Signal Group Sentiment Dashboard
DEMO DATA Synthetic 48h data showing how Kalman filters will track group activity once the buffer goes live.
Buffer Status
PENDING
Gateway restart at 19:12 loads the buffer code. No real data yet.
Messages (48h)
1396
Synthetic — simulating ~3 msgs/hr baseline
TollGate Mentions
192
Keyword hits in synthetic data
Anomalies Detected
7
Activity spikes Kalman flagged as unusual
1. Message Velocity — Kalman Filter Tracking
The Kalman filter smooths noisy 15-min samples into a stable velocity estimate.
The filter adapts: when activity spikes, gain increases and the filter responds faster.
Green dots = anomalies (activity exceeded 2-sigma above predicted).
2. Kalman Gain — Filter Responsiveness
High gain = filter trusts new observations more (during spikes). Low gain = stable periods where the filter trusts its own prediction.
3. TollGate Keyword Mentions
Counts of tollgate/bitcoin/lightning/mesh/esp32 keywords per 15-min window. Overlaid with activity baseline to see if discussions are TollGate-relevant.
4. Unique Senders
Approximate unique participants per window. More senders = broader discussion, not just two people arguing.
What Works / What Doesn't
✅ WORKS
- Admin sender filter — only you can command the bot in groups
- Message buffer — non-admin messages stored to SQLite with hashed senders
- Deduplication — UNIQUE index prevents duplicate messages
- Anonymization — SHA-256 hashes, no real names stored
- Summarizer scripts — tested and working
- Cron jobs — harvester (15min), summarizer (2h) scheduled
⚠️ PENDING (waiting for restart)
- Buffer code loading — gateway restart at 19:12 activates the SQLite buffer
- First real data collection — starts immediately after restart
❌ NOT BUILT YET
- Real Kalman filter on live data — this demo uses synthetic data
- Automated alerts — cron alerts you, but Kalman-based alerting needs building
- Multi-group comparison — currently single-group demo
- Trend prediction — Kalman estimates current state, but forecasting needs ARIMA or similar