[Python] trying to create a Telegram Call and Channel analyzer for Solana Memecoins

For weeks I have been trying to create a script to analyze telegram calls and channels for Solana memecoins. The call scraping function works fine but the metrics never come our right. No matter what API source I try to use I always get errors. Here is exactly what I need if there is anyone who can help..

build a bug-free production-ready Python script that analyzes Solana memecoin calls from Telegram channels. The script should combine Telegram scraping, price history analysis, ROI metrics, ATH detection, drawdown calculations, liquidity/safety checks, and per-channel aggregation.

Please follow this exact specification:

It must:

• Use Telethon with an existing session file (not new login).
• Be asyncio-based with efficient batching and checkpointing.
• Produce complete metrics without missing values.
• Output per-call audit CSV + per-channel leaderboard CSV.

Inputs

• CSV file usernames.csv with column username (Telegram handles).
• Existing Telethon session file for login.
• Configurable options: days of history, max messages per channel, concurrency, cache toggle, database toggle.

Telegram Scraper

• Use Telethon with session string/file.
• Scrape last N days of messages per channel (configurable).
• Extract all Solana mint addresses (32–44 base58).
• For each valid mint call, collect:
  • channel_username
  • message_id
  • message_timestamp
  • message_excerpt (truncate 500 chars)
  • deep_link
  • mint_address

Deduplicate calls on (channel_username, message_id, mint_address).

Per-Call Metrics (include all of the following)

• price_at_call
• price_at_call_ts
• price_at_call_source
• ath_price
• ath_timestamp
• current_price
• roi_ath
• roi_current
• ath_time_to_peak
• market_cap_ath
• volume_24h_ath
• liquidity
• safety_score
• rugcheck_score
• tokensniffer_score

Price & ROI Engine

• Use at least FOUR APIs (Birdeye, DexScreener, GeckoTerminal - SOURMUST HAVE HISTORICAL PRICE DATA FOR NEW AND/OR LOW CAP SOLANA MEMECOINS, BE ABLE TO OBTAIN PRICE AT SPECIFIC TIME, MUST BE ABLE TO GET ATH DATA).
• Apply retries + fallback if one fails.
• Compute metrics:
  • Entry price: last 1m candle before call timestamp.
  • ROI windows: 5m, 30m, 1h, 4h, 24h after call.
  • ATH:
    • Scan candles after call timestamp.
    • Record ath_price, ath_timestamp, roi_ath, time_to_ath.
  • Drawdown: lowest price in 6h window after entry.

Safety Checks

• Query RugCheck, TokenSniffer, Ave.ai, AND ANY OTHER ADDITIONAL SOURCES, etc.
• Normalize into 0–100 safety score.
• Collect:
  • rugcheck_score
  • tokensniffer_score
  • ai_safety_score
  • lp_lock_percent
  • top10_holders_percent
• Add heuristic:
  • If drawdown_6h <= -90% → set rug_flag=1.

Liquidity & Fundamentals

• Collect (from DexScreener / Raydium / [Pump.fun] AND/OR ANY OTHER SOURCES THAT PROVIDE THE NECESSARY INFO):
  • liquidity_usd
  • 24h_volume
  • market_cap_at_call
  • market_cap_ath

Per-Channel Aggregation

For each channel:

• total_calls
• successful_calls
• avg_roi_1h, avg_roi_24h, avg_roi_ath
• median_roi_1h, median_roi_24h, median_roi_ath
• win_rates (% reaching >1x, 2x, 5x, 10x ATH)
• rug_rate (% flagged rugs)
• avg_safety_score
• avg_time_to_ath
• avg_ath_market_ca
• calls_with_ath
• avg_roi_current
• ath_win_rate
• two_x_ath_rate
• five_x_ath_rate
• ten_x_ath_rate
• avg_time_to_ath
• survivorship_rate
• avg_market_cap_ath
• consistency_score
• outlier_dependence
• discrepancy_score

Outputs

1. Per-call CSV (per_call_audit.csv)
   • All raw + derived metrics per mint call.
2. Per-channel CSV (per_channel_leaderboard.csv)
   • Aggregated stats per channel.
3. Log file (analyzer.log)
   • Errors, retries, skipped calls.
4. Cache file (cache.json)
   • Prevent duplicate analysis of the same mint.

Robustness

• Retries (3×, exponential backoff).
• Fallback between APIs.
• Skip bad data gracefully (never crash).
• Detect and filter invalid data:
  • 
    ROI > 10,000× → ignore.
    
    
  • 
    Market cap > $1B → ignore.
    
    
  • 
    Checkpointing for resume support.
    
    
  • 
    SQLite or other DB backend optional, but CSV export required.
    

Deliverable

• Single Python script using asyncio.
• Runs with python script.py.
• Produces 2 CSVs every run with no missing data.
• Well-commented and easy to maintain.

Please generate the complete working script following this exact spec.


Problems Encountered

1. No Rug/Safety Info

• RugCheck and TokenSniffer often return None or partial data.
• APIs sometimes change endpoints or rate-limit heavily.
• Your pipeline falls back poorly: if RugCheck fails, it doesn’t retry another provider.
• Safety scores are often empty → rug rate and avg safety become useless.

2. Inaccurate ROI & Prices

• Price-at-call is unreliable:
  • Wrong candle resolution (using 1h instead of 1m).
  • Inconsistent timezones / unix timestamps.
  • Sometimes takes the nearest future candle instead of last closed before the call.
• Current prices overwrite entry prices, so ROI is skewed.
• ATH often equals current price because historical peak isn’t calculated correctly.

3. Missing ATH Data

• ATH fetches from DexScreener/Birdeye/GeckoTerminal are inconsistent.
• If one fails, script doesn’t aggregate others.
• ATH timestamp not properly aligned → “time-to-ATH” is missing or negative.

4. No Price at Call Time

• Many calls end up with price_at_call=None.
• Missing entry price → ROI is invalid or set to 0.
• Candlestick fetching window too narrow (±5 min) → often misses relevant data.

5. Channel Summary CSV Incomplete

• Some metrics (win rates, rug rate, time-to-ATH) are missing or always None.
• Survivorship calculation is inconsistent (sometimes divides by 0).
• Outlier filtering is weak → unrealistic ROI (10000x+) corrupts averages.

6. Other Issues

• Concurrency not tuned → occasional rate limits, skipped channels.
• Cache system doesn’t de-dupe well; duplicate mint calls are analyzed separately.
• Error handling inconsistent: some exceptions are logged, others silently skipped.
• No clear backup/fallback strategy for each metric.

I tried scripts from chatgpt, grok, copilot, deep seek, claude.. I was expecting csv output files with the metrics previously mentioned.

Continue reading...