VOID
I founded VOID to build operating plans, AI workflows and web systems. Atlas, a Python and SQLite audit console, connects website findings to the captured evidence behind them.
A business question, then a working system
Through VOID, I develop operating plans and software tools for small businesses and technology clients. The work brings together growth strategy, financial analysis, AI workflows and web systems. Atlas gives a concrete view of the engineering: a website audit workflow whose findings stay connected to the evidence behind them.
Atlas: a local evidence console
I built the Python crawler and local console around retained page state. URL normalization, deduplication, robots, and sitemaps feed a bounded run; SQLite keeps the records, and CSV/Markdown exports carry pages, links, findings, measurements, and collection events into review. A failed fetch remains distinct from a successful page with no extractable content.
An unchanged response can reuse cached work only while its artifact version is valid. Content hash, artifact type, and extractor/schema/scoring versions form the identity. A 304 can reuse a matching artifact; the same bytes must be parsed again when the interpretation changes.
A 200 response is not the whole page
The retained July 16, 2026 sample contains two successful responses with different evidence. Ten quote cards exist in static HTML; the JavaScript page embeds ten records but no source quote cards. The latter needs a render check before judging visible content. The implementation separately retains raw-only findings while using effective rendered facts for page conclusions.
| Capture | HTML quote cards | Embedded records |
|---|---|---|
| Static HTML | 10 | Not recorded |
| JavaScript source | 0 | 10 |
Saved HTML from July 16, 2026; not a fresh crawl, rendered-page measurement, or search-performance result.
Keep missing evidence visible
A collection gap must not become a bad website score. Provider states record skipped or partial collection, while findings retain page state and rule versions. The sample’s absent canonical is an observation without an assigned severity. This separation makes a recommendation traceable to the evidence actually collected.
Code and retained captures
The public repository documents the shipped local workflow. The pinned tests and July source sample establish inspectable behavior; hosted delivery, distributed crawling, rank tracking, and traffic gains are outside this case study.