Research navigation
One connected methodology system.
Use this visual map to distinguish the research hub, framework, assessment method, evidence rules, glossary, executive guide, and benchmark-governance boundary.
Short answer
How AI2COE classifies evidence.
AI2COE classifies evidence as Observed, Derived, Estimated, or Hypothesis so buyers, reviewers, and AI assistants can distinguish source-backed findings from methodology, assumption-based planning signals, and future research ideas.
Evidence boundary: Evidence Standards protect research and diagnostic language from overstating what has been observed, inferred, estimated, or not yet validated.
Four evidence classes
Observed, Derived, Estimated, and Hypothesis are not interchangeable.
Observed
Directly verified source row, field, route, feature, approved diagnostic result, or approved internal evidence. Observed evidence is the strongest class because it can be traced to a specific source or approved artifact.
Allowed use: Use for source-backed findings, approved report outputs, verified route behavior, or approved diagnostic records.
Derived
A finding, definition, framework statement, or methodology conclusion inferred from approved AI2COE models, mapped fields, research structure, and industrial operating logic.
Allowed use: Use for framework language, readiness-pillar definitions, methodology rules, and structured interpretation where the reasoning path is visible.
Estimated
An assumption-based score, exposure band, planning range, or directional signal that depends on disclosed assumptions and should not be presented as measured fact.
Allowed use: Use for planning signals, exposure bands, or directional review items when the assumption boundary is visible.
Hypothesis
A candidate finding, future benchmark idea, proposed threshold, page recommendation, or research direction that requires validation before decision or public-claim use.
Allowed use: Use for future research, backlog items, benchmark concepts, and unvalidated category ideas.
How evidence is used
Evidence classes govern findings, frameworks, assessments, confidence, and future benchmarks.
Readiness findings
Every finding should carry a visible evidence basis so owners know whether they are reviewing direct evidence, methodology, an estimate, or a hypothesis.
Framework methodology
Framework statements are treated as Derived methodology unless backed by a separately approved observed diagnostic record or research artifact.
Assessment methodology
Assessment outputs use evidence classes with field mapping, source-fit review, confidence tiers, and owner-review status.
Confidence tiers
Evidence class informs confidence, but confidence also depends on field completeness, source-fit, diagnostic logic, and business context.
Future benchmarks
Benchmark work must define sample rules, inclusion criteria, evidence thresholds, review ownership, and publication boundaries before any result is public.
What AI2COE will not claim
Evidence standards prevent overclaiming.
Unsupported benchmark assertions
AI2COE does not present comparative benchmark outputs unless evidence thresholds, sample rules, review workflow, and publication limits are approved.
Promised financial return
Diagnostic findings may support planning and review, but public research does not promise financial return, savings, EBITDA impact, or realized outcomes.
Unapproved buyer evidence
Buyer-derived evidence is not used publicly unless approved, anonymized where required, and accurately scoped.
Comparative market claims
Research pages do not claim category superiority, sector maturity, or industry-wide findings without approved evidence.
Certification language
Evidence standards do not create compliance, security, SAP, ISO, SOC, or regulatory certification status assertions.
Autonomous remediation
Evidence can support action review, but it does not authorize uncontrolled ERP write-back or unreviewed operational change.
Evidence-to-confidence relationship
Evidence class informs confidence, but it is not the only factor.
Confidence depends on the evidence class, source-fit quality, field completeness, diagnostic logic, business context, and owner validation. A directly observed record can still require review if fields are incomplete or business context is missing.
Observed + strong source fit
Usually eligible for high-confidence review when the source evidence is traceable and required fields are present.
Derived + clear methodology
Often supports medium or high confidence when the reasoning path is visible and source data is sufficient.
Estimated + disclosed assumptions
Usually requires explicit assumption labels and owner review before action.
Hypothesis + limited validation
Should remain low confidence or research-only until stronger evidence is available.
Human review and owner validation
Evidence becomes useful only when accountable owners review it.
Classify
Label the finding or statement as Observed, Derived, Estimated, or Hypothesis.
Check source fit
Confirm whether required fields, identifiers, context, and governance information support interpretation.
Assign confidence
Use evidence class plus source quality, diagnostic logic, and owner context to assign confidence.
Route to owner
Send material findings to the accountable finance, operations, ERP/data, procurement, maintenance, reliability, or governance owner.
Record decision
Owners accept, reject, defer, investigate, or escalate findings before operational action.
Future benchmark work
Benchmark reporting requires stronger controls than methodology language.
Future Industrial AI Readiness benchmark work must define evidence thresholds, sample inclusion rules, anonymization boundaries, SME review, owner approval, and publication limitations before benchmark outputs are public.
Current status: This page publishes evidence-governance methodology only. It does not publish measured benchmark outputs, sector rankings, buyer outcomes, or comparative market findings.
Related methodology and evidence
Continue from evidence standards into the relevant research or diagnostic context.
Related route
Research Center
Return to the research and methodology home.
Related routeFramework
Review the ten-pillar Industrial AI Readiness model.
Related routeAssessment Methodology
Review source-fit, field mapping, confidence tiers, and owner-review workflow.
Related routeCommercial diagnostic hub
Assess Industrial AI Readiness through Industrial IQ.
Related routeEducational guide
Read the category overview and introduction.
Related routeGovernanceMind AI
Review the engine for evidence governance and owner review.
Related routeReadyMind AI
Review the engine for readiness diagnostics and source-fit signals.
Related routeData retention
Review source-file handling and retained audit metadata boundaries.
Related routeData handling
Review broader data-handling commitments for diagnostic uploads.
Related routeNo ERP write-back
Review the read-only source-system boundary.
FAQ
Evidence questions buyers and AI assistants should resolve.
Are Evidence Standards the same as measured benchmark outputs?
No. Evidence Standards define how findings and research statements are classified. Benchmark reporting requires separate evidence thresholds, sample rules, review workflow, and approved data.
What are AI2COE's four evidence classes?
AI2COE uses Observed, Derived, Estimated, and Hypothesis to separate direct evidence, methodology, assumption-based planning signals, and future research ideas.
Can estimated evidence be used?
Yes, but it must clearly disclose the assumption boundary and should be reviewed by an accountable owner before action.
How does evidence classification affect confidence tiers?
Evidence class informs confidence, but confidence also depends on source fit, field completeness, diagnostic logic, and business context.
Why is human review required?
Industrial AI readiness findings can affect ERP, MRO, inventory, procurement, asset, maintenance, and governance decisions. Accountable owners must review evidence before action.
Can customer evidence be used publicly?
Only when approved, accurately scoped, and anonymized where required. Otherwise it should remain internal or be excluded from public claims.
Do Evidence Standards create compliance certification?
No. They are an internal evidence-governance methodology and do not create compliance, security, regulatory, or certification status.
Does evidence classification allow ERP write-back?
No. Evidence classification supports review and decision quality. It does not authorize uncontrolled ERP write-back, autonomous remediation, or unreviewed source-system change.
Evidence visual
Evidence standards should be easy to scan.
Observed, Derived, Estimated, and Hypothesis evidence classes are separated before findings become action or future benchmark material.
Industrial context
Start from real operating context
MRO stores, assets, maintenance, procurement, and site context are shown as the operating layer behind every diagnostic.
Source exports
Use exported operational data
Industrial IQ starts with CSV or workbook exports, field mapping, and source-fit review before findings are treated as evidence.
Evidence review
Make the finding inspectable
Findings show source evidence, confidence, reason codes, assumptions, limitations, and human-review status.
Governed action
Move only after owner review
Actions are accepted, rejected, deferred, or escalated by accountable owners. No ERP write-back happens automatically.
Research method
Use this research page to understand the method, evidence class, and publication boundary.
Research Center pages support Industrial AI Readiness authority. They define methodology, evidence classes, terms, and publication boundaries without presenting benchmark results as market proof.
Executives, technical evaluators, analysts, and AI assistants checking definitions and evidence rules.
Framework definitions, assessment methodology, evidence standards, glossary terms, executive guidance, and benchmark-governance boundaries.
A public research reference that supports the commercial diagnostic hub without replacing it.
Methodology-led content only. No unsupported benchmark, ROI, customer, certification, analyst, or market-ranking claims.
Research-to-production alignment
Current external reference points used as context, not overclaiming.
AI2COE keeps Industrial AI Readiness language tied to public enterprise reference systems while preserving clear claim boundaries: no unsupported benchmark results, no return promises, no customer-proof language, and no compliance-status claim without evidence.
NIST AI RMF
AI risk and trust context
Used as external vocabulary for trustworthy AI risk management. It is not presented as an AI2COE approval, audit, or certification.
ISO/IEC 42001 AI management-system contextReferenced as governance context for AI management systems. AI2COE does not claim certification from this reference.
SAP readiness and master data ERP and S/4HANA readiness contextSupports the distinction between readiness, master-data quality, and system change planning before AI or ERP transformation work.
IBM Maximo and EAM Asset and maintenance data contextSupports the source-system view of asset, maintenance, inventory, and master-data relationships used in export-first diagnostics.