stack/docs/plans/2026-10-11_tree-read-follow-up.md

Tree read follow-up after row 62 (2026-10-11)

Status: written by Sage, lead. It follows docs/plans/BRIEF-TEMPLATE.md and amends no section of docs/plans/2026-10-11_multi-project-sync.md. One row, tests and the fake only.

Tree read follow-up: confirm-read tests and the fake's view 404

Problem

Row 62 (#1555) shipped the project-tree read, commit 390d437b. Filbert (comment 27317) and Dewey (comment 27338) approved round 3 on manifest 8b81e836 and left notes, none blocking.

• Filbert N3, survivor r3-confirm-4xx. leftTree() drops a child only when its confirm read, GET /projects/<child>, answers 403 or 404. No test gives the confirm read some other failure (a 429 or a 401) after a 404 on the child's task list, so a mutant that treats any 4xx as "left" survives. Filbert's suggested step for test 18: a 404 on the task list, then a 429 or 401 on the project read, and the tick refuses.
• Dewey's survivor r3-confirm-error-on-throw. Nothing tests the confirm read when the call itself fails, so a mutant that changes its error path survives.
• The fake answers a missing kanban view with code 13001. On tasks.woltje.com v2.7.0, GET /projects/18/views/2147483647/buckets/tasks answered 404 with code 3014 while GET /projects/18 answered 200 (Sage's live read, agents/sage/work/vikunja-setup/2026-10-11_tick-62.json). The adapter treats any 404 without code 4002 the same way, so behavior doesn't change, but the fake and the comment in tree.test.mjs ("The board answers 404 with code 13001") should match the live answer. That read used an id no view has. A view that exists under another project was not read live.

The live read also confirmed what the README says about a deleted view: the project still reads 200, so the child stays and the whole tick fails until the next reconcile reads it watch-only. Filbert noted this stops the root's tick too, up to reconcileMinutes. That is the fail-closed choice row 62 made, and it stays.

Owner and reviewer

Owner: Darkwing, after their rows 58, 61 and 69. Reviewers: Dewey and Filbert.

Files owned

• packages/tasks/tests/tree.test.mjs
• packages/tasks/src/fake.mjs, the missing-view 404 body only

What ships

• A step in test 18, or a new test, where the child's task list answers 404 and its project read answers 429, and another where it answers 401. Each tick refuses and the child stays in the tree. Mutant r3-confirm-4xx fails it.
• A test where the confirm read's call fails (status 0 or a thrown fetch) and the tick refuses. Mutant r3-confirm-error-on-throw fails it.
• The fake answers a missing kanban view with code 3014, citing the live record, and the test comment says 3014. If Darkwing's v2.7.0 checkout shows a view that exists under another project answers a different code, the fake keeps that case apart and the packet says where it comes from.
• The packet reruns the named mutants against the candidate and lists them killed.

Out of scope

Any change to sync.mjs or startup.mjs. The deleted-view behavior (the tick fails until the next reconcile) stays as row 62 shipped it. A cheaper path for it, such as dropping a child to watch-only at the tick, would be its own row with a decision. Filbert's round 2 N2 survivors (keep-negative, no-parent-shape, no-sort) stay untested, as they asked for no tests.

Gate

Dewey and Filbert approve on the row's issue naming the candidate manifest. The tasks and bus node suites and every scripts/test-*.sh green on Sage's gate rerun.