Reliability is ordered from higher to lower, while comparable price is ordered from lower to higher.
Independent EV Evidence analysis
How the operator scorecard works
EV Evidence’s operator scorecard ranks two separately eligible components: higher reliability and lower comparable price. Exact ties share a midrank percentile, and an overall score is the mean of both component percentiles only when the same operator passes both eligibility gates. The current production build supports this reproducible method, not a league table.
Source: EV Evidence, Public charging reliability method.
Key facts and evidence
The published method keeps eligibility separate from scoring: reliability needs its own sample and coverage gates, price needs its own gates, and both components are required for an overall score. Exact equal values share a competition rank and midrank percentile. The current licensed scorecard contains no comparable operator observations, so it publishes no scores or rankings.
Source: EV Evidence, Public charging reliability method.
Each component applies its published sample-size and coverage gates before an observation can enter its cohort.
The current production scorecard contains no comparable licensed operator observations, so no scores or rankings are published.
Reproducible calculation
Component and overall formulae
Reliability is ordered from higher to lower. Comparable price in pence per kWh is ordered from lower to higher. Only eligible observations enter the relevant component cohort.
Tie midrank
Let pfirst and plast be the first and last one-based positions occupied by an exact-value tie.
rmid = (pfirst + plast) / 2Component percentile
N is the number of eligible operators in that component cohort.
P = 50 when N = 1P = 100 × (N − rmid) / (N − 1) when N > 1Overall score
The overall score exists only when the same operator is eligible in both components.
overall = (Preliability + Pprice) / 2Coverage duration
Coverage dates are inclusive. Duration is calculated before the coverage gate is applied.
days = coverage_end − coverage_start + 1 dayForm one comparable cohort per component
Keep only numeric observations with an identical complete metric definition and identical coverage dates for that component.
Apply the published sample and coverage gates
A row below either minimum remains disclosed but does not enter ranking. Neither sample size nor coverage duration changes its observed value.
Sort canonical values and assign ties
Sort reliability descending and price ascending. Exact equal values share competition rank, so positions are 1, 2, 2, 4. Operator ID may order display inside a tie but never breaks it.
Calculate each component percentile
Use the eligible cohort size and tie midrank in the formula above. A unique best value receives 100, a unique worst value receives 0, and a one-member cohort receives 50.
Calculate and rank complete overall scores
Average the two component percentiles only when both are eligible, then rank exact overall scores from high to low using competition rank again.
Published E2 gates
Eligibility, not hidden weighting
These values are read from the current scorecard artefact. If a gate is absent, EV Evidence does not supply a default on the page and must withhold the affected score or rank.
Reliability component
- Direction
- Higher is better
- Required unit
- Percent
- Minimum sample size
- ≥ 10 sample units
- Minimum coverage
- ≥ 30 inclusive days
Price component
- Direction
- Lower is better
- Required unit
- Pence per kWh
- Minimum sample size
- ≥ 1 sample units
- Minimum coverage
- ≥ 1 inclusive days
Overall requires both components: Yes — reliability and price must both pass. Sample size and coverage are pass/fail qualification gates and disclosures only. They do not weight, shrink, boost or estimate a component.
Missing remains missing
A missing value stays null with a reason. It receives no component rank and makes the overall score null; it is never filled with 0. A genuinely published numeric zero remains a numeric observation and is assessed normally.
Unformatted values decide order
Ranking and tie detection use canonical, unformatted values. Values that look equal after display rounding are not tied unless their canonical values are exactly equal. Exact rational overall scores are compared before display rounding.
Rounding is presentation only
Published scores may be displayed to four decimal places using round-half-up. That formatted value is never fed back into a component score, overall score or rank.
Time and release controls
Update cycle and data period
Build time, source check time and observation coverage answer different questions and are kept separate.
Cohort contract
Definitions must match exactly
Numeric rows are comparable only within a component whose complete definition and coverage dates are identical. A mismatch fails closed or must be published as a separate cohort.
Reliability
The unit must be percent and the direction must be higher is better. The definition must also identify its metric ID, description, sample unit, population and calculation. Different populations or denominators cannot share a ranking.
Price
The unit must be pence per kWh and the direction must be lower is better. A per-minute, per-session or subscription-only tariff, or a tariff with inseparable extra fees, is not converted into a guessed pence-per-kWh value.
Evidence attached to every row
Each row must retain operator identity, observed value or null reason, sample size, inclusive coverage period, observation time, provenance and the complete metric definition. Input order must not change the result. Charger, device and location are distinct sample units and cannot be silently mixed.
Interpretation boundary
What the score cannot tell you
A percentile is relative to its eligible cohort
It is not an absolute safety, service-quality or regulatory grade. Changing the comparable cohort can change percentiles even when an operator value does not change.
Source reporting is not an EV Evidence field measurement
Operator-reported and independently observed evidence must stay labelled as such. Admission checks provenance, reuse permission and comparability; it does not turn a source claim into an EV Evidence measurement.
Network reliability is not a session guarantee
The Public Charge Point Regulations 99% measure, if admitted as source evidence, concerns an operator's rapid network over a calendar year with the statutory scope and exemptions. It is not a promise that one charger, location, session or future visit will be available.
Price comparisons have a narrow scope
Payment channel, location, power, subscription, roaming and time window can change what a driver pays. A price described as current must have been observed within seven days and retain those scope fields, its source URL and observation time. Incomparable or older values remain missing, historical or unknown instead of being converted or averaged.
Sample size is not uncertainty adjustment
Passing the minimum does not make two sample designs identical and does not create a confidence interval. This method neither weights by sample size nor estimates an unpublished value.
Other evidence stays separate
Report discoverability, OCPI access or freshness, raw observed status and OPSS public outcomes are not silently folded into this two-component score. If available, they must be shown as separately sourced fields. A report not found after a documented search is not evidence that it was not submitted or that an operator acted unlawfully.
No payment changes the method
Sponsorship, affiliate relationships and commercial products do not add points, change eligibility, break ties or alter a safety classification.
Audit trail
Method version, licence and sources
Method identifiers and thresholds below come from the current E2 artefacts. Unknown values remain visibly unknown rather than being reconstructed from page copy.
Current scorecard evidence and licence record
uk-chargepoints-licence-audit
NCR official archives are OGL v3.0; no archive snapshot or separately licensed current CPO feed was admitted.
Current evidence warning: No verified official NCR archive snapshot or separately licensed current CPO feed was available to this build.
Method provenance
evcode-contracts
EVCode data contracts and configuration.
evcode-pipeline-code
EVCode data transformation code.
Evidence admission rule: operator-level public evidence, including statutory operator disclosures where applicable, must retain provenance and a verified reuse basis. Public availability alone is not permission to republish. Unlicensed feeds and fixtures cannot populate the production scorecard.
This page documents the calculation and its evidence boundary. It does not certify an operator, predict charger availability or replace the source's own scope and definitions.