Purpose & editorial responsibility
HOL maintains Throughline to help people understand AI standardization across independent organizations and open-source communities. It is a research index, not a standards authority, certification programme or ranking of organizations.
Throughline offers editorial suggestions for voluntary collaboration in the open-source ecosystem. Each community sets its own scope, governance and priorities. Inclusion does not imply participation, endorsement, an agreed liaison or an obligation to act.
From Patchwork to ongoing discussion
The Patchwork: Syncing on AI Standards event brought people together to discuss the landscape, overlaps and gaps. Throughline offers a continuing reference for that kind of discussion. Indexed organizations are not represented as event participants or supporters unless a cited source establishes that relationship.
What is in the index
The initial research baseline is dated November 11, 2026. Later source reviews and schema snapshots carry their own observation dates. The directory includes international, regional and national standards institutions, working groups, consortia, foundations, specifications, implementations, conformance tools and guidance frameworks.
Coverage is uneven. A national-body roster entry does not prove active AI standards work. An individual IETF draft is not an adopted IETF standard. Publication status and project activity are separate dimensions. The library’s coverage panel records reference sets and known gaps.
Evidence before inference
Entity records retain primary-source links, native status, examined versions, assertions, unresolved conflicts and milestones. Relationships distinguish documented, demonstrated, inferred and proposed claims. Conflicting sources remain visible rather than being silently reconciled.
Confidence assessments and lexical similarity are analyst aids, not calibrated probabilities. A shared topic or phrase is a reason to examine scope; it is not evidence that two protocols are equivalent.
How monitoring works
The collector checks enabled official sources on their configured cadence, records response hashes and observations, and preserves a revision history. First observations are baselines. A page fetched today may describe an older publication. Unreachable or partially collected sources remain marked as incomplete, with retry backoff.
A successful scan does not promote a draft to a published standard. Milestones require their own evidence. Automated overlap leads compare shared vocabulary in normative passages and require editorial review.
How comparisons work
Workflow comparisons identify responsibilities, selected implementations, actual source definitions and field inventories. Transfers are directional: each required target field needs a source, a transformation or an explicit external input. Information loss, retained fields, version mismatches and gaps are recorded with proposed acceptance tests.
Exports use the same selected participants and evidence as the workspace. An untested acceptance criterion remains untested. Missing workflow coverage can be exported as a research gap.
Corrections & updates
Public visitors can browse, compare and export. HOL administrators can add or archive records, update sources, run scans and record public editorial notes. Edits preserve revisions and evidence. To suggest a correction, contact HOL with the record URL, the proposed correction and a primary source.
If live storage is unavailable, public pages identify the published research snapshot they are showing. The monitoring API reports the failure; fallback content is not presented as a successful scan.