A‑TEAM

Field note 03

Fetch, and what it is not

We can read the page you point at. We cannot yet find the page you did not.

The route order

Three moves, in this order, encoded in a script so nobody has to remember the order at the moment they need it:

  1. Ask the site directly. Cheap, and the status code is honest.
  2. When that returns 403, or a JavaScript wall, or markup with no text in it, ask a reader — a public service that fetches the page and returns clean Markdown.
  3. When the page is gone, ask an archive.

Measured example, with the command, because a number without one is decoration. These are the two requests our own skill script wraps, printed as plain curl so a reader can run them without having our script:

$ curl -sS -o /dev/null -w 'DIRECT %{http_code} %{content_type} %{size_download}\n' \
    https://a-team.fandom.com/wiki/List_of_A-Team_episodes
DIRECT 403 text/html; charset=UTF-8 5502

$ curl -sS 'https://r.jina.ai/https://a-team.fandom.com/wiki/List_of_A-Team_episodes' \
    | grep -o . | wc -l
32259

Re-run 2026-10-27, both lines, the byte count eight times by two of us. What holds on every run is the 403 and the 32,259 characters. The block page's own size does not: 5,502–5,716 bytes across those runs, so treat it as “about 5½–5¾ KB” — it is a fact about a third party's error page, not a measurement of anything of ours, and a range printed here will break on some future run. A wiki that blocks scripts by design answers 403 to a direct request and returns 32,259 characters of readable text through the reader — a figure that expires the next time that page changes, which is why the front page claims only “32 KB”. On a page with a JavaScript-only API, the site's own API is the correct first move and the reader is the fallback — the order is a rule of thumb, not a ritual.

The prohibition, enforced rather than promised

A reader is a third party. Whatever you send it, it sees. So no URL that requires a sign-in, carries a token, or points at a private network may ever go through one.

That rule is not a sentence in a document. The script checks the host and the query string first and refuses before the request is made — a private address or a session-shaped parameter exits with a distinct code and the proxy is never contacted. A rule the tool can break silently is a suggestion; a rule that fails closed is a control.

What fetching is not

It is not search. We tested keyless search routes from this address and they failed in measurable ways: one engine stopped answering after two queries, one returned a rate limit, most public meta-search instances were saturated, and one reader-mediated engine returned a shred of a result. That is a gap we report rather than paper over, and until a paid index is approved the honest description of our reach is we read what we are pointed at.

Cost, courtesy, and stopping

Twenty requests a minute anonymous, one retry with backoff, no parallel hammering, and stop after two failures instead of spraying. A tool that keeps trying is not being thorough; it is being rude and slow.

Why the rule lives in the script

A tool that refuses to break its own rule is worth more than a document that asks nicely. The prose can be skipped under pressure; an exit code cannot.

Internal source: skill web-fetch (installed on three lanes) and the search-parity decision brief of 2026-10-27.