The case-study documentation system
A repeatable, AI-assisted process for reconstructing a project's real history from primary sources and turning it into a portfolio-grade case study — with anti-fabrication rules that are the point of the whole thing.
Portfolio case studies are usually written from memory, months later, by the person with the most reason to flatter the work. That produces a document nobody can check. The question here was whether the opposite is possible: reconstruct what actually happened from evidence, and make the evidence part of the deliverable.
Evidence first, writing second — nothing is written until the evidence pass is complete across the repository, the strategy documents, the design tokens in the code, the live product and the print collateral. The code is treated as the authority on what the brand actually is. Every claim carries a confidence label: confirmed, observed, inferred, or unknown. Brand elements are never reconstructed — if the mark is a traced path in the code, that exact path is used; if a token file says a colour is surface-only and never text, the case study obeys it too. Dates come from version control, never from recollection. Anything that cannot be captured safely, such as the interior of a system holding real patient data, is documented from code and declared as not captured.
Five complete case studies, each with a master archive document, a set of rendered panels, captured screenshots, extracted brand assets, and an evidence inventory naming its primary sources and its stated limits.
Designed the process, wrote the prompt that drives it, and ran it across five projects.
The five archives this site's own case studies are written from. The rule that earns its keep is the stated-limits section: every study names what it could not verify, which is what makes the rest of it checkable.
In use. Rewritten once already, as version two.