Every piece on The Scaffold is written with AI-based language models operating as a cognitive layer. The division of labor is fixed and does not vary by article.
I formulate the question, define the conceptual framing, choose the theoretical anchors and main references, and exercise full judgment over content, emphasis, and conclusions. The models perform structured drafting, synthesis, iteration, source search, and refinement of text in response to directed prompts.
Authority, judgment, and accountability for every claim published here remain entirely mine. If something on this site is wrong, it is wrong because I approved it.
No piece starts from a blank page
Each one begins with something that already exists: an artifact, a historical case, a framework someone built and abandoned. The work is altering it structurally until it says something about now. The models are useful in that alteration. They are not useful in choosing what deserves altering.
Every piece runs an adversarial review before it ships
Five standing positions, each with a fixed mandate:
A systems thinker who requires that complexity be acknowledged rather than flattened.
A methodologist who requires evidence for assertions.
A human advocate who requires the argument stay about people.
A political economist who requires distributional accounting of who gains and who loses.
An operator who requires technical accuracy.
None of them argue about whether the transformation is coming. They exist to make the argument survive someone who disagrees. This review is where the models do their most useful work, and it functions the way a second reader does in academic practice: to reduce the blind spots of any single perspective, including the model's own.
What the tools did not compress
Over the first year, cadence went the wrong way. Weekly in the fall, monthly by spring, and recent pieces run in three parts. Research and critique compressed enormously. The part where you apply something, watch it fail, and understand why did not compress at all. Twelve months is still twelve months. That gap is the subject of much of what gets published here, and it is also the honest limit of what the cognitive layer can do.
Modes
Each article carries a tag indicating which mode applied:
Directed Drafting. Framing, anchors, and judgment mine. Drafting and refinement performed by the models under directed prompts.
Research Assisted. Prose written by me. Models used for source search, fact checking, or synthesis of background material.
Unassisted. Written without AI involvement.
Why this page exists
Disclosure that says only "AI was used" tells a reader nothing worth knowing. What matters is which cognitive work was delegated, which was retained, and where accountability sits. This page is applied retroactively across the full archive. Where I could not verify how an older piece was produced, it is tagged conservatively.
Questions or corrections: [email protected]