An end-to-end Evidence Trail from everwished.

HumanVerified Chain of Custody An everwished proof-of-concept proposal

Proof of process, made readable.

A proposed end-to-end Evidence Trail for selected creation, editing, review and custody checkpoints.

Discuss funding or collaboration Read the research pack

No live Evidence Records are issued, and no verification, certification or legal chain-of-custody service is offered here.

The journey matters as much as the output

Finished files and physical objects often hide how source material, editing, review and custody were handled.

Isolated metadata, signatures and receipts can be useful, but ordinary readers may struggle to interpret them together or understand what each one leaves open.

HumanVerified proposes to record selected, policy-defined checkpoints in a readable sequence while showing the limits and gaps attached to each claim.

A first-mover end-to-end evidence category

HumanVerified is developing a first-mover, cross-domain, reader-facing Evidence Trail. Here, end-to-end means spanning selected checkpoints from source or capture to current status; it never means continuous or complete observation.

02

Content provenance

C2PA Content Credentials can bind signed assertions, ingredients and editing actions to supported assets.

03

Software provenance

Version control and signed attestations can relate source, builds and releases, but do not prove authorship or intent.

04

Evidence Trail

HumanVerified proposes to connect selected machine evidence, scoped human decisions, handovers and current status in one purpose-bound reader journey.

Starling Lab's capture-to-publication work demonstrates important adjacent practice. HumanVerified's proposal is positioned alongside, not as a replacement for or endorsement by, these systems.

Five checkpoints. One readable thread.

Only selected policy-defined checkpoints are recorded; gaps stay visible.

  1. 01

    Source / capture

    Where source material first entered the declared process, with fresh challenges or signed capture evidence where appropriate.

  2. 02

    Create / edit

    Named artifact versions, ingredients, declared actions and controlled-tool records.

  3. 03

    Human review

    A person's decision, the material considered and the exact review scope.

  4. 04

    Handover / custody

    Transfer of control or possession, with physical and digital evidence labelled separately.

  5. 05

    Current status

    The latest accepted, superseded, withdrawn or unresolved state at the time of checking.

What a reader could see

Fictional example A handmade silver pendant moves through five declared checkpoints.

  1. 01

    Source / capture

    Role
    Materials recorder
    Fictional date/time
    4 Sep 2026 · 09:14 UTC
    Event
    Silver sheet and findings entered the declared process.
    Digest
    sha256: 8f2c…a1e9
    Evidence state
    Declared
  2. 02

    Create / edit

    Role
    Maker
    Fictional date/time
    4 Sep 2026 · 14:32 UTC
    Event
    Cut, formed and soldered pendant version recorded.
    Digest
    sha256: 41bd…90c2
    Evidence state
    Evidenced
  3. 03

    Human review

    Role
    Workshop reviewer
    Fictional date/time
    5 Sep 2026 · 10:05 UTC
    Event
    Finish and declared material notes reviewed to a stated scope.
    Digest
    sha256: b730…6d4a
    Evidence state
    Reviewed
  4. 04

    Handover / custody

    Role
    Handover recorder
    Fictional date/time
    5 Sep 2026 · 15:41 UTC
    Event
    Tagged package transferred with a digital receipt.
    Digest
    sha256: 0ca7…e133
    Evidence state
    Evidenced
  5. 05

    Current status

    Role
    Status custodian
    Fictional date/time
    7 Sep 2026 · 11:20 UTC
    Event
    Continuous attachment of the physical tag to the same object is not established.
    Digest
    sha256: d219…7bc0
    Evidence state
    Unresolved
This example illustrates the proposed reader language and is not an issued Evidence Record.

A record is not a verdict

Could support if implemented and validated

  • Exact bytes match a recorded digest.
  • A registered credential completed a defined checkpoint.
  • Named artifact versions are related.
  • A receipt entered a stated log checkpoint.
  • Displayed status was current when checked.

Cannot establish by itself

  • Semantic truth or factual accuracy.
  • Complete human authorship or absence of AI and undeclared tools.
  • Ownership, copyright or consent.
  • Quality, safety or suitability.
  • Uninterrupted physical custody.
  • Legal chain-of-custody status.

Three tracks. One shared model.

01

Research, documentation and code

Follow a brief, sources, drafts, review, build and release through selected checkpoints.

Research question Does the trail keep human decisions separate from Git, in-toto/SLSA-style and document evidence?

02

Moving image and editorial production

Connect a brief, primary and witness capture, fresh session challenges, editing, review and export.

Research question Do attempted replay, virtual-input and omitted-edit attacks leave understandable evidence gaps?

03

Manual craft and physical custody

Relate materials, making, review, a tag and handover without collapsing distinct physical claims.

Research question Can readers distinguish making evidence, tag possession and continuous physical attachment?

Test before scale

The proposal is a conditional six-week lab programme, subject to funding, staffing, privacy, security and accessibility design, with go/no-go gates before wider work.

  1. Week 1

    Frame

    Scope, governance, threat model and fictional fixtures.

  2. Weeks 2–3

    Construct

    Canonical events, receipts, credentials, transparency checks and a readable status view.

  3. Week 4

    Connect

    Media, software, document and craft adapters with adversarial fixtures.

  4. Week 5

    Test

    Controlled demonstrations, accessibility and reader-comprehension testing.

  5. Week 6

    Decide

    Recovery exercise, evidence-gap report and go/no-go recommendation.

Read the proposal in full

v0.1 · 3 September 2026 · Proposed proof of concept

AI-generated under human direction. No named independent human review or approval is currently recorded.

The documents and checksums identify the current proposal and allow exact release files to be compared; they do not show that the proposed system has been implemented, undergone an independent audit or validated in operation. Checksums do not stop a file being changed or prove authorship.

Build the evidence, test the limits

01

Fund the controlled programme.

02

Strengthen secure capture, credentials, provenance and transparency engineering.

03

Test real editorial, research, software and craft workflows.

04

Challenge privacy, security, accessibility, identity and evaluation assumptions.

Discuss funding or collaboration

No live Evidence Records are issued, and no verification, certification or legal chain-of-custody service is offered here.