Q001 — Self-Audit¶


Research	R0048 — Corporate AI Training
Run	2026-04-01
Query	Q001

ROBIS 4-Domain Audit¶

Domain 1: Eligibility Criteria¶

Rating: Low risk

Criterion	Assessment
Evidence types defined before searching	Yes — publicly available training descriptions, frameworks, and curricula from named organizations
Criteria consistent throughout	Yes — same standard applied to all sources
Scope appropriate to query	Yes — covered corporate, government, and regulatory sources as requested

Notes: Eligibility criteria were defined by the query itself (named organizations, publicly available materials). Criteria did not shift during research.

Domain 2: Search Comprehensiveness¶

Rating: Some concerns

Criterion	Assessment
Multiple search strategies used	Yes — four separate searches targeting corporate, consulting, government, and DOL-specific content
Searches designed to test each hypothesis	Yes — searched for both evidence of comprehensive training and evidence of gaps
All results dispositioned	Yes — all 40 results across 4 searches dispositioned as selected or rejected
Source diversity achieved	Yes — government (DOL, GSA, NHS), regulatory (EU), corporate (Deloitte, NAVEX), tech (Microsoft)

Notes: Concern: internal training content (behind paywalls, corporate intranets) was not accessible. McKinsey, PwC, and KPMG training details remain underrepresented. These are search limitations, not methodology failures.

Domain 3: Evaluation Consistency¶

Rating: Low risk

Criterion	Assessment
All sources scored using same framework	Yes — GRADE reliability/relevance + bias domains applied consistently
Evidence typed consistently	Yes — Factual vs Reported distinction applied
ACH matrix applied	Yes — all evidence evaluated against all three hypotheses
Diagnosticity analysis performed	Yes — most and least diagnostic evidence identified

Notes: Consistent application across all seven sources.

Domain 4: Synthesis Fairness¶

Rating: Low risk

Criterion	Assessment
All hypotheses given fair hearing	Yes — H1 (comprehensive) given credit for DOL and NHS evidence; H3 (perfunctory) given credit for commercial training evidence
Contradictory evidence surfaced	Yes — evidence that contradicts the researcher's prior expectation (government programs being genuine) is prominently presented
Confidence calibrated to evidence	Yes — Medium confidence reflects the known gap between public descriptions and internal content
Gaps acknowledged	Yes — internal training content gap explicitly identified

Notes: Active compensation for researcher's declared bias toward viewing training as superficial.

Domain 5: Source-Back Verification¶

Rating: Low risk

Source	Claim in Assessment	Source Actually Says	Match?
SRC04	DOL framework explicitly names "hallucinations and accuracy limits"	Campus Technology reports Area 1 includes "Hallucinations and accuracy limits" as a named subtopic	Yes
SRC05	NHS framework addresses "automation bias and rejection bias"	NHS search summary states "cognitive biases including automation bias and rejection bias can affect decision-making with AI"	Yes
SRC01	NAVEX describes "possibilities and limitations"	NAVEX page states "Help your people understand the possibilities and limitations of AI tools"	Yes
SRC03	Deloitte covers Trustworthy AI with seven dimensions	Deloitte page lists seven dimensions of Trustworthy AI framework	Yes

Discrepancies found: 0

Corrections applied: None needed

Unresolved flags: None

Notes: All claims verified against source content. The characterization of training depth (shallow vs comprehensive) is an analytical judgment, not a source claim, and is clearly marked as such.

Overall Assessment¶

Overall risk of bias: Low risk

The research process was methodical and compensated for the researcher's declared bias. The main limitation is structural: internal training content is not publicly accessible, creating an inherent information gap.

Researcher Bias Check¶

Confirmation bias risk: The researcher expects training to be superficial. Evidence partially confirms this for commercial programs but government programs exceeded expectations. Assessment gives fair weight to both findings.
Availability bias risk: Publicly available training descriptions may not represent the full landscape. This is acknowledged as a gap.
Anchoring risk: Low — hypotheses were generated before evidence was collected, and the ACH matrix was used to evaluate systematically.