Platform Vision

Human Layer · Constitution · Get involved

PLATFORM VISION / A PROPOSAL FOR DISCUSSION

Show your work.

Imagine a curated home for longform video where viewers can understand who stands behind a work, how its media were produced and where its factual claims come from. Documentaries, essays, education and creative work could share a common foundation: meaningful human agency, visible provenance and accountable publication.

This page describes a possible future platform. Human Layer is currently in Phase Zero: gathering collaborators and testing the idea. There is no video upload service, verification system or operating moderation team here yet. The following workflows and labels are proposals, not existing features.

Two questions. Two kinds of evidence.

Media provenance: Where did the footage, images and audio come from? What was captured, generated, edited or restored? What evidence supports that production history?

Information provenance: Where do factual claims come from? Can viewers inspect the relevant sources, distinguish evidence from interpretation and find corrections?

A genuine camera recording can contain a false claim. A human author can mislead. Neither human authorship nor a cryptographic signature guarantees truth. Both kinds of evidence matter.

Begin with a small, curated community.

A pilot could invite a few dozen creators rather than open unrestricted registration. Creators would agree to clear production and sourcing rules, appropriate upload limits and a review process. Applications would help establish who takes editorial responsibility and what kind of work they want to contribute.

Longform is welcome; length alone is not a quality test. The aim is to make thoughtful work easier to inspect, not to manufacture another engagement race.

From upload to accountable publication

Upload is the start of review, not automatic publication. A proposed workflow:

Creator and production declaration. Identify the responsible creator, contributors, source material and relevant AI assistance.
Receive and preserve the submitted original. Record an integrity hash and retain the uploaded source under a defined access and retention policy. A hash detects changes to a file; it does not prove how the file was created.
Hold for review. New submissions remain unpublished while the required checks are completed. Quarantine is a workflow status, not an accusation.
Check available provenance. Examine metadata, Content Credentials, archival references and the declared production history. Missing metadata alone does not justify rejection.
Screen and assess policy. Automated tools may identify inconsistencies or material needing attention. Review also considers sourcing, rights, misleading presentation and compliance with the published rules.
Human decision. Approve, request clarification, or decline with reasons. Disputed or uncertain cases require human judgment and an appeal route.
Prepare and publish. Create streaming versions and publish a viewer-facing provenance summary and source card. Relevant credentials must be preserved or appropriately linked across derivatives.
Continue accountability. Accept reports, publish corrections, conduct proportionate audits and reconsider decisions when new evidence appears.

Some checks could run in parallel. Technical transcoding may happen earlier; the essential requirement is that public release remains gated by the agreed review process.

Evidence labels, not a truth score.

Possible labels would describe the actual evidence available for each work or component:

Capture credentials checked: specified credentials and their signatures were validated against an explicit trust policy. This does not certify everything depicted.
Production evidence reviewed: a reviewer inspected relevant source files or production records, with the scope and date stated.
Creator declaration supplied: a production statement is available; independent verification has not necessarily occurred.
Archival source documented: material is traced to an archive, collection, item or other identifiable source.

These labels can coexist. A documentary may combine signed contemporary footage, historical film and declared illustrations. A film from 1944 cannot acquire an original camera signature retrospectively; its archive and digitisation history are the relevant evidence.

Source cards: let viewers inspect the claims.

A source card could connect important claims and media excerpts to specific references. Listing a famous archive or a bibliography is a starting point, not sufficient support for every statement.

Illustrative source card — not an actual reviewed video:

Claim at 03:12: source title, author, date, page or item identifier, link where available, and a note distinguishing direct evidence from interpretation.
Footage at 11:04–11:31: archive, collection and item identifier; relevant licence and editing or restoration notes.
Production: named responsibilities for research, writing, narration and editing; tools used and what they contributed.
Corrections: dated changes, reasons and affected claims.

Sources may themselves be mistaken or disputed. The card makes scrutiny possible; it is not a platform endorsement of every conclusion.

AI assistance and human authorship

Our starting principle is that an identifiable human author exercises meaningful editorial judgment and accepts responsibility for publication. AI may assist that work; declaring a tool does not automatically make its use acceptable.

Transcription, accessibility, translation, editing and research assistance need clear rules appropriate to their effects. AI research output is not a substitute for a checked source. Restoration must disclose changes that alter what the evidence appears to show.

Synthetic footage presented as documentary evidence, fabricated testimony and automated content farming conflict with this vision. Whether declared synthetic narration, generated illustrations or generative editing belong in particular categories remains an open policy question. A stricter human-capture category could coexist with other clearly described categories.

Human Layer's own generated website illustrations are disclosed as conceptual illustrations. They are not evidence of real events. The platform's eventual category rules must be equally explicit and consistent.

Detection, audits and fair decisions

Automated forensic screening would provide review signals, not a verdict about authorship or truth. A flagged work might prompt a request for original material or an explanation. No creator should be publicly accused solely because a detector returns a score.

Audits could combine risk-based checks and random sampling. Requests must be proportionate: older work, confidential sources, licensed material and missing project files require workable alternatives. Production files may expose private information, so reviewers need restricted access, secure handling and clear deletion rules.

Creators should receive the relevant reasons for a decision, an opportunity to respond and a route to a second review. Public verification records should state their scope and limits, avoid exposing private identity documents, and never become an opaque social score.

Build on existing infrastructure.

PeerTube is one candidate foundation. Its documented moderation controls include holding new uploads for approval and limiting registration; its plugin hooks offer ways to extend upload acceptance and review. Our provenance summaries, source cards and appeals workflow would still require design and implementation.

An initial pilot could operate without following external instances. Later federation would require explicit partner standards and decisions about how remote content is labelled and reviewed. Following another server does not automatically verify its content.

C2PA / Content Credentials could supply verifiable provenance assertions where available. Validation depends on trusted signers and what they attest to. Credentials can be incomplete or removed; they do not establish factual accuracy or human authorship by themselves.

What needs to be solved before a pilot?

Precise category rules and workable definitions of assistance versus substitution.
A moderation team, independent review capacity and protection against abuse.
Storage, streaming, backups, security and a sustainable budget.
Privacy, creator rights, retention limits and safe handling of source material.
A limited test set that measures review time, false flags, costs and viewer understanding.

Open-source infrastructure reduces some development work. It does not remove the continuing costs of hosting, moderation, support and governance. We have not yet established a launch budget or promised a launch date.

Help shape the human layer.

We welcome creators, PeerTube developers, provenance specialists, journalists, historians, accessibility experts and people experienced in moderation and Trust & Safety. Challenge the assumptions, bring practical constraints and help define a pilot worth building.

Start a conversation · Read Constitution V0.1

Technical references

Working proposal · 30 September 2026. Developed from Paul-Lukas Hertz's project direction and material discussed with ChatGPT; structured and drafted with AI assistance. Technical references describe existing tools, not an implemented Human Layer platform.