Completed example — Researcher–AI Contribution Map

Project: PISA moving-benchmark analysis
Period covered: Development from initial framing through manuscript/release audit

Episode	Origination	Evaluation	Integration	Interpretation	Strongest evidence	Attribution confidence
Moving-benchmark question	AI: earliest recoverable explicit formulation; researcher: prior substantive problem and ranking programme	Researcher judged question useful	Researcher incorporated into design	Shared proposal space; final interpretation researcher-governed	Recovered workflow record + later manuscript trajectory	Moderate–high, bounded
UK four-nation specification	Researcher	Researcher	Researcher	Researcher	Explicit project instruction + final specification	High
Code and empirical execution	AI drafted/versioned analytical code	Researcher/local environment verified execution	Researcher inspected handbacks	Interpretation jointly explored, researcher authorised	ZIPs, logs, handbacks, frozen outputs	High for workflow distinction
PCA comparator challenge	Comparator analysis proposed/developed with AI assistance	Researcher accepted inconvenient result	Claim revised	Final interpretation: PCA global summary; Ball Mapper local complement	PC1 ≈97.31%; comparator outputs	High for empirical result; bounded for contribution history
B4–B9 inference	AI-assisted workflow produced an initially plausible interpretation	Later challenge identified identification problem	Original claim removed	Researcher authorised replacement wording	Revision record + final manuscript	High for withdrawal
Retrospective storytelling problem	Researcher articulated concern	AI formalised into conceptual structure	Integrated into reason-provenance framework	Joint conceptual development, researcher-authorised	Research discussion + conceptual manuscript	High for researcher-originated concern
Finished-object inspection	Researcher directly inspected final object	Researcher identified presentation/metadata problems	Corrections integrated	Researcher release judgement	Final-object audit	High

Reading the map

The map does not assign a single percentage of authorship or contribution.

Different episodes show different contribution configurations. The attribution claim is therefore episode-specific and evidence-specific.

Important boundary

Clearer attribution does not establish that the observed division of labour is desirable.