When the model goes away: single-provider dependency at scale-of-one

## Spark

Follow-up to #253, closed as obsolete. That issue wanted a dated [short] on the US
export-control directive that took Fable 5 and Mythos 5 offline for 19 days
(2026-06-12 → 2026-07-01). The news cooled before it could be drafted; the question it
raised about my own stack didn't.

## Why it could be interesting

"Off the Desk" ends with everything consolidated onto one provider, because Claude was
the only model I ever reached for. The claim here is that the consolidation carries a
failure mode I never priced: a provider can lose a model on five hours' notice for
reasons that have nothing to do with the provider, the model, or me. The post isn't the
politics — it's the inventory. Most of what I built around Claude is portable (standing
instructions, skills, conventions, the habit of driving work from a phone); some of it
isn't; and a one-person operation has to decide whether carrying a warm second provider
is worth it, or whether the honest answer is to accept the outage and wait it out.

## Open questions

- [ ] What actually breaks if the primary model vanishes for 19 days? Inventory the
tooling and sort it portable vs. not — that inventory is probably the spine.
- [ ] Is a second provider worth carrying at scale-of-one, or is "accept the outage"
the correct answer? Needs an honest cost read, not a reflexive redundancy answer.
- [ ] Stay personal-inventory, or borrow the resilience frame from #339 (Hollnagel:
respond, monitor, learn, anticipate)? Risk of the essay swallowing the inventory.
- [ ] Sequencing: publish after "Off the Desk" (2026-08-09), which sets up the
consolidation this post examines.

## Source material

- #253 (closed as obsolete) and its source cluster — the Anthropic statement, Simon
Willison, Snyk, and the "export controls harm US cyber defense" piece
- — the suspension statement
- — the restoration
- `src/data/blog/off-the-desk.md` (scheduled 2026-08-09) — the setup this builds on
- Related: #190 (supply-chain hardening), #128 (repos as a fleet), #339 (resilience
engineering as a personal-ops lens)