AI Content Disclosure and Provenance Policy

This public archive uses AI tools as research, drafting, and organizing assistants. Mike Hall is the human operator and curator of the articles and materials that carry this disclosure. He supplies the intent, lived memories, source material, architectural direction, and active curation.

To be clear: saying Mike personally line-edited every file or guarantees that he has read every word is not accurate. He is reviewing the process, working in and through the tool, and curating this digital garden with care. He verifies a good deal of what comes out, but parts of the garden grow faster than others, and there are no guarantees that every claim is an absolute factual statement.

We are in a distinctly weird moment in computing. Using an AI assistant to warn readers about the epistemic limits and hallucination traps of AI carries an unmistakable irony: at this point, what is even entirely “pure” text anymore? This archive is an open experiment in learning in public. I am doing my best to apply rigorous safety protocols, contract tests, and physical ground truth, but it remains a continuous learning process. If you spot holes, ungrounded leaps, or opportunities for tighter boundaries, suggestions and corrections are genuinely welcome.

AI tools can help organize a large archive, suggest structure, summarize source material, and draw diagrams. They cannot independently verify a memory, hear a recording, establish a physical invariant, or take responsibility for an engineering claim.

The publication rule

Every article must make its authorship and production process legible. A reader should be able to answer four questions without guessing:

Who is responsible for the article?
What source material shaped it?
How did AI assistance affect the result?
Where should a reader go to verify an important claim?

The visible article notice is backed by frontmatter, provenance links, the canonical Markdown source, and automated validation. The goal is a strong, repeatable publication contract. No single policy can promise universal legal or platform compliance in every jurisdiction, so material claims still require human review and source-level verification.

Disclosure levels

Level	Use	Public treatment	Required disclosure
Organic writing	Mike wrote and edited the article without generative AI	Normal publication	Author and ordinary source notes when useful
AI-assisted, human-led	Mike provided the subject, evidence, direction, corrections, and approval. AI helped organize, draft, or illustrate	Public and indexable when the content is ready	Visible notice, human responsibility, process description, and source links
Exploratory AI synthesis	AI produced substantial structure or prose while Mike is still evaluating the result	Quarantined under `/ai/`, `noindex,follow`, and excluded from the sitemap	Visible notice, explicit uncertainty, provenance, and primary-source links
Transcript support	AI helped transcribe, normalize, or align a recording	The transcript remains a best-faith record of the speaker or speakers	Identify the speaker model, preserve the source recording and its timestamps, and never present AI as a speaker or author

“AI-generated” is retained as a technical flag for substantial machine assistance and quarantine. It does not mean that AI is the author. Where the human-led relationship is known, the article also records human_led: true and source_kind: ai-augmented-human-led.

For this archive, “Strong Enough to be Gentle” is the last organic article in the recent writing run. Articles published from 2026-08-29 onward that were created with substantial AI assistance are classified as exploratory AI syntheses, even when Mike supplied the memories, evidence, direction, and final judgment. They live under /ai/, receive the standard visible notice, use noindex,follow, and stay out of the sitemap.

The early LinkedIn essays from late 2024 and early 2025 are a different case. Mike developed them through interactive ChatGPT prompting and human editorial direction. They remain public and indexable, but now carry the same visible ethics notice and an explicit historical-production note. “The Sound of Tokens” is restored to the public archive as a human-led personal essay with its former /ai/ route retained as a redirect. “From Y2K Panic to AI Anxiety” remains quarantined because its personal detail exceeds the public lesson’s useful surface area.

What the notice means

The site notice is intentionally plain:

Mike Hall led the article.
Mike supplied the subject, source material, corrections, and editorial judgment.
AI helped organize or draft the material.
AI is not an independent historical authority.
Published claims remain Mike’s responsibility.
Important claims should be checked against the linked primary archive.

For quarantined material, the notice also states that the page is exploratory, non-canonical, and excluded from search indexing. A transcript gets a different provenance treatment from an essay because the recording is the primary source. The transcript is a best-faith attempt to preserve the words, speaker attribution, turn boundaries, and timing of a historical recording. AI may be used in service of that fidelity. It does not become a speaker, author, or independent witness.

The independently inspectable evidence is the recording hosted by the original service and the timestamps visible in that player. A reader can compare the transcript against the recording. If a transcript time label is inferred from text alignment rather than supplied by the source, the implementation must not present that label as a frame-accurate measurement. A wording, attribution, or turn-boundary question should be reported with the recording timestamp so the archive can be corrected against evidence.

Verification checklist

Before publication, the article must pass these checks:

The human author is identified as Mike Hall.
The frontmatter identifies the assistance level and source kind.
The article has a title, description, normal heading structure, and source note where historical claims need one.
The article distinguishes what happened, what is interpretation, and what is uncertain.
Important dates, names, numbers, quotes, and causal claims resolve to a public source or are labeled as personal recollection.
AI is not listed as the author, speaker, witness, or independent authority.
Exploratory AI synthesis is under /ai/ with noindex,follow and sitemap: false.
Public AI-assisted work is not silently presented as organic writing.
Markdown, YAML, links, rendered HTML, and prose quality checks pass.

The validator runs this checklist against the repository’s flagged articles. The layout test verifies that every flagged article receives the same visible notice. This creates an auditable floor. Human review remains the final gate.

External guidance

This policy follows the direction of primary guidance rather than treating a disclosure as a marketing label:

Google Search guidance on helpful, people-first content recommends clear authorship and context about how automation was used when a reader could reasonably ask how the content was created.
Google Search guidance on generative AI content emphasizes accuracy, quality, relevance, and added value over the mere use of an AI tool.
UNESCO’s Recommendation on the Ethics of Artificial Intelligence centers human dignity, transparency, fairness, and human oversight.
C2PA Content Credentials provides a machine-readable vocabulary for AI disclosure, provenance, and levels of human oversight. It is especially useful for media assets. For Markdown articles, frontmatter, stable source links, and visible notices are the practical equivalent.

This policy is itself human-led and AI-assisted. Mike Hall supplied the intent, the publication requirements, and the final editorial judgment. AI assistance helped organize the policy and its checklist.