Technical Specification (Draft Version)

Project: Telegram Message Analysis System for Finding the Most Important and Significant Information in the Cryptocurrency Sphere

1. System Goal

Create a system that connects to a separate working Telegram account and:

• reads all messages from specified private chats and channels;

• automatically determines which of them contain:

  • important information (important messages);

  • very important and/or "perishable" information (urgent messages);

• publishes:

  • important messages and digests in one channel;

  • urgent messages — immediately and separately, in another channel;

  • and also uses special channels Good/Bad for self-learning from examples.

2. Data Sources

• One dedicated working Telegram account (UserBot).

• It is part of a number of private chats and channels where insights, schemes, and situational methods of earning in crypto appear.

• All these chats and channels are data sources.

3. Types of Messages and Classification

Each incoming message is analyzed and classified into one of three categories:

3.1. Ordinary (Unimportant)

• Any discussions, news, reactions, chatter that do not describe a specific method of earning.

• Examples: "the market is falling", "Bitcoin is flying", "I made 200 bucks" — without explanation of how exactly.

Action: ignored (optionally saved to the database).

3.2. Important Message

Messages that contain a description, hypothesis, discussion, or hint at a specific method of earning, even if not yet verified.

Signs:

• contain phrases or action structures ("if you do X, you get Y");

• include words like "mechanics", "scheme", "method", "farm", "arbitrage", "bypass", "instruction";

• no indication that this is advertising or empty theory.

Examples:

• "There is a new mechanic on DEX X — if you deposit token A and lock it, you earn points."

• "You can arbitrage token Z between exchanges A and B."

Action: marked as important, included in the digest, published in the "Important" channel.

3.3. Urgent Message

Messages where the described method is already working right now, meaning it is not just an idea or discussion, but a fact of an active earning opportunity.

⚠️ Important:
Urgency = relevance, not just immediacy.
We are looking for signs that the scheme is already yielding results at the time of discussion, not words like "hurry" or "by the end of the day".

Signs:

• present tense is used: "works", "gives", "farms", "checked", "pays", "drops";

• results are mentioned: "gives 15% profit", "drops token X", "we are already farming";

• an active process is described, not a supposition ("it will be possible", "possibly");

• words like: "still works", "still farming", "is currently ongoing", "verified", "really drops" are encountered.

Examples:

• "On exchange A you can rotate between tokens X/Y, it is currently working — gives about 5%."

• "Farming on protocol Z is still active, token drops every 10 minutes."

Action: published immediately in the "Urgent" channel, and also included in the next digest as a priority element.

4. Telegram Channels

Four private channels are created, grouped into one folder:

5. System Operation Logic

1. Collection — the system receives all new messages from connected chats and channels.

2. Analysis — determines:

   • whether the message contains a description of a method of earning;

   • if yes, whether this method is already working now.

3. Classification:

   • NONE — ordinary → not published;

   • IMPORTANT — important → in the "Important" channel and in the digest;

   • URGENT — urgent → in the "Urgent" channel + in the digest.

4. Publication:

   • important → immediately in "Important";

   • urgent → immediately in "Urgent";

   • every 4 hours the system publishes a summary (digest) in "Important" from all sources.

5. Feedback:

   • the user forwards individual messages to "Good" or "Bad";

   • the system saves these notes for further model training.

6. Digest Format

Frequency: 6 times a day (approximately every 4 hours).

Each digest:

• is published in the "Important" channel;

• is structured by sources (chats/channels);

• includes brief summaries of all important messages for the period.

Example:

Digest 12:00–16:00

Chat: DeFi Hunters

1. Farming token X on DEX Z — works, gives 12% per day.

2. Mechanics with bridge Y — token B is credited until it is turned off.

Channel: CeFi Alpha

1. Arbitrage between Binance and OKX for token A — 4–5% per round.

—
Period: 12:00–16:00, 9 sources, 17 important messages.

7. Feedback and Self-Learning

• The user manually forwards publications from the "Important" and "Urgent" channels:

  • if it worked correctly → to Good;

  • if there is an error → to Bad.

• The system must:

  • link forwarded posts to their original messages;

  • save pairs:
    text, type (important/urgent), label (good/bad);

  • use these pairs for training and adjusting filters.