█████                        
 ░░███                         
 ███████    ██████   █████ ████
░░░███░    ░░░░░███ ░░███ ░███ 
  ░███      ███████  ░███ ░███ 
  ░███ ███ ███░░███  ░███ ░███ 
  ░░█████ ░░████████ ░░████████
   ░░░░░   ░░░░░░░░   ░░░░░░░░ 

Tau is a temporal database built around a single invariant:

Data is never mutated — only extended through time.

Rather than overwriting state, Tau stores deltas that are valid over explicit time ranges. This makes history, auditing, and point-in-time reconstruction intrinsic to the model instead of bolt-on features. Tau is designed for domains where correctness over time matters:

• Time-series and sensor data
• Financial systems and audit trails
• Simulations and historical modelling
• Systems requiring deterministic replay

• Tau — an immutable delta valid for a specific time range
• Schedule — an ordered collection of related Taus
• Frame — a grouping of schedules that belong together
• Lens — a transformation that acts as both a view and a migration

These primitives allow Tau to represent evolving state without ever losing information.

Traditional databases force trade-offs between:

• Write performance vs historical queries
• Storage efficiency vs duplication
• Flexibility vs auditability

Tau avoids these trade-offs by treating time as a first-class dimension and immutability as a hard constraint.

Tau is early-stage and research-driven.

The focus is on:

• Correctness and clear invariants
• Minimal, explicit abstractions
• Deterministic behaviour
• Long-term evolvability

APIs and on-disk formats may change.

Tau is opinionated by design:

• Immutability over convenience
• Explicit time over implicit state
• Simple primitives over complex magic