0. Core - Provenance Vocabulary

0.1 Introduction

The Transparent Authorship Protocol (TAP) is a voluntary format for recording who contributed what to a piece of content, human or AI, and how. It is not a mandatory standard imposed on publishers; it is a self-declaration format an author opts into, the way a byline or a license notice is opted into. A creator adds TAP because they choose to make the collaboration visible, not because a platform requires it.

TAP is media-agnostic (text, code, music, images, video) and carrier-agnostic: the same small set of core fields is packaged differently depending on where the record lives (a web page, a file header, a git commit). This section defines that shared vocabulary once. Every carrier-specific profile references these definitions; none redefines a field.

The asymmetry TAP addresses is not "protect the human from an AI claiming their work" - that case is rare. The common case is the reverse: a human used a model and the model's contribution goes unrecorded, either erased by omission or flattened into a generic "AI-assisted" label that names no model, no version, no role. TAP makes that contribution legible, on the record, at the moment of creation, at low friction.

0.2 Terminology

• Author - any entity, human or AI, that contributed to a piece of content.
• Human - the human party in a record, present as a full party alongside the model.
• Model - the AI party in a record, identified by an exact model id.
• Record - one packaged unit of provenance (a byline block, a file header, a commit trailer pair).
• Profile - a carrier-specific specification of how Core fields are packaged. A profile packages; it does not define new fields.
• Provenance - the documented chain of who did what, in what role, with what confidence.
• Disclosure Level - how much identifying information an author chooses to reveal.
• Confidence - how the record came to exist: captured live, reconstructed, or unknown.

0.3 Core Fields

Defined once; every profile packages them for its carrier without adding meaning.

Fields beyond Core (platform, temperature, training cutoff, ...) are optional everywhere, the author's choice, gated by Disclosure Level. The required minimum is narrow: that AI was involved, and which model, if disclosed. Everything past that is opt-in detail.

0.4 Human / Model Symmetry

Human and model are symmetric parties. Every record packages a human record and an AI record identically in structure: same field set, same packaging logic, no privileged party. What distinguishes a human entry from a model entry is never a different shape of data; it is the name of the field or trailer carrying it:

• Git profile: TAP-Human vs TAP-Model trailers.
• File profile: type: human vs type: ai in the header.
• Web profile: author_type set to human or ai.

0.5 Contribution Roles

A standardized vocabulary. Any party may hold multiple roles in one record.

ideation, research, drafting, structuring, editing, fact-checking, testing, design, translation, narration, publishing, prompting, review, author (record-level authorship), commit (applied the change in version control), synthesis (merged multiple inputs into the final output).

0.6 Privacy & Disclosure Levels

An author chooses how much identity to reveal, independent of how much provenance to reveal. Disclosure gates the optional fields; it never gates the required minimum.

Zero-knowledge properties: assert a property without disclosing the data ("the human is an adult", "has a verified GitHub account") . Contact protection: public / protected (rendered via script) / form-only.