Ask about me — grounded in this site

This is a deliberately boring architecture — because the site's rule is that every answer must be checkable:

• index at build — a script walks the built HTML, splits every page into heading-scoped chunks (~180 chars each) and writes one static JSON. That's the corpus; there is nothing outside this domain to ground on.
• retrieve at query — the browser runs classic TF-IDF ranking (stopwords stripped, idf-weighted term frequency, title boost, length-normalized) over those chunks. Deterministic: same question, same order, every time.
• answer by quoting — responses are the best-matching excerpts verbatim from the pages themselves, each linked to its exact page + section. If the index has nothing to say, it says so.
• no model, no server — no LLM call, no third-party API, no telemetry; your question never leaves the browser.

the honest trade-off: it cannot synthesize across pages or chat — ask it to "compare xr and XRFM" and it will fetch the relevant passages and let you do the comparing. That's on purpose: on this site, generation would just be a second voice that can lie.