Turning Legend IP Announcements into Risk Ledgers

I compared manually transcribing announcements with using Tencent Cloud AI to extract fields, and actually ran it through once. The trigger was an announcement from Kingnet on the evening of September 11th, stating their intention to acquire a 49% stake in Neosphere Co., Limited through their wholly-owned Hong Kong subsidiary, Cypress Technology HK Limited. For beginners, every word in this paragraph sounds like a tongue twister. IP stands for intellectual property; here it can be understood as whether the characters, rules, names, etc., related to the "Legend" game can be legally adapted and licensed. A 49% equity stake means holding nearly half the shares, which doesn't necessarily equal full control—it depends on the agreement and board arrangements.

Having done enterprise solutions for a long time, my first reaction was to ask if this announcement could be turned into a queryable table. Technical feasibility is no problem, and client willingness to pay is obvious. Legal, investment, and publishing teams all want to know what problems this deal solves and what risks remain. This tutorial breaks down one announcement into a risk ledger.

First, prepare the original text. Open Feishu (Lark), click "New Document," and enter "Legend IP Announcement Breakdown" as the title. Paste the original announcement or news summary into it. As long as you see complete text in the white editing area, you're good. The expected result is that you can select all and copy.

Next, create the table. Open Excel and click "Blank Workbook." Enter six columns in the first row: Entity, Action, Ratio, Basis, Risk, To Confirm. In plain language: Entity is who is doing it, Action is what they are trying to do, Ratio involves how much equity, Basis is what the announcement or court said, Risk is what might go wrong, and To Confirm is who to ask next.

Then, use AI for extraction. I've been testing Tencent Cloud AI's dialogue or document understanding entry points these past few days. Paste the announcement into the input box, then enter the prompt: "Only extract information explicitly stated in the announcement, do not infer. Output a table with six columns: Entity, Action, Ratio, Basis, Risk, To Confirm." Click send. The expected result is that the AI returns a draft table.

Here's where people stumble. Models tend to write "intends to acquire" as "has acquired," and interpret a 49% equity stake as "controlling interest." My first run came out exactly like that. The solution is to add a line to the prompt: "Whenever 'intends,' 'plans,' or 'in progress' appears, the status must be written as undetermined; do not judge control rights." If the table gets messy, enter "Output only the table."

Finally, manually supplement the basis. Don't just write "there is a dispute." According to public reports, the Legend IP dispute has lasted for years. At the end of 2021, the Shanghai No. 1 Intermediate People's Court rejected all litigation claims by the Legend IP company, recognizing that Zhejiang Huanyou bears independent responsibility for debts with all its assets. Put this section in the "Basis" column. The Risk column can say "Whether the new transaction structure affects existing long-term licenses." The To Confirm column should say "Ask legal to confirm if Shengqu and Kingnet's existing licensing agreements are still valid."

After running it, I did a simple comparison.

Method	Time Spent	Output	Main Issues
Manual Transcription	~40 mins	One paragraph summary	Fields easily missed
AI Extraction	~5 mins	Six-column draft	Status and control rights need manual correction

The value of this table lies in turning vague announcements into objects that can be questioned. Clients buy clarity on liability boundaries. The next step is to try connecting this table to RAG. RAG lets the AI first check your organized materials before answering. For example, if a business colleague asks, "Which licenses does this 49% equity acquisition affect?", the system first finds the basis in the table, then provides an explanation. Technical feasibility is fine; the difficulty in implementation lies in field standardization.