Landvex — Data Quality

How we ensure data quality.

Every observation in Landvex passes through a multi-stage verification chain before it influences a score. Confidence is earned, not assumed.

See the process → Enterprise API

Five stages from observation to score

Every data point undergoes sequential verification. A failure at any stage removes the observation from the scoring pool.

1

Contributor identity verification

Each contributor is identity-verified before any observations are accepted. Device binding, location history, and activity patterns are monitored continuously for anomalies.

2

GPS validation

GPS coordinates are validated against declared location, movement patterns, and historical data. Spoofed, interpolated, or implausible locations are automatically rejected.

3

AI image review

Computer vision models review submitted images for specification compliance, content authenticity, and data integrity. Manipulated or non-compliant images are flagged and excluded.

4

Specification compliance

Observations are checked against task specifications: correct asset type, required angles, mandatory data fields, and temporal constraints. Incomplete submissions are rejected, not approximated.

5

Human spot-check (borderline cases)

Edge cases, borderline confidence scores, and flagged observations are reviewed by trained human analysts before entering the scoring pipeline. Automated decisions are not final.

Targets we hold ourselves to

Published targets create accountability. These are the standards Landvex commits to across every data collection mission.

>85%
Approval rate target
Observations that pass all verification stages. Below-threshold batches trigger quality review.
<10m
Geo-accuracy
GPS accuracy tolerance. Observations outside this threshold are flagged for manual review.
<90 days
Temporal freshness
Maximum observation age per data point before requiring refresh. Older data is down-weighted in scoring.

Every data point has a confidence score

Landvex attaches a 0–100 confidence score to every data point. Scores are transparent, composable, and auditable.

87 confidence

Confidence score: 87/100

Example: Norrmalm commercial vacancy observation, June 2026. Scored across four independent factors.

Recency of observation 32 pts
Source count 28 pts
Cross-validation against other sources 15 pts
Reviewer consensus 12 pts

Flagging what doesn’t add up

When observed data conflicts with official sources by more than 2 standard deviations, the observation is automatically flagged for review rather than silently averaged away.

>2σ

Automatic contradiction flag threshold

Any observation diverging from official sources by more than 2 standard deviations triggers a structured review process before inclusion in scores.

Flagged for human review
Source conflict logged
Contradiction report generated
Included in client delivery

Full provenance, every observation

Landvex stores observations with a complete provenance chain. Nothing is anonymised away if it’s needed for audit.

Contributor ID hash

Each observation is linked to a hashed contributor identity. Allows quality tracking without exposing personal data in standard outputs.

Timestamp

Precise capture timestamp stored with every observation. Temporal accuracy is critical for decay weighting and freshness scoring.

Device metadata

Device model, OS version, and sensor data retained for quality auditing and anomaly detection. Used internally; not surfaced in standard reports.

GPS trace

Full GPS trace stored per observation session. Enables movement-pattern validation and retroactive quality review if anomalies are detected.

Deeper access for enterprise clients

Standard Landvex output includes final scores and contradiction flags. Enterprise clients can go deeper.

Full audit trail on request

Enterprise clients can access raw confidence scores via API, review individual observation-level provenance, and receive structured audit trails for regulatory or investment committee use.

Raw confidence scores via API
Full audit trail on request
Observation-level provenance
Regulatory-grade documentation
Discuss enterprise access →

Questions about data quality?

Whether you need technical documentation, API access, or audit trails — get in touch and we’ll respond within one business day.

We don’t spam. Your information stays private.