The Drift Race: methodology and measured results

Adding is safe. Moving is what breaks a site.

For eleven operations the two sites are indistinguishable. Then comes the first reorganisation that moves pages other pages point at, and the raw site never recovers — not even after a final operation whose only job was to find and fix broken links.

In raw files a link is a literal path, so moving a page silently invalidates every reference to it. In RIFT a link is a reference to a page, so moving it repoints every inbound link automatically.

The residue is specific. Eight of raw's dead links point at two pages — a glossary and a support page — from four concept pages each. Those two moved, and nothing that pointed at them was told.

The failure a review would not have caught

The worst damage in the raw run was not a missing page. It was this, written into eight pages:

Backslashes where quotes belong. The stylesheet never loads. One page in six was serving with no design system at all — and still returning 200. Those eight pages are the difference between raw's eleven broken links and its score of nineteen. RIFT rejects that markup at the moment it is written, so the governed run could not produce it.

Where we were wrong

We predicted RIFT would reduce style forks. It tripled them — 28 against raw's 9, in both governed runs. The published snapshots say precisely where they came from.

Through twenty-eight operations the governed site carried six. Operation twenty-nine took it to twenty-eight. That operation asked for one thing:

The design system had no component for a review stamp. So the model wrote one by hand, eleven times — once per API page — and each copy carried its own inline style rule and its own hard-coded colour. Two forks per page, eleven pages, twenty-two forks in a single operation.

Governance stops a model writing a broken link. It does not stop a model inventing styling it cannot find. That is a gap in the design system, not in the governance — and the fork count is better read as a measure of design-system coverage than of governance.

Removing the twenty-two forks attributable to that one operation would leave the governed run at six against raw's nine. We are not claiming that as a result: it did not happen, and the missing component is our responsibility. But it does locate the fix — one component, not a change to how governance works.

What the test measures

After every one of the thirty operations, the whole published site is crawled and audited from scratch. Not a diff against the previous version — an independent examination of the site as it stands, so a fault introduced at operation twelve and never repaired is still counted at operation thirty.

Seven measures are recorded. Three are reported here because they are the ones a content team feels: broken references, style forks and token cost. The others cover header and footer consistency, pages nothing links to, how many pages each change disturbed, and whether the change that was asked for happened at all.

That last one matters more than it sounds. An arm that quietly skips an operation damages nothing, and would otherwise score beautifully. So each operation carries a check written from its own instruction — a page that was supposed to move is confirmed moved — and an arm cannot look good by doing nothing.