ASYMMETRY FIELD NOTES · EXPERIMENT 001
The HK IPO
Blind Box.
I researched the company.
The allocation algorithm researched my patience.
One human. One AI. Six serious questions.
An unreasonable amount of joy about possibly receiving nothing.
An investment committee walks into a toy shop.
It began with IPO notices, company research, and increasingly elaborate conversations about robots. Soon we had six lenses, a LABUBU on the committee, and a theory about lucky stock-code digits. The company name occasionally achieved escape velocity before its business model did.
This experiment keeps the jokes and puts the predictions on the record. We make a dated guess, seal it, and let the first three trading days answer. Every miss stays visible. Personal accounts and the private research machinery stay behind the curtain.
“We researched like an investment committee.
We waited like children outside a toy shop.”
Curiosity, with a checklist.
Timing
Why this industry, now?
Position
What advantage can survive competition?
People
Who runs it, and whose interests come first?
Proof
Show the product, customers and cash.
Valuation
A small share price can hide a large price tag.
Market setup
Who can buy, who can sell, and when?
Names, lucky 6s and 8s, and cheap-looking tickets belong in the hypothesis notebook. They have not earned a predictive edge.
Will the box pay for its own confetti?
Robin requested one button: if allotted, do we predict a +10% limit sell within the first five trading sessions? The answer is YES or NO. The allocation algorithm already supplies enough rejection.
YES is a frozen forecast, not a promised fill. Default participation is SUBSCRIBE unless the registered screen finds a serious reason to skip. A NO forecast alone does not veto a box. This separate layer has no demonstrated forward edge yet.
Timetable update · 20 October 2026: DBS lists a 22 October, noon HKT subscription deadline for Microintelligence (03275) and Dongshan Precision (02384). The HKEX public deadline remains 23 October, noon HKT (White Form applications: 11:30). The earlier date is a DBS broker deadline, not a verified IBKR cutoff; IBKR availability and cutoff remain unknown. Frozen forecasts and conditional sell targets are unchanged. DBS timetable ↗
Shenzhen Kinwong Electronic Co., Ltd.
SUBSCRIBE · simulated participation
Conditional sell limit: HK$76.90
Based on HK$69.88 offer terms; recompute from the final issue price, rounding up to a valid tick.
The sealed receipt
Frozen 2026-09-23T03:33:00.620845+00:00
Version binary-exec-v1.0
Expected listing 2026-09-29
7773d51fc34515f59b624211581f3a6417a68c95274e9ef4624af37be7fcc9f1Issuer prospectus ↗Red Avenue New Materials Group Co., Ltd.
SUBSCRIBE · simulated participation
Conditional sell limit: HK$42.90–48.40
Based on HK$39.00–44.00 offer terms; recompute from the final issue price, rounding up to a valid tick.
The sealed receipt
Frozen 2026-09-23T03:33:00.620845+00:00
Version binary-exec-v1.0
Expected listing 2026-09-29
4860dd64d72bdd30a2061e30db471d9d4aca040c35d3105f1befc62cbbfb6062Issuer prospectus ↗Direct Drive Tech Limited
SUBSCRIBE · simulated participation
Conditional sell limit: HK$23.76
Based on HK$21.60 offer terms; recompute from the final issue price, rounding up to a valid tick.
The sealed receipt
Frozen 2026-09-23T03:33:00.620845+00:00
Version binary-exec-v1.0
Expected listing 2026-09-29
4fc922e6e0b2bf57eab70fcdf6a407894a4cc2a54cc2e4b3e97fcc1dda6e8393Issuer prospectus ↗RoboTechnik Intelligent Technology Co., Ltd.
SUBSCRIBE · simulated participation
Conditional sell limit: HK$479.60
Based on HK$436.00 offer terms; recompute from the final issue price, rounding up to a valid tick.
The sealed receipt
Frozen 2026-09-23T03:33:00.620845+00:00
Version binary-exec-v1.0
Expected listing 2026-09-29
b33c322671d664fade57699af870f45b9c4878ad97b3e39de35f05c757ee0e33Issuer prospectus ↗Shenzhen Camsense Technologies Co., Ltd.
SUBSCRIBE · simulated participation
Conditional sell limit: HK$64.75
Based on HK$58.85 offer terms; recompute from the final issue price, rounding up to a valid tick.
The sealed receipt
Frozen 2026-09-23T03:33:00.620845+00:00
Version binary-exec-v1.0
Expected listing 2026-09-30
83c1a5f212ce2e5c43da9ed62856320327a710de47ccf3cc11c9b14cfc32d42dIssuer prospectus ↗Beijing ESWIN Computing Technology Co., Ltd.
SUBSCRIBE · simulated participation
Conditional sell limit: HK$1.63–1.75
Based on HK$1.48–1.59 offer terms; recompute from the final issue price, rounding up to a valid tick.
The sealed receipt
Frozen 2026-09-28T01:34:00.165348+00:00
Version binary-exec-v1.0
Expected listing 2026-10-09
27799398786e865fd3c381ce649aec04d3edac7ceabde64ea62d9d025c852871Issuer prospectus ↗Shenzhen Transsion Holdings Co., Ltd.
SUBSCRIBE · simulated participation
Conditional sell limit: HK$38.84–42.68
Based on HK$35.30–38.80 offer terms; recompute from the final issue price, rounding up to a valid tick.
The sealed receipt
Frozen 2026-10-07T02:17:57.839681+00:00
Version binary-exec-v1.0
Expected listing 2026-10-15
7b6079e0053259bdfed0405f6a86a60c5f7d1b9ab4d50020cfe96916abd9b49aIssuer prospectus ↗Jiangsu Slender West Lake Culture and Tourism Co., Ltd.
SUBSCRIBE · simulated participation
Conditional sell limit: HK$4.315–5.58
Based on HK$3.92–5.07 offer terms; recompute from the final issue price, rounding up to a valid tick.
The sealed receipt
Frozen 2026-10-14T01:33:00.552133+00:00
Version binary-exec-v1.0
Expected listing 2026-10-23
b71c6d23bc0aedf87a8b3353ee4c382020ac8acc44d233ab5f73ae0f9fb9d6c2Issuer prospectus ↗Changzhou Microintelligence Co., Ltd
SUBSCRIBE · simulated participation
Conditional sell limit: HK$18.30–23.62
Based on HK$16.63–21.46 offer terms; recompute from the final issue price, rounding up to a valid tick.
The sealed receipt
Frozen 2026-10-20T01:19:35.145023+00:00
Version binary-exec-v1.0
Expected listing 2026-10-28
cac448f6415fe16b847f8072ea3bb9dbc666feb747c7690ee40123689541af60Issuer prospectus ↗Suzhou Dongshan Precision Manufacturing Co., Ltd.
SUBSCRIBE · simulated participation
Conditional sell limit: HK$103.20
Based on HK$93.80 offer terms; recompute from the final issue price, rounding up to a valid tick.
The sealed receipt
Frozen 2026-10-20T01:19:49.467155+00:00
Version binary-exec-v1.0
Expected listing 2026-10-28
1d4a2644fb8bdfd39ebb35d444dbfb95c059cb7e2cbe0b468360cf037b806f65Issuer prospectus ↗The exit rule: keep the +10% limit through sessions 1–5. From session 6, use an OCO +10% target / −10% stop. A gap below the stop can lose more than 10%. A price touching a limit does not prove an order filled.
Forward evidence: 3 YES outcomes, 3 NO outcomes and 4 still pending across 10 frozen calls. 7 paths have completed-session evidence. An early target hit can resolve YES; a first-day miss cannot resolve NO. This small, partly observed sample does not establish a forecasting edge.
Single-vendor completed daily bars; official final issue prices and independent opening-price checks. Market counterfactuals, not personal fills. The original D3 results are reported separately below.
“The lottery handles rejection.
We handle the receipts.”
The five-day rehearsal, with the awkward bits left in
The 2024-09-23–2026-09-22 inventory contains 249 listings. Under the declared daily-bar execution assumptions, 177 of 248 resolved five-day labels reached the +10% limit in time; 71 did not. One recent listing is still pending. This is historical reconstruction, not unseen prediction or a personal trading win rate.
The binary prediction recipe scored 70.2%, versus 71.4% for always saying YES. The fixed screen excluded no IPOs, so it selected the same portfolio as subscribing to all. No screening advantage has been demonstrated. There are 241 modeled exits, 6 unresolved execution histories, and 2 known open positions. Missing evidence stays missing; we do not publish a complete portfolio return from an incomplete ledger.
Only redacted calls and aggregate evidence appear here. Private model details, capital assumptions and personal logs remain private.
The next boxes.
These are forecasts for the highest traded price within the first three trading days, relative to the final issue price. The initial probabilities are uncalibrated research priors. Matured results are descriptive; this small sample cannot establish calibration or a forecasting edge.
3 correct return bands out of 7 matured D3 forecasts (42.9%). Coverage: 7/10 frozen calls scored; the remainder have not completed three sessions. 95% Wilson interval: 15.8%–75.0%. Multiclass Brier: 0.713. Completed daily highs use single-vendor evidence with archived official issue prices and independent opening/date checks. These are market-price labels, not subscription profits or fills. Frozen calls and models remain unchanged.
Open the original three-session research (unchanged)
Shenzhen Kinwong Electronic Co., Ltd.
Expected listing ·
D3 return-band forecast: missed. Comparison target touched; touching is not a fill.
Red Avenue New Materials Group Co., Ltd.
Expected listing ·
D3 return-band forecast: missed. Comparison target not touched; touching is not a fill.
Direct Drive Tech Limited
Expected listing ·
D3 return-band forecast: correct. Comparison target touched; touching is not a fill.
RoboTechnik Intelligent Technology Co., Ltd.
Expected listing ·
D3 return-band forecast: missed. Comparison target not touched; touching is not a fill.
Shenzhen Camsense Technologies Co., Ltd.
Expected listing ·
D3 return-band forecast: missed. Comparison target touched; touching is not a fill.
Beijing ESWIN Computing Technology Co., Ltd.
Expected listing ·
D3 return-band forecast: correct. Comparison target not touched; touching is not a fill.
Shenzhen Transsion Holdings Co., Ltd.
Expected listing ·
D3 return-band forecast: correct. Comparison target not touched; touching is not a fill.
Jiangsu Slender West Lake Culture and Tourism Co., Ltd.
Expected listing ·
Changzhou Microintelligence Co., Ltd
Expected listing ·
Suzhou Dongshan Precision Manufacturing Co., Ltd.
Expected listing ·
History has already read the ending.
The 24-month reconstruction runs from 2024-09-21 to 2026-09-20. This is the original three-day price-band test, separate from the five-day +10% question above. It uses dated offering evidence and chronological evaluation; historical reconstruction is not unseen forward performance.
44 right out of 79. That is what 55.7% means: the retained historical baseline correctly predicted the return band of the first three sessions’ highest traded price. It does not mean that 55.7% of subscriptions made money, received shares, or filled a +10% sell order.
Assessment 95% interval: 44.7%–66.1%. Multiclass Brier score: 0.621 (lower is better). The baseline survived the predeclared validation gate. A more complicated story has to earn its place.
What this score means
The four bands are below 0%, 0–<20%, 20–<40%, and 40% or more, measured against the final issue price. Predicting 0–<20% when the three-day high was +12% counts as correct; predicting 20–<40% does not. All 247 historical three-day outcomes have now been recovered. The 0 here counts missing D3 labels only; it does not erase the separate five-day/OCO ledger’s 6 unresolved execution histories. Direction, probability error and target touches have separate scores.
238 vendor series, six exchange-derived additions, and three independently corroborated public high observations. Source limitations remain in the private audit. Recovered labels score existing frozen forecasts; no retroactive retraining or model reselection. Trading costs and allocation uncertainty are separate from price prediction. Quarterly reviews may introduce a new model version; they cannot revise old answers.
Keep the curiosity.
Keep the receipts.
A robot can refuse to pick us. A stock can refuse to rise. We can still leave behind an honest record of how we thought—and how much we laughed.
Bubu can guess wrong. Bubu cannot secretly edit the exam paper.