timedb

TL;DR

timedb is a opinionated schema and API built on top of PostgreSQL design to handle overlapping time series revisions and auditable human-in-the-loop updates.

Most time series systems assume a single immutable value per timestamp. timedb is built for domains where data is revised, forecasted, reviewed, and corrected over time.

timedb lets you:

• ⏱️ Retain "time-of-knowledge" history through a three-dimensional time series data model;
• ✍️ Make versioned ad-hoc updates to the time series data with annotations and tags; and
• 🔀 Represent both timestamp and time-interval time series simultaneously.

Why timedb?

Most time series systems assume:

• one value per timestamp;
• immutable historical data; and
• no distinction between when something was true vs when it was known.

This pattern is a major drawback in situations such as:

• forecasting, where multiple forecast revisions predicts the same timestamp;
• backtesting, where "time-of-knowledge" history is required by algorithms;
• data communication, where and auditable history of updates is required.
• Human review and correction, where values are manually adjusted, annotated, or validated over time
• Late-arriving data and backfills, where new information must be incorporated without rewriting history

In practice, teams work around these limitations by overwriting data, duplicating tables and columns, or encoding semantics in column names — making systems fragile, opaque, and hard to reason about.

timedb addresses this by making revisions, provenance, and temporal semantics explicit in the data model, rather than treating them as edge cases.

Installation

pip install timedb

Basic usage

TBD

Tables

runs_table

Field	Type	Purpose
run_id (primary key)	attribute	Unique identifier for the run (generated by the API)
workflow_id	attribute	Identifier for the workflow that produced this run
run_start_time	time dimension	When the workflow started
run_finish_time (optional)	time dimension	When the workflow finished
run_params (optional)	attribute	Parameters/configuration used for this run (JSON object)
inserted_at	time dimension	When the row was inserted (default now())

---

values_table

Field	Type	Purpose
value_id (primary key)	attribute	Unique identifier for each version of a value
run_id (foreign key)	attribute	References the run that produced this value (runs_table.run_id)
valid_time	time dimension	Timestamp the value is valid for
valid_time_end (optional)	time dimension	Optional interval end time; NULL means point-in-time at valid_time
value (optional)	measure	The numeric value (nullable; NULL can be a valid stored value)
annotation (optional)	attribute	Optional human annotation (whitespace-only disallowed)
tags (optional)	attribute	Optional semantic labels / quality flags (empty arrays disallowed; use NULL)
changed_by (optional)	attribute	User or service responsible for the change
change_time	time dimension	When this version row was created (default now())
is_current	attribute	Whether this row is the active version for its key (default true)

---

metadata_table

Field	Type	Purpose
metadata_id (primary key)	attribute	Surrogate primary key for metadata rows
run_id (foreign key)	attribute	References run context (runs_table.run_id)
valid_time	time dimension	Time context for the metadata (joins onto values via (run_id, valid_time))
metadata_key	attribute	Name of the metadata field (e.g. contractId, deliveryStart)
value_number (optional)	attribute	Numeric metadata value (exactly one typed value must be set per row)
value_string (optional)	attribute	String metadata value (mutually exclusive with other typed values)
value_bool (optional)	attribute	Boolean metadata value (mutually exclusive with other typed values)
value_time (optional)	attribute	Timestamp metadata value (mutually exclusive with other typed values)
value_json (optional)	attribute	JSON metadata value (mutually exclusive with other typed values)
inserted_at	time dimension	When the metadata row was inserted (default now())

Changed

Three-dimensional time series data model

Every time series value is described using three independent timelines:

Time dimension	Description
knowledge_time	The time when the value was known
valid_time	The time the value represents a fact for
change_time	The time when the value was changed

Additional attributes

Schema columns provides additional attributes to the values according to:

Column name	Description
value	The numeric value being stored (may be NULL)
tags	Semantic labels and quality flags applied to the value
annotation	Optional human annotation explaining the value or change
run_id	Reference to the workflow run that produced the value
run_params	Parameters and configuration associated with the producing run
is_current	Indicates whether this row is the active version for its key
changed_by	User or service responsible for the change

Roadmap

• Decouple the knowledge time from the run_time
• Python SDK that allows time series data manipulations, reads and writes
• RESTful API layer that serves data to users
• Handle different time zones in the API layer while always storing in UTC in the database.
• Support for postgres time intervals (tsrange/tstzrange)
• Built in data retention, TTL, and archiving
• Support for subscribing to database updates through the API
• Unit handling (e.g. MW, kW)