Read at claim level
Open the cited source, locate the supporting passage and confirm that it says what the argument says it says.
07LAB INSTRUMENT
EVIDENCE SPINE / 2026
Original points of view. Primary-source discipline.
Every claim remains open to challenge, every number to recalculation and every assumption to revision.
THE VERIFICATION STANDARD
Convincing is not the same as correct.
Autonomous agents can hallucinate with confidence. An agent checking another agent may improve coverage, but it does not create independent assurance.
Recalculate deterministically
Inspect inputs, formula, units, denominator and date. Do not ask another language model to merely agree.
Test permissions and reversibility
Know what the system can change, how it can be stopped and which evidence survives the action.
Name the accountable owner
Make authority, rationale, residual risk and escalation legible before consequences compound.
PRIMARY SOURCE REGISTER
Evidence that can be opened and challenged.
These sources inform the institutional context. The Lab’s frameworks and interpretations remain original arguments, not claims of source authorship.
NIST AI Agent Standards InitiativeAgent security, identity and interoperability
↗02Stanford AI Index 2026AI capability, adoption and institutional change
↗03OECD AI and SkillsSkills, work and the changing human contribution
↗04World Economic ForumSafeguarding and reinventing early-career pathways
↗If the same model family produces the claim, checks the claim and summarizes the check, the organization has layered fluency, not established independence.
CLOSE THE LOOP