Confidence Scoring

How to evaluate the reliability of accessibility measurements.

Overview

Every surface measurement receives a confidence score from 0–100%, indicating its reliability for OSM contributions. The current tooling calculates this score for single rides; rides over the same segment can additionally be combined into a multi-ride consensus.

Single Ride Scoring

Confidence for a single ride multiplies three factors, each capturing a different threat to data reliability. The product expresses how trustworthy one ride is for mapping surface accessibility – a weakness in any single factor pulls the whole score down.

Scoring Factors
FactorDescription
Classification ClarityScores how clearly the Z-axis standard deviation falls within the calibrated smoothness thresholds
Speed AppropriatenessAssigns the full speed factor within the 2–6 km/h assessment range and reduces it outside that range
Movement ContinuityReduces confidence when recent samples contain stops

Confidence Over Time

The following chart tracks confidence scores for a single ride, showing how speed, classification clarity, and movement continuity affect the final reliability before aggregation at the street (“way”) level. High-confidence periods indicate strong alignment with wheelchair-relevant conditions, while low-confidence intervals signal stops, abnormal speed, or potential sensor noise.

Loading chart data...

Multi-Ride Scoring

Multi-ride consensus can combine rides over the same segment using spatial clustering and temporal distribution, so that agreement between independent riders, rather than one strong ride, carries the weight of a proposal.

Spatial Clustering
FactorDescription
Tolerance radius10 meters (spatial buffer for GPS error)
Minimum rides3 rides for consensus scoring
Temporal spreadData from different days/times

Clustering alone does not make rides trustworthy. A second set of indicators grades each cluster on internal consistency, timing spread, and rider variety before any consensus value can be proposed.

Quality Indicators
MetricDescription
Classification consensusProposed minimum: ≥70% of rides agree
Spatial coverageAt least 30% of segment length sampled
Ride diversityDifferent users and e-scooter types

Confidence Thresholds

Confidence thresholds determine how OSM proposals are handled. Scores above 80% are generally considered safe for contribution, while everything below that bar earns progressively more scrutiny. The confidence in that bar comes from testing against OpenStreetMap itself: where a tag was already present, the algorithm's classification agreed with it in more than 8 out of 10 cases. While end-to-end automation is the goal, until the method is validated, pilot proposals are subject to especially conservative rules.

General Proposal Rules
Confidence LevelScore RangeRecommendation
High>80%Direct OSM contribution
Medium60–80%Manual review recommended
Low40–60%Additional validation needed
Insufficient<40%Exclude from proposals
Pilot Proposal Rules
Repetition: Evidence from at least two distinct rides
Gap-filling only: No existing smoothness=* tag on the way
Way types: Only footway, cycleway, path, or pedestrian
Bad: At least 30 m of continuous bad or very-bad evidence
Intermediate: At least 70% of evidence non-bad
Before upload: Fresh OSM check, manual review, create-only

Practical Application

Single rides are enough to surface local accessibility gaps. The multi-ride workflow can analyze larger datasets systematically for city-wide coverage. The thresholds are the same in both cases; what changes is how much data stands behind each proposal.

Single Ride Processing
Apply Filters: At least 70% coverage and 70% point agreement
Focus Areas: Segments without existing smoothness tags
Multi Ride Processing
Proposed Automatic Threshold: 75% consensus
Proposed Minimum: 3+ rides per segment

Wheelchair confidence score tiers above are guidance for review priority. The current single-ride export and apply workflow defaults to coverage ≥70% and point agreement ≥70% (apply_smoothness_batch.py). High (>80%) remains the recommended bar for direct contribution without extra review until the method is validated.

Until the method is validated, more conservative rules were applied for the pilot.