the data foundation you plug into an agentic stack to give it EV charging awareness out of the box here
kwwhat is an open-source data pipeline that transforms raw OCPP logs into a structured, queryable context layer for EV charging. It models charge attempts, visits, uptime, and session outcomes — so whatever agentic use case you are building has reliable, grounded domain knowledge from day one.
This project is designed for AI engineers, data practitioners, and CSMS providers building support agents, analytics agents, or agentic ops workflows who want to give their agents a deep understanding of charger behavior.
- Source modeling for raw OCPP logs (StatusNotification, Heartbeat, Start/StopTransaction)
- Charge attempt and visit models
- Interval data for other reporting use cases that require energy delivery by 15-min time slices
- Core metrics:
| Metric | Description |
|---|---|
uptime |
Average fraction of commissioned time a Port was online and not in a Faulted state |
total_visits |
Total number of charging visits |
total_charge_attempts |
Total charge attempts across all visits |
first_attempt_success |
Count of visits where the first attempt succeeded |
troubled_success |
Count of visits that succeeded but required more than one attempt |
failed_visits |
Count of visits that did not end in a successful charge |
first_attempt_success_rate |
Proportion of visits where the first attempt succeeded |
troubled_success_rate |
Proportion of visits that were troubled success |
failed_rate |
Proportion of visits that failed |
average_attempts_per_visit |
Total charge attempts divided by total visits |
The fastest way to see what kwwhat can do is the self-contained Chat BI demo — ask plain-English questions about EV charger reliability and get answers grounded in real OCPP data. No cloud account needed, runs locally. Or explore a live Tableau dashboard if you want to see the metrics without running anything.
| Term | Definition | Sanity check |
|---|---|---|
| Charger (Charging Station) | Physical system where EVs can be charged. Has one or more Ports (EVSEs). | What a driver perceives as a single charger |
| Port (EVSE) | Independently operated and managed part of a Charger that can deliver energy to one EV at a time. | "How many vehicles can charge simultaneously?" = number of Ports |
| Connector | Independently operated electrical outlet on a Port. A Port may have multiple Connectors (socket types or tethered cables) to support different vehicle types (e.g. four-wheeled EVs and electric scooters). | "What vehicle types can charge here?" = Connector types |
Reliability and utilisation metrics are tracked at Port grain.
Source: OCPP 2.1 Edition 1 — © Open Charge Alliance 2025, Definitions
A charge attempt captures intent to charge on a Port. It begins when bay occupancy changes or a driver plugs in, and ends when the connector is no longer performing any charging related tasks. If all prerequisites align, a charge attempt wraps around a transaction. A charge attempt may or may not result in energy transfer — if the transaction was successful, there was meaningful energy transfer (more than 0.1 kWh).
A visit is a single driver trip to a charging location. It may span multiple charge attempts — for example, when a driver unplugs and retries on the same or a nearby port.
Operational models reconstruct individual charge attempts from OCPP logs. But a driver's experience at a charging site is rarely a single attempt. They park, plug in, face an error, unplug, try another port, eventually get a charge. Visit modeling turns that sequence of technical events into a single, coherent unit of analysis: one driver, one stop, one outcome.
Two grouping strategies are used depending on whether the driver successfully authorised.
For authenticated drivers, all charge attempts by the same driver at the same location within a 30-minute window belong to the same visit. A gap of 30 or more minutes, or a different location, starts a new visit.
For unauthenticated drivers, attempts on the same port within a 2-minute window belong to the same visit. Different ports always start a new visit. When an anonymous attempt immediately precedes an authorized attempt on the same port within 2 minutes, the driver identity is inferred retroactively.
This model unlocks four driver-centric metrics: first attempt success rate (charging worked on the first try), troubled success rate (succeeded after retrying), failure rate (no successful charge in the visit), and average attempts per visit (a guardrail for how much effort drivers expend).
Together, these shift the question from "did this transaction complete?" to "did this driver get a charge?" — a behavioral measure of real impact of reliability improvements on driver outcomes. It is behavioral in the sense that in the absence of a clear failure signal, a driver who keeps retrying is telling you they have not reached success yet. Repeated attempts within a visit are evidence of friction, not ambiguity.
Visit is successful when the last charge attempt is successful.
Charge attempt is successful when:
- there is a transaction (energy transfer)
- next connector status is not ‘Faulted'
- transaction stop reason is 'Local’ or ‘Remote’ or ‘EVDisconnected'
- energy transferred is above 0.1 kWh
Success criteria are partially borrowed from https://github.com/chargex-consortium/OCPP-2.0.1-Interim-KPI-Calculator and/or are a modification of visit success when at least one charge attempt is successful here Customer-Focused Key Performance Indicators (KPIs) for Electric Vehicle Charging
git clone https://github.com/YOUR_USERNAME/kwwhat.git
cd kwwhatThen update your profiles.yml to point to your raw data location (e.g., DuckDB, BigQuery, Snowflake, Redshift, etc.).
-
OCPP 1.6 logs were kindly donated by Epic Charging
-
OCPP 2.0.1 logs were borrowed from OCPP-2.0.1-Interim-KPI-Calculator
Seed data was generated with OCPP synthetic log generator
Use dbt seed command to add seed data to your warehouse.
Designed based on industry frameworks and academic research to align metrics with real-world expectations:
-
The Public EV Charging Infrastructure Playbook by U.S. Joint Office offers guidance for performance evaluation in EV infrastructure EV Charging KPI Playbook
-
The Sage‑published journal article Novel Methodology to Measure the Reliability of Public DC Fast Charging Stations proposes a data-driven framework for charger reliability, which informed the visits logic in kwwhat. Novel Methodology to Measure the Reliability of Public DC Fast Charging Stations
-
Uptime calculations modeled in part after NEVI guidelines
This project was created independently and outside of any prior employment. It does not include any proprietary information, logic, or data.
The kwwhat project is licensed under the MIT License. External datasets and tools used in this repo follow their respective licenses as noted above.
Open to contributions from the EV data community. If you’re building in this space and want to improve reliability tracking, user experience analytics, or charger diagnostics — join in!
Questions? Ideas? Drop an issue or find us on LinkedIn kwwhat