[taras]

Why

The Implementation fixture intermittently exceeds its 120-second deadline in CI (#1143). Its workflow repeatedly checks HEAD and tracked changes. The staged and unstaged reads were independent but ran in sequence.

What changes

Ticket.Checkout runs its existing staged and unstaged Git reads in two Markdown <Spawn> branches under <All>. It joins their JSON in authored order and returns the existing HEAD, clean flag, and change pages. Each branch has its own durable coroutine, so an unfinished REPL entry can replay either completion order.

What must stay true

• The five fixture groups, assertions, real process paths, and deadlines are preserved.
• Each Git read retains its contained observation and private index. Every existing role transition still checks fresh source.
• Failure and cancellation join both branches.
• The Markdown component works in hosts that load Markdown components.

How to verify it

• The scheduling regression fails against the original serial component and passes with both reads active. The focused Deno suite checks checkout values, fresh reads, index purity, foreign-checkout refusal, joined cleanup, and exact filenames through JSON.
• RGO6 runs the actual component in an unfinished REPL entry and reopens it with the read arrival order reversed. It fails against the first draft’s TypeScript helper, reproducing the reviewer’s divergence finding, and passes with separate durable identities and zero native reads on reopen.
• The two REPL files pass on Deno, Node, and pinned Bun 1.4.0. The Bun run passed all 16 tests.
• The attributed local Bun fixture fell from 103.4 to 91.7 seconds with the same 92 checkout checks, 184 staged and unstaged reads, and 690 process calls. Checkout time fell from 21.0 to 14.4 seconds. The uninstrumented Node fixture passed in 93.6 seconds.
• Delivery, Artifact Return, Artifact Return Recovery, and Decision Resolution passed under Deno. Focused type, Git observation, formatting, and dead-code checks passed.

Risks and limitations

The first draft introduced loader and replay regressions. Independent Fable review identified both, and CI reproduced the loader failure across Deno, Node, and Bun. Commit a40e68b replaces that helper with the Markdown composition and adds the discriminating replay regression.

The original causal requirement remains open. In a matched, instrumented Linux pair on one worker, the original serial carrier passed in 93.1 seconds and this correction passed in 90.7 seconds. Both made the same 92 checkout checks and 690 process calls. Time inside checkout fell from 8.95 to 8.30 seconds, a 0.65-second reduction. The baseline did not time out, and this single fixed-order comparison is subject to cache and instrumentation effects. It does not establish the historical timeout’s cause or a reliable remedy. This PR remains a draft while the original acceptance and delivery gates are unresolved.

Scope confirmation

• Every changed file supports this checkout optimization and its regression coverage.
• Unrelated cleanup and formatting changes are excluded.
• The description matches the current diff and focused results.