_Rank 1, on the pullback family that grades best here, and the only name screening as genuinely UNDERVALUED - but it is blocked from a new position for a third consecutive session by the quality screen, and the fundamentals back that call: free cash flow is NEGATIVE (-HK), revenue is shrinking 9.2% with forward revenue -20%, F-Score 5, quality rank 21.. Cheap and shrinking is not the same as cheap and compounding. Adjusted conviction 94 = 86 raw + 3 valuation (halved from +6 because quality fails) + 5 scorecard - the highest reading of the day by 16 points and the most consequential rule in this book. Entry_Distance_ATR -0.69 means it would have been QUEUED at 146.94, not bought at the 150.00 close, even if the gate had passed. It has run 148.60 -> 150.00 since the first block. RE-ARM: a Quality_Pass flip on the next fairvalue refresh with the PullbackBuy print intact. Logged as gate_blocked so the counterfactual grader can eventually price what this gate costs or saves - see systemic finding 3; three sessions of blocking the day's best candidate with zero measurement is the real defect. |
| 3690 | MEITUAN-W | 93.90 | WATCH | 88.39 | 84.86 | 93.53 (+5.8%) | 92.47 (+4.6%) | 85.06 (-3.8%) | Meituan triggered a buy signal on the book's best family and would otherwise have taken the third acquisition slot at adjusted conviction 75, but it is the worst-quality name in the portfolio on the numbers - F-Score, operating income -HK8bn, free cash flow -HK1bn, quality rank - and its own forward view is negative, with the 12-week estimate at 85.06 against a 93.90 close (-9.4%). Blocked. JUDGEMENT CALL STATED: Val_Signal is NA (EPV per share is negative, so no fair value can be struck) and the Step 6 rule says NA_* rows pass with the gate off, but Quality_Pass here is not missing - it is computed and FALSE on real fundamentals. The gate is applied, because 'valuation could not be computed' and 'quality was computed and failed' are different facts, and waving the book's worst-quality name through on a valuation NA while 0316 is blocked on a computed False would be incoherent. The spec is genuinely ambiguous and should be tightened (systemic finding 3). Declined on merit independently: Entry_Distance_ATR -1.77 would have queued it anyway, and the 4-week view is also negative (-1.5%). |
| 0388 | HKEX | 416.80 | WATCH | 416.05 | 400.47 | 416.50 (+0.1%) | 416.06 (+0.0%) | 415.53 (-0.1%) | The exchange operator triggered a breakout buy but fails on both of the checks that matter: it screens as overvalued against a residual-income model for a financial (trading at roughly 2.1x estimated intrinsic value), and the breakout family it triggered on has a 33% one-week hit rate across ten graded trades in this book. Its own 12-week projection (415.53) sits BELOW today's close. Adjusted conviction 55 = 64 raw - 6 valuation - 3 scorecard, ten points below the 65 floor; Quality_Pass False independently blocks a new entry. Interim results land 19-Aug. No action. |
| 2888 | STANCHART | 236.00 | NO ACTION | 236.00 | 224.83 | 236.44 (+0.2%) | 234.61 (-0.6%) | 227.76 (-3.5%) | One profit-take tick fires - the 12-week projection 227.76 sits below 1.02x the 236.00 close - so a 30% trim is technically due, but at it banks only HKand loses the third disposal slot to the leveraged SK Hynix ETF on size by about HKIt is marginal on the merits anyway: conviction recovered from no signal to a bullish HOLD 58, RSI 63.9 is not extended, and the shares sit far above their 200-day average (236.00 vs 190.72) after a +1.6% session. Ledgered optional so the counterfactual grader prices the miss (HK-5: this book's untaken trims have tended to be right, median +2.3% at one week over 38 samples). Hold 348, stop held at 224.83 - the model's 223.61 is lower and a stop is never lowered. |
| 0700 | TENCENT | 481.40 | HOLD | 481.40 | 466.94 | 477.52 (-0.8%) | 477.53 (-0.8%) | 510.26 (+6.0%) | Hold through the print. Tencent reports Q2 on 12-Aug after the Hong Kong close - two sessions out, inside the five-session catalyst window where the rule is to trail rather than pre-trim - and the case for holding has strengthened anyway: conviction recovered from no signal on 08-07 to a bullish HOLD 63, and no profit-take tick fires because the 12-week view (510.26) clears the close by 6.0%. Stop held at 466.94; the model's 456.48 is lower and a stop is never lowered. |
| 0005 | HSBC HDGS(USD0.50) HK O'SEAS | 162.50 | HOLD | 162.50 | 159.50 | 162.06 (-0.3%) | 161.55 (-0.6%) | 163.43 (+0.6%) | Stays retired to a watch line at 159.50. The dead-12-week tick fires again (163.43 below 1.02x the 162.50 close) but re-issuing a sell on a name already retired is exactly what HK-11 forbids, and the arm rests only on T12_Progress 99.43, which is above 90 on every held row in this book and therefore carries no information. The shares rose 0.8% on a bullish HOLD 51. IMPORTANT: HSBC goes ex-dividend on 14-Aug (second interim, record 14-Aug, pay 25-Sep) - the mechanical gap is NOT a stop break. Model stop 156.24 is below the 159.50 watch line; the higher line stands. |
| 2513 | KNOWLEDGE ATLAS | 1252.00 | HOLD | 1252.00 | 862.00 | 1235.51 (-1.3%) | 1175.38 (-6.1%) | 1215.46 (-2.9%) | Stays retired to a watch line at 862.00 and the model agrees - AVOID at conviction 60 with 0.00 and no Sell_Rank, after a quiet +0.5% session to 1252.00. One profit-take tick fires (12-week 1215.46 below the close) but it is the known degenerate-projection artifact affecting the whole book. Two reasons to stand down: the Hang Seng quarterly review announcement (14-21 Aug) is inside the catalyst window with this name a named inclusion contender, and the model's Stop_Price 1591.67 prints 27% ABOVE the close, which is HK-12 degenerate and rejected. Against that, the US$4bn share sale remains a real dilution overhang - hold one lot, never add. Position -23.4% from a 1635.00 cost. |
| 0522 | ASMPT | 167.50 | HOLD | 167.50 | 145.50 | 169.15 (+1.0%) | 185.39 (+10.7%) | 191.82 (+14.5%) | Declined on merit, and it is the one held name where the profit-take machinery behaves correctly: ASMPT is 14.7% underwater (Gain_ATR -2.53, T12_Progress 87.32), so it is the only held row below the arming line and the only one NOT flagged PROFIT_ARMED - which is precisely why the flag firing on the other ten carries no information. The forward view is the strongest in the book: 4-week +10.7% (185.39) and 12-week +14.5% (191.82) against a 167.50 close, after a +3.0% session. Hold 1,100, stop 145.50. |
| 1347 | Hua Hong Semiconductor | 137.50 | HOLD | 137.50 | 118.00 | 135.56 (-1.4%) | 147.84 (+7.5%) | 140.99 (+2.5%) | Declined on merit for a second session, not displaced. The only profit-take tick that fires is repeated NO TRADE; the 12-week projection 140.99 clears 1.02x the close, the 4-week view is +7.5% (147.84 vs 137.50), and the shares remain far above their 200-day average (137.50 vs 114.55) despite a -2.6% session and a wide 144.00-135.10 intraday range. Trimming a name into a rising forward view on a label-absence tick alone is the error this book made with the 07-15 re-entry. Hold 3,000, stop 118.00. Position -14.1% from a 160.00 cost. |
Outcome & track record
Accuracy at the time of this record.
152 Graded decisions Past recommendations old enough to score against what actually happened (target hit / stopped out / closed). This is the evidence base — the bigger the number, the more history behind the calls. Click to inspect every one. See the decisions →
-0.249 1-week calibration How much of the eventual 4-week move typically shows up in the first week (0 = almost none, 1 = most of it). Use it to read the 1-week estimates: a low number means prices drift toward the target slowly, so don't over-read a soft week-1.
49 Published records How many dated end-of-day verdicts exist for this book — the length of the public track record you can browse in the Archive below.