MADE: Measure of AI in Design & Execution

Purpose & Philosophy

MADE (Measure of AI in Design & Execution) is a voluntary, author-declared rating framework that enables creators of software, writing, art, music, and other public works to transparently communicate how generative AI was used in their projects. What sets MADE apart from previous AI disclosure scales is that it rates original human ideas separately from mechanical implementation, rather than reducing an entire project to a single number.

Why Single-Number Scales Fail

Most disclosure scales and frameworks^[1] collapse all AI usage onto a single numeric spectrum or a single category label. Such scales routinely place brainstorming and ideation at a lower number than mechanical implementation. This distorts how AI usage is represented:

1. A creator who prompts an AI model for all creative inspiration, and then manually types or illustrates the result, receives a "low" AI rating, despite the core concepts originating entirely from AI.
2. A creator who conceives a 100% original design, hand-crafts the primary work, and uses AI strictly to handle repetitive procedural tasks (such as writing file-conversion utilities, or cleaning up audio and image noise) is assigned a "high" AI rating, despite only using AI for mechanical tasks.

MADE's Balanced Alternative: Separating Ideation from Execution

MADE resolves this problem by evaluating a project across two distinct rating categories, each scored from 0 to 4:

• Ideation (I0 to I4) measures who conceived the ideas, concepts, structure, design, composition, or algorithms.
• Execution (E0 to E4) measures how the tangible work (code, prose, visual art, music, etc.) was mechanically produced.

Both categories use the same five symmetrical, whole-number levels:

• 0: Unassisted
• 1: Assisted
• 2: Collaborative
• 3: Directed
• 4: Autonomous

Example Usage: Assume a creative project was 100% designed by a human, who then collaborated with AI to produce the supporting material. The project's author could then self-disclose the following MADE score:

MADE: I0 / E2
(or:)
MADE: I0 / E2 (Ideation: Unassisted; Execution: Collaborative)

Ideation Levels (I0 to I4)

The Ideation level measures the intellectual and creative origin of the work, including its core concepts, narrative themes, visual design, musical composition, or software algorithms.

Level	Title	Definition	Examples Across Creative & Technical Fields
I0	Unassisted	100% human-originated concepts, designs, compositions, and algorithms. Zero AI consultation.	Programming: Designing an algorithm or system architecture solely from personal knowledge and non-AI documentation. Writing: Conceiving a novel's plot, characters, and chapter structure independently. Art: Designing characters, layouts, and color palettes entirely from imagination and traditional references. Music: Composing melodies, harmonies, and arrangements without querying any AI tool.
I1	Assisted	Core concepts and designs are human-originated. AI used only for background research or minor conceptual refinements.	Programming: Inventing an original algorithm while using an LLM to review mathematical formulas and test minor edge-case tweaks. Writing: Plotting an original historical novel while using AI to summarize archival documents or verify period timelines. Art: Designing an original illustration while using AI to search for historical costume references. Music: Composing an original song while using AI to brainstorm lyric rhymes, thematic phrasing, or instrumentation references.
I2	Collaborative	A human set the creative vision and partnered with AI to brainstorm specific elements, sections, or outlines.	Programming: Designing an application architecture through an iterative dialogue where human and AI jointly propose subsystem designs. Writing: Co-developing worldbuilding lore or secondary subplot beats in a back-and-forth brainstorming session. Art: Iterating on rough composition thumbnails and visual motifs in collaboration with an AI tool before illustrating. Music: Composing a main theme by hand and collaborating with an AI tool to explore alternate chord progressions or counter-melodies.
I3	Directed	A human provided a high-level goal; AI proposed the primary structure or design choices, curated by the human.	Programming: Prompting an AI with desired input/output goals and allowing the AI to select the underlying data structures and architecture for human approval. Writing: Prompting an AI to generate several chapter outlines from a one-paragraph premise and curating the best outline. Art: Generating multiple concept layouts from a text prompt and selecting one as the blueprint for production. Music: Prompting an AI for song structures and harmonic progressions in a specific genre and curating the arrangement.
I4	Autonomous	AI conceived the project concept, design, and creative direction with minimal or no human conceptual input.	All Fields: Prompting an autonomous system to invent a software project, story concept, visual theme, or musical composition from a minimal, open-ended request.

Execution Levels (E0 to E4)

The Execution level measures how the finished work (source code, prose, visual artwork, or audio recordings) was produced and reviewed.

Level	Title	Definition	Examples Across Creative & Technical Fields
E0	Unassisted	100% human-produced code, prose, illustration, or musical performance by the human creator.	Programming: Writing source code or assembly routines completely by hand. Writing: Typing and editing a manuscript manually without generative text tools. Art: Drawing, painting, or modeling artwork entirely by hand. Music: Performing, sequencing, mixing, and mastering audio manually using only traditional (non-AI) tools, or read-only meters that do not modify the output signal.
E1	Assisted	100% human-authored primary output. AI is used strictly for diagnostics, proofreading, syntax checking, or minor restorative cleanup.	Programming: Writing code manually and using an LLM to scan for off-by-one bugs or syntax errors. Writing: Writing prose by hand while using AI as an interactive phrase thesaurus or proofreader. Art: Hand-painting digital art, then using AI tools strictly for automated dust removal or minor margin bleed extension for print preparation. Music: Recording live instruments and using AI-assisted plugins strictly for corrective signal cleanup (such as spectral denoising, dynamic EQ, pitch correction, or restoring a clipped sample) without generating new musical material.
E2	Collaborative	Shared human-AI production, such as human-crafted primary work paired with AI-generated supporting assets, boilerplate, or utilities.	Programming: Hand-coding the core application engine while using AI to generate test harnesses, build scripts, or CLI parsers. Writing: Writing the primary text of a book or article while using AI to generate summary tables, indexes, or data charts. Art: Hand-drawing all characters and foregrounds while using AI to assist with repetitive background textures or in-between animation frames. Music: Performing lead vocals and primary instruments by hand while using AI to generate secondary accompaniment stems or drum fills.
E3	Directed	AI produced the component's output strictly from detailed human specifications and constraints, verified by the creator.	Programming: Providing an LLM with a detailed, human-authored specification and algorithm so it writes a companion utility in a secondary language, followed by human testing. Writing: Translating a human-authored book into another language using AI under strict glossary constraints, followed by human review. Art: Generating rendered assets using a workflow tightly constrained by human-drawn control maps, depth maps, and manual paint-overs. Music: Synthesizing a vocal or orchestral performance from a human-written score and lyric sheet, followed by manual mixing.
E4	Autonomous	Generated end-to-end by AI from minimal prompts with little or no human specification or review.	All Fields: Accepting bulk AI-generated code without reviewing the logic ("vibe-coding"), or publishing raw AI-generated text, images, or music directly from a prompt.

Disclosure Format & Syntax

At its simplest, a MADE disclosure always begins with the MADE: prefix (including the colon), followed by a single Ideation level (I0 to I4) and a single Execution level (E0 to E4) separated by a forward slash (/). For example:

MADE: I1 / E0

For works with uniform AI usage throughout, this single rating is all that is necessary. When first introducing MADE in prose alongside a rating, authors are encouraged to spell out its full name (MADE (Measure of AI in Design & Execution)) or link to the specification.

MADE also defines a layered disclosure format that can scale with project complexity. Hybrid projects (where there were multiple levels of human and AI involvement for various pieces) can summarize multiple levels in the overall score, while optionally documenting individual components and source files underneath.

Note: You do not need to format complex disclosures by hand; the accompanying Interactive Badge and Disclosure Builder can automatically generate multi-level scores, badges, component breakdowns, and file header tags.

Single-Rating Formats

1. Shorthand Format: Start with the MADE: prefix, then separate the Ideation (I) and Execution (E) levels with a space, a forward slash (/), and a space.
   • Example: MADE: I1 / E0
2. Expanded Human-Readable Format: Start with a shorthand rating, then append the parenthesized Ideation: and Execution: titles immediately afterward, separating the two clauses with a semicolon (;):
   • Example: MADE: I1 / E0 (Ideation: Assisted; Execution: Unassisted)

Multi-Component Projects

When a project contains multiple components with different AI involvement levels, combine all applicable levels within each category from lowest to highest (0 to 4). Always include the letter (I or E) before every number.

• Discrete Levels: When components use non-consecutive levels with nothing in between, list only the levels used, separated by a comma without spaces:
  • Example: MADE: I1 / E0,E3
• Contiguous Ranges: When components span consecutive levels with no gaps (such as adjacent levels I0 and I1, or every level from E0 through E2), always join the first and last level with a hyphen rather than commas:
  • Examples: MADE: I0-I1 / E0 or MADE: I1 / E0-E2
• Combined Categories: Ideation and Execution can apply commas and hyphens independently. (Both can be combined within a category if needed, such as E0-E1,E3):
  • Example: MADE: I0-I1 / E0,E3

When writing multi-component ratings in the expanded format, append the applicable level titles in ascending order separated by commas after each category label, such as: MADE: I0-I1 / E0,E3 (Ideation: Unassisted, Assisted; Execution: Unassisted, Directed).

Component-Level Breakdown

Whenever a project includes multiple components with different ratings, authors are encouraged to include a brief component breakdown where each component declares both its Ideation level and its Execution level:

AI Disclosure ([MADE](https://mobygamer.github.io/MADE-specification/)): `MADE: I0-I1 / E0,E3`
- **Core Application Engine (`I0 / E0`, Ideation: Unassisted; Execution: Unassisted):** Designed and hand-coded 100% by the human author with zero AI involvement.
- **Data Import & Conversion Utilities (`I1 / E3`, Ideation: Assisted; Execution: Directed):** Algorithms designed by the author (`I1`), with code generated by an LLM from detailed human specifications and verified by hand (`E3`).

Or for a creative publication or album:

AI Disclosure ([MADE](https://mobygamer.github.io/MADE-specification/)): `MADE: I0-I1 / E0,E2`
- **Novel Manuscript & Dialogue (`I1 / E0`, Ideation: Assisted; Execution: Unassisted):** Plot and prose written 100% by the author (`E0`), using AI only for historical period research (`I1`).
- **Chapter Header Illustrations (`I0 / E2`, Ideation: Unassisted; Execution: Collaborative):** Character line art drawn by hand (`I0`), paired with AI-assisted background textures (`E2`).

Per-File Source Code & Asset Metadata Headers

To prevent individual hand-crafted files, manuscripts, or digital assets from being misattributed to AI (or vice versa) when viewed in isolation, authors may place a MADE: line tag in source file headers, document frontmatter, or asset metadata comments. Because a header tag describes how a single file or asset was made, it specifies a single Ideation level and a single Execution level:

Python / Shell / YAML / Makefile / Markdown / Text Metadata:

# MADE: I1 / E1 (Ideation: Assisted; Execution: Assisted)

C / C++ / Rust / Go / JavaScript / C#:

// MADE: I1 / E3 (Ideation: Assisted; Execution: Directed)

Assembly (.asm, .inc, .s):

; MADE: I0 / E0 (Ideation: Unassisted; Execution: Unassisted)

HTML / XML / SVG Artwork / EPUB / XMP Metadata:

<!-- MADE: I0 / E0 (Ideation: Unassisted; Execution: Unassisted) -->

Optional Tool & Model Attribution

Creators who want to note which specific AI tools or models were used can optionally list them in square brackets at the end of any standalone rating or file header tag, separated by commas:

# Example 1
MADE: I1 / E2 [Claude 3.5 Sonnet, Midjourney v6]
# Example 2
MADE: I1 / E3 (Ideation: Assisted; Execution: Directed) [GPT-4o]

Machine Verification

For the complete formal syntax rules, see the RFC 5234 / RFC 7405 ABNF grammar in MADE-GRAMMAR.abnf.

Licensing & Governance

MADE is free to use. The specification prose is licensed under CC BY-ND 4.0, and everything else (ratings, badges, templates, header tags, and software code) is dedicated to the public domain under CC0 1.0 Universal (see LICENSE for full details):

1. Everything Else (CC0 1.0 Public Domain): Assigning MADE ratings, embedding badges, copying disclosure templates or MADE: header tags, and using, modifying, or redistributing all accompanying software scripts, regular expressions, and builder code (scripts/sync_spec.py, scripts/sync.bat, and the builder CSS/JavaScript in docs/index.html) are completely free of restrictions.
2. Specification Prose (CC BY-ND 4.0): The normative specification text (MADE-SPECIFICATION.md, the specification prose in docs/index.html, MADE-GRAMMAR.abnf, and README.md) is licensed under CC BY-ND 4.0. Republishers may distribute the specification verbatim, but may not modify the definitions of I0 through I4 or E0 through E4 while calling the resulting scale MADE.

Express Permission Addendum: Notwithstanding Section 3(a)(4) of CC BY-ND 4.0, the Licensor grants perpetual, irrevocable, royalty-free permission to: (1) translate and distribute the specification in any natural language, provided the translation preserves the canonical I0 through I4 and E0 through E4 identifiers and definitions, identifies itself as a community translation, and links to the canonical English specification URI; (2) quote and display unmodified level titles and definitions in user interfaces, tooltips, and documentation; and (3) publish separate, externally authored extension profiles that add auxiliary metadata fields without altering I0 through I4 or E0 through E4.

Questions and Answers

If I use an AI tool that only checks or cleans up my work and doesn't create anything new, would I still be obligated to disclose it?

While international laws and industry standards differ, in general you would not be obligated. Under the European Union AI Act (Regulation (EU) 2024/1689, Article 50(2)), tools that "perform an assistive function for standard editing" or do not substantially alter the input data or its semantics are exempted. Similarly, the United States Copyright Office registration policy (88 FR 16190) establishes that authors are only required to disclose AI-generated content that exceeds de minimis amounts; using AI merely as an assistive tool to edit, touch up, or format work that remains human-authored does not require disclosure. However, this distinction changes once a tool synthesizes novel material that goes into the finished release, such as generative inpainting of new visual elements, rewriting passages, generating code, or resynthesizing vocal takes.

MADE does not evaluate or dictate how an author chooses to measure or disclose their level of AI usage. That said, an AI tool used strictly for diagnostics, proofreading, or corrective signal cleanup (such as denoising, dynamic EQ, or pitch correction) would generally be classified as E1, whereas tools that do not modify the output signal remain E0.

Which other AI disclosure standards separate Design from Execution just like MADE, but provide even more functionality?

With at least 14 competing standards for AI disclosure in circulation, as of publication there are at least two specifications that allow separating conceptual design from execution while providing additional specialized features:

AI-DECLARATION.md is more comprehensive than MADE, providing extensive features relevant to software development: a formal machine-readable YAML specification, tracking AI involvement across six discrete software lifecycle processes, and granular path-based component mappings. However, you may still prefer MADE if your work spans non-software disciplines like creative writing, art, and music, or if you prefer a compact human-readable shorthand (such as "MADE: I0 / E2").

The AI Attribution Toolkit has multiple orthogonal declarations, allows for independent disclosure of design and execution, and is applicable to non-software works. Attributions are represented as tagged shorthand, such as AIA Ph SeCe Hin R v1.0. They offer a tool to expand attributions back into natural language.

Does MADE align with disclosure rules on commercial platforms?

Mostly yes. Major digital storefronts already draw a practical line between assistive workflow tools and directly generated output. Amazon Kindle Direct Publishing (KDP) exempts AI-assisted brainstorming, outlining, and grammar editing from disclosure, while requiring authors to declare AI-generated text and images. Valve's Steam Content Survey focuses disclosure on pre-generated or live-generated player-facing assets. Medium requires explicit transparency labels whenever AI assistance is used. MADE mirrors these by letting creators disclose individual pieces or components of their project separately from generative execution, without over-reporting routine operations as "AI generation".

Does MADE compete with cryptographic provenance standards like C2PA (Content Credentials)?

No, because C2PA and MADE solve different problems. C2PA embeds cryptographic assertions inside binary files to verify editing history and tool provenance, while MADE is used to self-declare how much AI was involved in the creation of an asset.

If I prompt an AI tool with my own original idea, can I use I0 (Unassisted) Ideation?

I0 is meant to represent unassisted ideation; a wholly original idea represented in great detail. I0 means you personally originated the concrete structure, plot, composition, or architecture before any tool execution. Simply entering a short prompt and letting an AI invent the underlying structure, details, or creative direction is I3 (Directed) or I4 (Autonomous) Ideation, not I0.

Authorship & Provenance

The MADE standard was created by Jim Leonard (mobygamer@gmail.com) to provide programmers, writers, visual artists, musicians, and independent creators with a human-readable self-disclosure standard that distinguishes human conceptual originality from mechanical AI assistance.

Notes & References

1. Examples of single-spectrum or single-label AI disclosure frameworks include the REAL Rating, the AI Assessment Scale, W3C's AI Content Disclosure for HTML, and RAID - Responsible AI Disclosure. ↩