# Review brief: temporary static sites implementation plan

You are an independent second reader of an implementation plan for the
CARLOS platform (this repository). Read cold and adversarially.

**Plan:** `docs/superpowers/plans/2026-09-26-temporary-static-sites.md`
at the commit named in the prompt.
**Spec it implements:** `docs/superpowers/specs/2026-09-25-temporary-static-sites-design.md`
(rev 9). The spec was reviewed by you over nine rounds and accepted, and
the product owner has approved it. **Do not re-review the design.** Its
rulings (§1) and mechanisms are fixed.

## What to check

The plan will be executed task by task by implementers who see only
their own task. Check:

1. **Fidelity.** Does each task implement the spec's mechanism as
   written? Flag any place where the plan weakens, reorders or omits
   something the spec requires, especially the orderings (§6 restore
   steps, §5 expire steps, settle-first, the tripwire rules) and every
   row of the §3 and §4 refusal tables.
2. **Will it compile and pass?** Code in the plan is written against
   this repo but has not been compiled. Check signatures, imports,
   receiver types, helper names and test fixtures against the real
   code. Cite `file:line`.
3. **Cross-task consistency.** Names and types a later task consumes
   must match what an earlier task produces.
4. **Tests that would pass for the wrong reason**, or that do not
   exercise the ordering or race they claim to.
5. **Existing behaviour broken.** Box code must not change, and no
   bucket object may change shape. Flag any existing test the plan
   would break without saying so.

## Output

A numbered list of findings. Give each one a severity (P1 means the
build is wrong or a spec guarantee is lost, P2 means wrong behaviour or
a test that proves nothing, P3 means clarity), the task and step, the
evidence with `file:line`, and a concrete fix. Then give a one-line
verdict: accept or revise. Do not rewrite the plan, and leave out
praise.

In later rounds you will be given your earlier findings files. Start
with a disposition table, then list only new findings.