Data Collection
Sources
Repute AI collects data exclusively from public Telegram channels and public creator activity. No private data, private groups, DMs, or personal account information is accessed at any point.
What Is Collected
| Data Type | Description | Source |
|---|---|---|
| Subscriber count | Total members at time of snapshot | Public channel metadata |
| Join/leave delta | Net membership change per time window | Public channel metadata |
| Message frequency | Count of messages per time window | Public channel |
| View counts | Views per message | Public channel |
| Reaction distribution | Count per reaction type | Public channel |
| Forward counts | Number of forwards per message | Public channel |
| Reply counts | Total replies per message (if replies enabled) | Public channel |
| Creator overlap | Cross-references between channels via shared content | Public channel + forward graph |
What Is NOT Collected
- Private message content
- Private group activity
- Personal account message history
- Phone numbers or contact information
- Any data from private or invite-only channels
- Device identifiers or location data
Repute AI operates exclusively on the Telegram Bot API’s public data access layer. No credentials that would allow private channel access are used or stored. The system is architecturally limited to public data.
Data Freshness and Update Cadence
Repute AI processes data at multiple cadences depending on signal type:
| Signal Type | Update Cadence |
|---|---|
| Subscriber count | Every 6 hours |
| View and reaction data | Every 4 hours |
| Message frequency | Real-time ingestion, aggregated hourly |
| Join/leave delta | Every 2 hours |
| Creator graph | Daily full refresh + real-time event triggers |
| Model score outputs (AVI, SIS, NSM, BPM, TRI) | Recomputed every 12 hours |
| REP composite score | Recomputed every 24 hours |
| On-chain REP token metadata | Updated on REP recompute if delta > threshold |
Data Retention
Repute AI retains historical signal data to enable trend analysis and temporal weighting in model computation:
- Raw signal data: 90-day rolling window
- Aggregated model scores: 12-month history
- Creator graph snapshots: 6-month history
- REP score history: Indefinite (on-chain record)
No personally identifiable information is retained beyond the Telegram user ID and public display name of channel owners, collected only when a user explicitly connects their channel to the protocol.
Privacy Architecture
The data collection system is designed with privacy as a structural constraint, not a policy:
- Public-only access: The bot operates via Telegram’s public API. Private channel data is architecturally inaccessible.
- No content storage: Message text content is processed for frequency and volume signals but is never stored. Only metadata is retained.
- Hashed identifiers: Telegram user IDs used in the REP soulbound token metadata are stored as one-way hashes, not raw IDs.
- No cross-platform tracking: Repute does not correlate Telegram identity with activity on other platforms.