9 September 2026. Completed root lane of the OSDI comparison. These findings supplement the r21 graph; they are not yet graph-integrated. OSDI, CDDP and the Safety Tech Challenge Fund are distinct programmes. Advisory access, supplier delivery accounts and actual evaluated outputs retain different evidence states.
The government's 2 July 2021 account identifies a ten-person Online Harms Expert Group convened and chaired by the Centre for Data Ethics and Innovation (CDEI) for OSDI. Its role included advising on project scope and direction, privacy-preserving and legally robust data sharing, and data-stewardship practice. CDEI says the group had already encouraged attention to users' online experience alongside the technology industry's needs. That is an official account of advice and input, not acquired meeting minutes or an approval instrument.
The roster includes Poppy Wood, then Senior Advisor at Reset Tech, alongside ICO technology-policy head Ali Shah, Ofcom technology-policy lead Fred Langford, NSPCC policy head Andy Burrows, Open Data Institute programme lead Jack Hardinges, BBC product manager Jon Howard, Web Foundation adviser Azmina Dhrodia, Bristol researchers Asma Vranaki and Claudia Peersman, and Oxford researcher Jun Zhao. Roles are as listed in July2021, not claims about current employment. Poppy Wood's later Reset UK directorship/member-control records are separate dated relationships already in the case.
This is a concrete government-programme advisory route for a named Reset representative. It does not grant Reset a procurement vote, veto over deployment, ownership of government data, or approval of any later content decision. Those would require their own evidence.
Source: [CDEI's expert-group account](http[local research file] Full article and roster read; original retained as cdei-governance.html, 53,671 bytes, SHA256 5e0d8421c0f20775de28d1db2ed837351e9fee559d2b38029a8017c6c66660ed.
The [official OSDI description](http[local research file] describes a fifteen-month project to improve access to data for technologies identifying and removing illegal and harmful content. It identifies government, OSTIA, Faculty Science and PUBLIC as participants. Shared Outcomes Fund financing is named; the fund-wide GBP200m is not OSDI's own allocation.
The [June2021 longlist](http[local research file] describes nearly seventy discovery interviews, workshops involving OSTIA members, and a Whitehall roundtable. Its candidate responses include data repositories, privacy-preserving access, common taxonomies, benchmarking/accreditation and procurement/marketplace support. These were twenty-three candidate solutions awaiting shortlisting, not twenty-three delivered products or enacted requirements. The separate [security workstream account](http[local research file] describes early security, privacy and legal design, including ICO guidance. It is an account of approach, not an independent compliance audit.
The advisory group, discovery participants and government team therefore had a route to shape which industry problems became publicly funded projects. This locates influence in priority formation, before later procurement or deployment. The minutes, shortlisting matrix and specific feedback dispositions remain unacquired.
PUBLIC's G-Cloud14 service definition uses OSDI as a case study. It says it delivered a TRL4 evaluation sandbox to DCMS for pilot review/testing, focused on text models moderating suicide and self-harm. It describes a labelled dataset, technical and qualitative evaluation frameworks, three pilot firms, a DPIA, user guidance and scaling recommendations. Its final stage proposes progression to live operation; this is not proof of a live national service.
This is first-party evidence of a claimed delivery in a commercial service description, not an inspected sandbox, replicated classifier test, named pilot roster or government acceptance certificate. Its reported two hundred-plus interviews and three roundtables describe a broader programme than the earlier seventy-interview discovery account; differing scope or timing is possible. Do not treat these figures as contradictory counts of the same measured event or sum them.
[Original service definition](http[local research file] final PDF page13. Retained public-gcloud-sandbox.pdf: 3,142,085 bytes; SHA256 db602fec0ede3076b488f78be0bcd5698a16cf3c0cfb4306d8eb3ad641be4174. Root read the final case study and targeted surrounding methodology; not every service offering was evaluated.
The separate evaluation-output note distinguishes PA Consulting's programme-assurance role from PUBLIC's product-evaluation role. It also identifies a later PUBLIC-authored STCF2 report explicitly reusing OSDI practices, with self-reporting and testing limitations. No original PA final assessment or original OSDI framework was located in that bounded search.
A DSIT-commissioned, PUBLIC-produced [Data Landscape Review 2024–2025](http[local research file] published26March2026, gives a later account. It explicitly says PUBLIC produced earlier internal data-access reviews for DCMS. Its research comprises thirteen interviews between28January and5March2025, nine survey responses (six relevant to this workstream), and more than fifty-five sources. The appendix describes a targeted stakeholder tracker with selection reasons, tailored interview scripts and an affinity map; these underlying artifacts were not acquired. This small, purposive sector study is not a representative population survey or a direct causal evaluation of OSDI.
The report's acknowledgements include AVPA, Ofcom, OSTIA, Resolver, Yoti and other suppliers. That establishes participation or research input at the stated combined level, not that each answered every question, endorsed all conclusions or belonged to the earlier CDEI group. Its contextual citations include a Paladin/PUBLIC/Perspective Economics market study and a Tony Blair Institute paper. Citation does not establish commissioning by either organization.
Its findings complicate any picture of a seamless apparatus. Providers report persistent data costs, commercial/IP barriers, platform reluctance and inconsistent classification practices. Market demand and customer awareness are described as especially pressing. Some interviewees report moving toward brand protection where demand is stronger. Government programmes are appreciated, yet respondents describe slower momentum and uncertainty about enforcement. The report treats regulation as a demand driver; that is attributed market analysis, not proof that any firm secretly authored legislation.
One consequential reported development is sharing processed signals or classification outputs instead of raw datasets. That can preserve data custody while spreading judgments between services. The report gives Checkstep/Sightengine as one example and discusses other collaborative models. These are research-report claims requiring provider agreements, interface specifications and actual downstream decisions before promotion into an exercised cross-platform control claim. No WEF reference was found in the retained report's text search; no exclusive taxonomy lineage is inferred.
Retained report HTML: 336,941 bytes, SHA256 71fd0eb7b10593d398668bda52d09c45b66b414dada9640d3ca6274e2ae29e1f. Substantive reading: introduction/objectives and methodology; selected market, barrier and classification passages; intervention-impact section6.2; conclusion; appendix/acknowledgements and selected citations. A full HTML capture is not evidence every case study was independently checked. The link labelled as the full2022 Data Landscape Review resolves to an unrelated RDI organizational review; that page was not treated as the missing original.
We can now connect a named NGO representative to a specified government advisory group, an industry-informed selection process, commercial prototype development and later reuse of evaluation practices. That is more explanatory than a generic membership map. The field also contains legal constraints, resource shortages, competing interests and explicit limitations on measured performance.
The next discriminating records are ordinary project artifacts: group minutes and advice dispositions; the shortlisting/selection matrix; original evaluation instruments and government acceptance; and, for the signal-sharing branch, contractual permission and downstream action rules. PUBLIC's exact CDDP commission remains unresolved in the separate contract review. The OSDI consortium cannot fill that gap by association.