The site republishes itself every half hour, and a broken link stops the publish.
IN PRODUCTION
This runs in client systems today. The sample below was regenerated for a fictional company.
A manifest diff decides what needs rebuilding, only that gets regenerated, and then every link is audited. A blocking fault ends the run instead of shipping. Afterwards the hashes of what was built are compared to what the host is actually serving, because a successful publish and a correct site are not the same claim.
TRY IT / RUNS IN YOUR BROWSER
Use it here.
The mechanism is the production one. The company, the catalogue, and every figure are fictional.
Loading the build manifests, the link table, and what the host is serving.
THE OUTPUT / GENERATED FOR APEX INSTRUMENTS
What comes out.
MJS
The pipeline, four named stagesData · Node ESM, no dependencies. diff, regenerate, audit, parity. Exits non zero when the audit finds a blocking fault, which is the whole mechanism. node run.mjs writes data/run-report.json and exits 1. node run.mjs --fix-fixtures applies every remedy the audit computed, writes data/links.fixed.json and data/run-report.clean.json, and exits 0. node run.mjs --sync-widget copies the core into widget.html.Open the file ↗
JSON
Two build manifests, thirty minutes apartData · 232 entries in build 14:00 and 233 in build 14:30, each with a path, a content hash, a byte size and the time it was last built. 93 pages and 140 assets. The current manifest also carries the rename the asset step recorded and the page the content step marked as superseded, which is how the audit can compute a fix rather than only report a fault.Open the file ↗
JSON
Every link the build emitted, 417 of themData · Page, target, kind and link text for 234 internal links, 117 asset references, 29 in page anchors and 37 external addresses. Carries the heading ids each page emitted and the status the external link check recorded at 14:12, so anchors and external addresses are resolved by lookup rather than asserted. Nine faults are planted in it.Open the file ↗
JSON
What the host is serving right nowData · 233 paths with the content hash the host answers with, read at 14:29. Two hashes differ from the last build that published and one path is served by no build at all.Open the file ↗
JSON
The run that abortedData · Written by run.mjs, not by hand. Every stage's counts, all 9 faults with the page, the target, what the audit found and the fix it computed, all 3 parity mismatches, and the verdict with exit code 1.Open the file ↗
JSON
The run that publishedData · The same four stages against data/links.fixed.json. 415 links, 415 resolve, 0 faults, exit 0. Parity still reports the same three mismatches, because fixing links does not fix a host.Open the file ↗
JSON
The link table with the remedies appliedData · 7 targets rewritten and 2 links taken off the page, computed by the audit and written by run.mjs --fix-fixtures. This is the second fixture set, so the clean run is a real run and not a claim.Open the file ↗
MJS
The fixture generatorData · Deterministic. Every hash is a sha256 of a stable seed string and every planted fault is placed by a named function, so re-running produces byte identical files.Open the file ↗
JSON
What the generator plantedData · The audit never reads this file. verify.mjs uses it to confirm the audit found exactly these faults and nothing else, which is what stops the demonstration from being a list of findings written by hand.Open the file ↗
Compare the manifests and decide what changed. Nothing else is touched.
02Regenerate
Rebuild only the changed pages, and say which and why.
03Audit
Walk every link. A dead internal link, a renamed asset, a missing anchor are all faults.
04Abort
A blocking fault ends the run with a non-zero exit. Nothing is published.
05Check parity
Compare built hashes against what the host serves, so drift is visible rather than assumed.
WHAT IT READS
The previous and current build manifests
Every link the build contains
The hashes the host is serving
WHAT IT CHECKS BEFORE IT SHIPS
A dead link stops the publish instead of reaching a reader
Only changed pages are regenerated
Hosted content is compared to built content, not trusted
Every fault names the page and the target
In production: a self-publishing pipeline that runs on a half-hour cycle, diffs the manifest, regenerates what changed, and aborts on a link audit failure.
The mechanism runs inside client systems. Everything on this page was regenerated for Apex Instruments, a company that does not exist. Client names, source files, and internal wording stay private.
