Crystallize the cross-domain "same rot" argument in the Story

> _Posted by Claude (AI assistant) on behalf of @jmchilton — not authored by them personally._

The Story section **"The cost of letting the skill be the source"** used to carry two worked illustrations — *workflow construction* (hallucinated tool IDs, dropped revisions, parser-rejected docs) and *statistical method selection* (an agent inventing a plausible-named method and certifying its own output). Dropped them: the specifics are real but read as silly/unillustrative and aren't convincing.

The argument underneath is worth keeping:

- the skill-rot failure is the **same in every structured domain**, not a quirk of one;
- the failures are **mechanically detectable**;
- prose papering over them **isn't durable** — because the thing doing the papering is also the thing being trusted.

**Want:** a crisper, more convincing way to make that cross-domain point, ideally without two arbitrary domain vignettes.

Directions to weigh:
- one sharper single illustration instead of two;
- an abstract statement of the invariant (no vignette);
- lean on real evidence already on the site — the two actual Foundries in Instances / the Diff demonstrate "same substrate, same rot" concretely; point there instead of inventing examples.

Start with the Story section, especially “The cost of letting the skill be the source,” then compare the two actual Foundries in Instances / Diff. Revise the section so the cross-domain invariant and mechanical detectability are convincing without relying on arbitrary vignettes, and confirm the resulting argument remains consistent with the site’s evidence.

Written by the indexing model from the issue text.