An AI governance platform is software that operationalises an organization’s AI risk and compliance obligations - inventories, risk assessments, policy enforcement, documentation. Most platforms govern models and use-cases; almost none govern the training-data layer where model behavior is decided. LabelFort operates as the governance-of-training-data layer: versioned guidelines, role separation, per-cohort IAA, immutable audit trails and Article-mapped evidence exports, feeding the documentation your model-level governance platform expects.
The category (model registries, risk-assessment workflows, policy engines, documentation generators) answers board- and regulator-level questions: what AI do we run, what risk class is each system, who approved it, where is the technical file. These are necessary functions - and they sit downstream of the data.
Ask a model-governance platform which guideline version produced the labels in the training set, or what inter-annotator agreement was on the pediatric cohort, and the honest answer is that the platform records whatever the annotation vendor attested. If the attestation is a process claim rather than an export, governance inherits a gap it cannot see. Annex IV Section 2 asks for labeling procedures specifically - someone has to actually have them.
The training-data governance layer produces: versioned ontologies and guidelines; enforced role separation; per-cohort agreement statistics against contracted thresholds; complete, tamper-resistant event logs; provenance and transfer records; and a per-dataset evidence bundle. LabelFort ships these as deliverables - the inputs your governance platform’s technical file needs.
Whether evaluating LabelFort or anyone else: (1) Is the audit log complete and immutable, or sampled? (2) Is agreement measured per cohort with a contracted threshold? (3) Are guidelines versioned with per-record applicability? (4) Is the evidence one export or a support ticket? (5) Can the vendor show a redacted evidence bundle from a real engagement? Marketing survives four of the five; only an export survives all of them.
Agreement between two annotators, corrected for the agreement expected by chance. κ = 1 is perfect; κ = 0 is no better than chance. Regulated annotation QA contracts above 0.90.
Read More
How consistently independent annotators label the same data. High IAA evidences that labels reflect a shared, documented standard rather than individual opinion.
Read More
Reliability for any number of annotators, tolerant of missing data, across nominal, ordinal, interval and ratio labels - the general case where kappa handles exactly two.
Read More
Software that operationalises AI risk and compliance management - system inventories, risk classification, policy enforcement, and regulatory documentation across the AI lifecycle.
Generally only by attestation. The annotation stage, guidelines, agreement measurement, annotator identity, and audit trails need their own governed tooling that exports evidence into the model-level platform's documentation rather than a claim the platform has to take at face value.
LabelFort is the governance layer for training data specifically: governed annotation with Article-mapped evidence exports. It complements model-level governance platforms rather than replacing them.
Not automatically. A platform that logs "training data reviewed" without a per cohort agreement score, a guideline version, or an exportable record is logging an attestation, the same gap described throughout this page, just formatted as a checkbox inside the governance tool rather than absent from it entirely.
LabelFort provides IAA scores per cohort, audit trails for every action, & evidence exports mapped to EU AI Act Articles 10 & 12. Experience this on your own data in an evidence-grade proof of concept.
Certifications & readiness
