backpacks founded, with 21 packs and a source registry that fetches nothing

backpacks is now part of Bussetech Software Studio: an aggregator and exploration tool for technical and boutique carry — makers, packs, and the lines they belong to, with every spec traced to the page it was read on.

The founding corpus is 21 packs from six makers — Mystery Ranch, Evergoods, GoRuck, Alpha One Niner, Filip Raboch, and Sample — resolved from 105 signals, each one a single claim from a single source.

Three things are worth saying plainly at the start.

The source registry fetches nothing. We reviewed ten domains before recording a single spec: read each robots.txt, read each terms of service, tried each feed. Every one of them ended up marked manual. Evergoods and GoRuck prohibit crawling in their terms. Alpha One Niner asks that its materials not be copied. Mystery Ranch publishes no feed at all. And Carryology, The Perfect Pack and Suburban — who publish perfectly permissive feeds — are the people who have documented this subject for years, so we read them and credit them and do not aggregate them on a cron. No scheduled fetcher will ever touch a source in this registry.

The corpus itself was gathered by an AI agent, not by a person. It fetched each cited page once, on request; nothing was crawled and nothing was polled. The first version of this post said the dataset was “hand-carried”, which sounded better and was not true. It also leaves a question we have not settled: Evergoods and GoRuck forbid “spider, crawl, or scrape”, and a single agent fetch is not spidering — but we wrote the rule that a permissive robots.txt does not license what an operator forbids in prose, and we owe it to that rule not to grade our own homework. The question is recorded as an open studio decision. If either maker would rather we had not, their specifications come down.

The gaps are on the record. A pack whose weight the maker never published says so, on its page, in a list titled what we don’t know. An empty field reads like an answer; a named gap reads like the truth.

Two of our own errors are on the record too. Our founding brief named a designer who does not exist — the maker is Filip Raboch, not “Robach”. And it described SAMPLE and ARTICLE as two product lines, when Sample is the brand and Article is how Sample numbers its models. Both were caught by reading sources before seeding data, and both corrections are visible rather than quietly swept up. If this dataset is going to ask anyone to trust its provenance, it can start by showing its own.

There is a fourth thing, and it is the one we like most. Sample’s product page for the Article 404C says 20L in its title and 18 L in its spec block. The same page, disagreeing with itself. We record both, cite the one URL twice, and resolve nothing. That is the whole design in miniature: signal > canon.

No affiliate links. No monetization plumbing. Whether there should ever be any is a real question, and the studio has written it down without answering it.

Start with the packs, or read how the data was collected.