The public status page delegates all its numbers off-site, and nothing checks the destination

• Type: task — one shippable deliverable, scoped below
• Track: D — D owns the prose that makes the promise. The publishing machinery is W's and the generators are T's; this ticket touches neither.
• Found: 2026-08-30 by frankD, spinning a concrete unit out of idea-public-status-page [D p25].

First, the part of the parent that is already done

The parent idea (filed when status.md was hand-written prose) says:

"Good as a narrative, but the numbers rot — they are a manual snapshot of gates that move every day."

That is no longer true, and the fix was the parent's own option (a). ce89ff14b — "docs(D): stop hand-writing numbers in status.md" — removed the figures. The page now opens:

"For current numbers, read the live status pages, not this page. https://pxxc.org/status/ is generated from the test manager's own output on every content pull ... Deliberately no figures — a number written here is a number that starts going stale the day it is written."

So the parent's gap item 1 (publish path) and item 2 (freshness) are both discharged: the live pages exist, docs/** links to them from two files, and status.md is now the stable narrative the idea proposed it become. It even carries a "What this page can still get wrong" section naming the residual rot direction correctly (prose understates what works, because a feature ships and the sentence denying it is never revisited).

The parent's remaining concern is not this. It also asked that the claims discipline "survive templating" in whatever is auto-generated. Checked: tools/twatch_web.py and tools/testmgr.py contain no "byte-identical", "clone", "compatible" or parity language at all — the generated pages carry numbers, the claims live in docs/** prose, and that prose was audited against CLAUDE.md on 2026-08-29–30. Nothing to do there either.

Recommend the parent be resolved rather than worked. Not doing so here: the dispatch was explicitly to spin out a unit, not to claim the idea.

What actually remains — the exposure the fix created

Removing the figures was right, and it traded one failure mode for another:

That is the better trade. It is not a free one, and the residual has never been checked once.

The measurement

• docs/** references six distinct pxxc.org/status/* URLs across two files — /status/, /status/conformance/ (twice), /status/tests/, /status/benchmarks/ from docs/reference/status.md, and /status/flow/ from docs/targets/nil-python.md:29.
• Plus three github.com/yoctobyte/pxx links, two of them to blob/master/LICENSE.md.
• tools/ contains no link checker — nothing matching link, url or href. tools/docaudit.py checks internal citations (compiler/…, lib/…) and tools/docsnip.py compiles code blocks; neither looks at an http URL.
• The generated HTML is not in this checkout. twatch_web.py --static writes it inside the watcher's own clone, and devdocs/progress/tstate/ here holds only the TSV/JSON inputs. So there is no local artefact to diff against either.

status.md does not merely link to those URLs — it delegates its entire numeric content to them. If /status/conformance/ is missing, the public compatibility page has no figures and no working pointer to any, while explicitly telling the reader that is where the figures are. That is strictly worse than the hand-maintained snapshot it replaced, and it is the one outcome nobody would notice from inside the repo.

The deliverable — one thing

A link check over docs/**, wired into Track D's gate. Modelled on the two checkers already shipped for this lane, and deliberately no larger:

1. Extract every http(s):// URL from docs/** (the extractor must skip markdown artefacts — a naive regex picks up https:// from inline code fences, and example.com appears twice as an intentional placeholder).
2. Request each one; report non-2xx, with the file and line.
3. For the pxxc.org/status/* set only, assert the response actually looks like the report it is promised to be — a conformance page containing pass/fail counts, not a landing page or a redirect to one. This is the half that matters: a 200 that renders the wrong thing is exactly as bad as a 404 and a plain status check will not see it.
4. Run it from docs/ work, not from every gate — it is the only check in this lane that needs the network, and it must degrade to SKIP (never FAIL) when offline, or it converts a lane that works on a train into one that does not.

Scope notes

• Not a website change. If a URL is wrong, the fix is either the link in docs/** (D) or the publish path (W) — this ticket only makes the breakage visible, and files what it finds into the owning lane.
• Not a general docs link-farm. Ten URLs across 41 files; the value is entirely in the six that carry delegated content, and the GitHub ones are nearly free to include.
• The offline-SKIP rule in step 4 is load-bearing, not politeness. A network check that hard-fails is a check people delete.

Gate

The checker runs, reports the six status URLs and the GitHub links, and skips cleanly with no network. No compiler build, no lib/**.

Done — tools/doclinks.py, 2026-08-30

Every URL was verified by fetching it, not by reading what the doc says it links to. All eight distinct external links in docs/** resolve, and all five content-marker assertions pass:

docs: 13 external link(s), 8 distinct
  ok   https://github.com/yoctobyte/pxx
  ok   https://github.com/yoctobyte/pxx/blob/master/LICENSE.md
  ok   https://pxxc.org
  ok   https://pxxc.org/status/              [backlog resolved]
  ok   https://pxxc.org/status/benchmarks/   [fib sieve]
  ok   https://pxxc.org/status/conformance/  [pass fail]
  ok   https://pxxc.org/status/flow/         [filed closed]
  ok   https://pxxc.org/status/tests/        [GREEN RED]
checked 8, BROKEN 0

So the feared failure is not occurring — status.md's delegation is honest today. /status/conformance/ carries real per-category pass/fail/skip counts across 25 t* categories, /status/tests/ the four watcher hosts with SHAs and GREEN/RED verdicts, /status/benchmarks/ per--O-level timings against FPC, /status/flow/ the filed-vs-closed curves. That does not retire the check; it establishes the baseline the check defends.

All four paths were exercised, not just the green one

Publishing an untested failure path would repeat the defect found the same morning in the fact sheet — a command nobody ran reads as verification.

The third row is the one the tool exists for and the one a plain status check cannot see. The fourth is why it will still be here in a month.

Also confirmed ignored, both of which would otherwise be reported forever: the two example.com placeholders, and the https:// fragment a naive regex lifts out of an inline code fence.

One divergence, reported not fixed

The published dashboard says 338 backlog; tools/factsheet.sh says 351. Both are defensible — factsheet adds backlog_new/ (13), the dashboard appears to break those out separately alongside "20 experimental". Not a defect, and not D's call, but two public surfaces answer "how big is the backlog" differently and whoever owns the generator (T/W) should decide which is meant. Filed here rather than as its own ticket because it is one sentence and may be intentional.

Not done, deliberately

Not wired into a gate. It is the only check in this lane needing the network, it takes ~8 round trips, and its value is weekly rather than per-commit — running it from docs/ work is what the ticket asked for and what the exit codes support.

Resolves the spin-out. The parent idea-public-status-page is separately discharged (see this ticket's opening section) and is a resolve, not work.

Provenance

Spun out of idea-public-status-page [D p25] on 2026-08-30. Every claim above about status.md's current content is from the file at HEAD, and the "no claim language in the generators" finding is from grepping tools/twatch_web.py and tools/testmgr.py directly — not from the parent ticket's description, which is two rewrites out of date and is the reason this spin-out reads the way it does.

Log

• 2026-08-30 — resolved, commit 3496da031.

2026-09-05 — the accepted risk fired, for the first time, six days in

This ticket's central trade has now been tested by events. ce89ff14b removed the figures from status.md and priced the exchange explicitly:

It was the better trade and still is. The residual — the right-hand column — had never once occurred when this ticket was resolved. On 2026-09-05 it did: https://pxxc.org/benchmarks — precisely, /status/benchmarks/ — began answering 503 while /, /status/ and /status/tests/ all served 200. Filed as [[bug-w-status-benchmarks-503s-while-every-sibling-page-serves]].

The baseline above is what made it diagnosable, and that is the reusable part. This ticket did not merely record that the URL resolved; it recorded the content marker — ok https://pxxc.org/status/benchmarks/ [fib sieve] — so six days later the claim available was not "a page is down" but "a page that served the real per--O timings on 2026-08-30 now 503s". The first is unbounded: was it ever up, is the URL right, did the docs invent it. The second has a bounded window and two named suspects, the generator or the deploy, and the marker is what rules out "served a shell". A verification record is worth recording even when it is green, because green is the baseline a later red is measured against.

Two data points for whoever revisits the trade, which is the whole reason to write this here rather than leave it in the newer ticket:

1. It fired once in six days of the new regime. That is a real rate, not a theoretical exposure, and it is the number this decision was made without.
2. doclinks.py caught it — the mitigation this ticket built is the reason the failure was visible at all, which is the assumption the trade rested on and is now evidence rather than intent.

So the trade stands, and it stands for a better reason than when it was made. The checker remains deliberately un-gated (network, ~8 round trips, weekly value); the offline-SKIP rule is what kept it in the tree long enough to be the thing that noticed.