Build Index is a publication about moving from something that demonstrably works to something that operates. It pairs four short guides with a private brief builder, and it exhibits three public working sites that show the kind of ground the method stands on. Nothing here is offered as a client result, a saving, or a credential.
A demo proves the idea survives contact. An operating system is what remains after the demo is someone else's problem. This guide separates the two and asks what the demo genuinely demonstrated.
An acceptance test written after the build tends to describe the build. Writing it first forces the question of what would count as proof, and of what you would accept as a fair failure.
Automation rarely removes a person from a workflow. It moves where the person stands. This guide is about naming that new position, what it must watch, and how it recovers.
A pilot is a bounded test with a defined owner and a stated horizon. This guide treats scope as a promise you can keep rather than a surface you can cover.
03
Working exhibits
Public pages, linked as they are published. Each annotation reports only what the page itself states. Nothing here is a client, a case study, or a measured outcome.
Publishes a worksheet titled “One outlier can move an average” in which a user enters 3–200 finite numbers and compares mean and median with and without one identified outlier removed. The page states the worksheet runs entirely in the browser, with entries not uploaded or stored, and that its outputs are illustrative rather than evidence about the real world.
Annotation Removing the outlier is presented as a judgment call, shown to demonstrate effect and not as a recommendation. Optional resampling accepts sample sizes 2–100 and 50–500 repeats with a seed that reproduces the same plot for the same settings.
Publishes a small, MIT-licensed reference implementation of store, retrieve, provenance and consolidate operations, grounded in a reproducible test fixture. Store is described as an append-only ledger with explicit event IDs and timestamps; retrieve as BM25 or tag-based ranked results; provenance as source, event ID and insertion time per item.
Annotation The page reports fixture observations on a 16-entry synthetic corpus (seed 7, k=5) and states plainly that these are not a benchmark. Full JSON output is stated to differ between runs, with the cause not yet diagnosed; no bit-reproducibility, Python 3.9 compatibility or external performance guarantee is claimed. The reference is described as intentionally small and not production-grade.
Publishes “Handoff”, a desk for mapping one workflow by minutes: work minutes and wait minutes per stage across two to eight stages, returning totals, a wait share, the largest wait with ties named, a text table and a downloadable Markdown plan. The page states nothing is uploaded and nothing is saved.
Annotation The before-and-after invoice example on that page is explicitly stated to be invented, unmeasured, and not the result of any engagement. Naming the largest wait is offered as a candidate for a small test, not a measured saving.
04
Write the brief, keep the gaps
The builder is private and client-side. Its value is not a polished plan but an honest one.
Six inputs, then two optional numbers. Anything you leave blank appears in the generated Markdown as an explicit unknown rather than being filled with a plausible assumption. You can download the brief and reset the form.
