Every tick is a learning opportunity. Every pattern has a rhythm. Temporal intelligence that never forgets.
Part of the [&] Protocol stack · Ecosystem overview · Three-protocol stack · Stack status
Welcome to the documentation hub for TickTickClock — the temporal intelligence layer for the [&] Protocol ecosystem. TickTickClock provides time-series anomaly detection, forecasting, and pattern recognition as continual learning services.
TickTickClock is the &time primitive provider, exposing &time.anomaly, &time.forecast, and &time.pattern capability contracts. It does not store business data, host agents, or manage workflows — it ingests temporal signals, learns their structure continuously, and returns intelligence to upstream consumers.
Time-series intelligence in production systems is fragmented. Anomaly detection runs as a batch job. Forecasting lives in a notebook. Pattern detection is a dashboard someone checks on Mondays. None of these systems learn continuously, and none of them talk to each other.
Meanwhile, underlying models suffer from catastrophic forgetting — retrained monthly on a window of recent data, discarding everything about long-range seasonality. Multi-timescale temporal behavior requires multi-timescale memory, and no production system provides it.
TickTickClock solves this with:
Continual learning — models update at inference time, no batch retraining
Multi-timescale memory — fast (seconds), medium (hours), slow (days), glacial (months)
Streaming-first — designed for unbounded streams, not fixed datasets
Edge-capable — SQLite + Bumblebee on constrained devices
Homepages
Root Docs
TickTickClock Docs
| Capability | Operations | Description |
|---|---|---|
&time.anomaly | detect, enrich, learn | Adaptive anomaly detection via selective state-space models |
&time.forecast | predict, explain, enrich, learn | Multi-timescale forecasting with continual learning |
&time.pattern | detect, summarize, enrich | Cycle and motif detection via spectral analysis |
| Tool | Description |
|---|---|
anomaly_detect | Run anomaly detection on a temporal stream |
anomaly_subscribe | Subscribe to real-time anomaly events |
forecast_predict | Generate forecasts with confidence intervals |
forecast_explain | Explain forecast reasoning |
pattern_detect | Detect cycles and motifs in time series |
pattern_summarize | Summarize detected patterns |
temporal_enrich | Enrich external data with temporal context |
temporal_learn | Submit outcome data for model updates |
| Component | Role | OTP Pattern |
|---|---|---|
| Anomaly Engine | Adaptive SSM (Mamba-style) with selective state gating | GenServer per stream |
| Forecast Engine | Multi-timescale CCO prediction, no forgetting | GenServer + Nx |
| Pattern Engine | Spectral + autocorrelation cycle detection | Concurrent tasks |
| Multi-Timescale Store | Fast (ETS) → Medium → Slow → Glacial memory | Tiered storage |
| Stream Dashboard | Anomaly timeline, forecast explorer, pattern viewer | Phoenix LiveView |
| Tier | Latency | Storage | Contents |
|---|---|---|---|
| Fast | seconds | ETS | Per-tick state snapshots |
| Medium | hours | SQLite/TimescaleDB | Aggregated state vectors |
| Slow | days | SQLite/TimescaleDB | Consolidated summaries |
| Glacial | months | SQLite/TimescaleDB | Long-range seasonal models |
| Product | Relationship |
|---|---|
| GeoFleetic | Spatial intelligence — when + where = complete situational awareness |
| Graphonomous | Continual learning substrate for temporal knowledge |
| Delegatic | Governance enforcement for temporal operations |
| Layer | Technology |
|---|---|
| Language | Elixir 1.17+ |
| Framework | Phoenix 1.8+ |
| Database | PostgreSQL 16+ + TimescaleDB |
| Edge Storage | SQLite |
| ML | Bumblebee / Nx (SSM, CCO models) |
| Replication | Epoch-aware delta-CRDTs |
Spec: Technical Specification
[&] Protocol ecosystem: AmpersandBoxDesign/
[&] Ampersand Box Design — ticktickclock.com