[fadi-dev-1]

Workstream

D210, its row in docs/governance/workstreams.md.

What it changes

On a project, nstack claim D5 in an accepted cycle answered work continues on d5-…, which exists: no other branch is made. That branch was the start branch of an earlier cycle's D5. It had another title, its only commit was the empty Start D5 — …, and it had never been deleted on the remote. Since D149, any d<n>-… branch counts as the deliverable's work. Once that branch was deleted, claiming again only renewed the hold, and still made no work branch.

flowchart LR
    C["nstack claim D5"] --> E{"d5-… branch?"}
    E -->|"first commit: Start D5 — another title"| X["an earlier cycle's: ignored"]
    X --> M
    E -->|"hand-named, or this title's start"| K["work continues on it"]
    E -->|"none"| M["work branch d5-‹this title› made"]
    R["renewed hold"] --> E2{"work branch left?"}
    E2 -->|"yes"| S["nothing said"]
    E2 -->|"none, not merged"| M
    E2 -->|"none, merged this cycle"| N["none made"]

Loading

Legend: on the left, a claim. Below, a renewal by the holding session. Before this change, the renewal never reached the branch, and every d5-… branch was taken.

• hold._started_for_another: an empty branch (one commit off the default branch, its tree unchanged — the test _empty_starts reads, now the shared _empty) whose commit is a claim's start for another title of the same id is an earlier cycle's. Neither the remote's copy nor this checkout's is taken. A hand-named branch (d1-log-audit, D149), this deliverable's own start, and a branch with work whose deliverable was retitled since are still taken.
• hold._work_branch(renew=True): a renewed hold with no work branch left gets its first one again, unless this cycle's default branch has already merged the deliverable. Without that guard, a renewal after the merge would bring back an empty start branch. With a branch, the renewal prints nothing about it, as before.
• skeleton/.nstack/handbook/project.md, Who holds a deliverable, says both.

What a project receives

• Engine: claim ignores an earlier cycle's start branch, and a renewal can make a work branch.
• Skeleton: nstack update merges two sentences into project.md.
• Nothing that was green turns red.

Definition of Done

• The behaviour is covered by a test that failed before — committed first, seen red (P5)
• Every failure names the rule, the place and the action (P6) — no new failure message
• uv run bash platform/tests/run.sh green, on a committed tree
• uv run nstack fitness green
• Affected documentation updated, diagrams and legends included; no dead reference
• The workstream's row updated in the register; its plan deleted, if it had one
• Brought current with main as the last thing before opening
• No private project named, no path of a workstation, no link into an assistant's session

Summary

DONE

• An earlier cycle's start branch of the same id is no longer taken as the deliverable's work.
• A renewed hold with no work branch left gets one again, unless the deliverable is merged.
• project.md says it; register row D210.

VERIFIED

• Six new tests in platform/tests/test_hold.py:
  • seen red before the fix, then green: another cycle's start branch ignored; a renewed hold with none makes one;
  • green before and after, pinning what must not change: this title's own start branch still taken; a renewal with a branch prints nothing;
  • the merge guard: seen red with the guard disabled.
• test_hold.py: 208 passed.
• uv run bash platform/tests/run.sh: 1269 passed, Platform tests: OK, at the head.
• uv run nstack fitness --root .: compliant. nstack pr-scope: inside the review budget.

ASSUMED

• A claim's start commit subject, Start D<n> — <title>, identifies the cycle well enough. The two cycles' D had different titles.

NOT VERIFIED

• Not run against the project that exposed it; the tests reproduce the case.
• Not checked by an independent verifier.

RISKS

• These branches are still taken: an earlier cycle's D with the same title, or a stale branch with real work on top of a start of this title. Start commits do not name their cycle.
• If the new title's 40-character slug equals the stale branch's name, the push is refused and the claim says NOT pushed. Not tested.
• nstack plan's Empty start branches left (_empty_starts) still counts such a stale branch of a takeable id as this cycle's. Left unchanged, out of this scope.

🤖 Generated with Claude Code