Source ledger · Reviewed August 2026
Use this AI recruiting source ledger to separate binding rules, government guidance, professional practice, voluntary frameworks, technical standards, and first-party research. Each entry states what it can support and what it cannot establish for a specific workflow.
Public frameworks and guidance
These sources come from U.S., UK, EU, and international public bodies. Scope and legal effect differ. Read the linked page and applicable law for the role and jurisdiction; do not turn this summary into a universal compliance conclusion.
Useful for: organizing governance and lifecycle risk work around Govern, Map, Measure, and Manage. Does not prove: that a hiring system complies with employment law or performs well on a specific role. NIST describes the framework as voluntary and use-case agnostic; it is being revised.
Useful for: turning a high-level risk framework into documented actions and outcomes across an AI system lifecycle. Does not prove: that every suggested action is necessary, sufficient, or ordered for a recruiting pilot.
Useful for: understanding that selection procedures can raise discrimination concerns, including when a procedure disproportionately excludes a protected group without sufficient justification. Does not prove: that a particular model or workflow is compliant; employers must assess their actual use.
Useful for: reading the federal text on documentation, impact, and validity evidence for covered employee selection procedures. Does not prove: coverage, statistical sufficiency, defenses, or obligations in a particular matter.
Useful for: reviewing accessibility, accommodation paths, and whether a hiring technology measures job skills rather than disability-related characteristics. Does not prove: that a product is accessible for every person or use case; the guidance stresses examining technology before and during use.
Useful for: defining role tasks, competencies, context, and assessment content before evaluating a screening or ranking system. Does not prove: that a private-sector role brief or automated assessment is valid.
Useful for: predetermined job-related questions, consistent administration, and common rating standards. Does not prove: validity, fairness, accessibility, or the correctness of a hiring decision.
Useful for: procurement and deployment questions about purpose, governance, accessibility, assurance, testing, pilots, transparency, and monitoring. Does not prove: legal assurance or fitness for every deployment.
Useful for: observed privacy and information-rights issues in recruitment sourcing, screening, and selection tools. Does not prove: prevalence, legal status, or performance of another product.
Useful for: testable web-content accessibility criteria for candidate-facing interfaces. Does not prove: that the full hiring process is accessible or lawful.
Useful for: official Local Law 144 materials on covered AEDT use, bias audits, public summaries, and notice. Does not prove: that a system falls within the definitions or satisfies the requirements.
Useful for: current Commission explanations of scope, risk classification, employment use cases, obligations, and implementation timing. Does not prove: classification for a particular intended purpose or facts.
First-party product research
Metix AI publishes technical and research material about its own system. These pages are appropriate sources for what Metix says it built, measured, or learned. They are not independent validation of a buyer's deployment.
Useful for: understanding Metix's described agent boundaries, recruiting-native retrieval and matching design, evaluation layer, and reported business metrics. Test in a pilot: whether the evidence chain and quality gains hold for your role, market, and approval model.
Useful for: separating component, trajectory, and outcome evaluation; understanding golden sets, deterministic checks, calibrated judges, and online evaluation. Test in a pilot: the coverage, calibration, versioning, and connection between evaluation results and live corrections.
Useful for: framing drift across prompts, architecture, evaluation, models, and context. Test in operations: whether change control, golden sets, observability, and fallbacks keep quality stable after the pilot.
Useful for: reported retrieval and reranking metrics, dataset protocols, and boundary-aware modeling in a recruitment use case. Does not prove: live-role quality, fairness, or general superiority outside the reported protocol.
Useful for: a lifecycle-oriented literature review spanning job posting, matching, assessment, bias, explainability, and human oversight. Does not prove: the effectiveness of any one vendor or product configuration.
Useful for: a first-party argument for judging recruiting systems by delivered hiring progress and operating burden. Does not prove: return on investment or fit for every employer.