Review mitigation for the q29p/n698/x0wf batch (nxf 6j6v.q29p, n698, x0wf)

[cabcookie]

Stacked on #48 (→ #47 → #46); this PR is the one to merge — it contains all four.

Mitigates the independent review of PR #48 (.nxf/review/2026-10-11T05-26-00Z.md) for the q29p / n698 / x0wf batch. The eight decisions of docs/specs/derived-ranking-keys.md §12 are untouched: §12 gains one bullet clarification (the untyped rank key is refused in the [ranking.list] default order too), and the now bullet now says the filters narrow the rows before they are ranked.

Mitigation Summary

Finding (source · severity)	Disposition	Notes
Derived keys read raw parent edges, not the projected single parent (Integrity #1 · Medium)	Fixed	Store::present_edges_of_kinds reads parent from the present_parent LWW view (one query, UNION ALL); core test + facade test with two concurrent re-parents (count and related … pick = "max")
list_query ranks before it filters (Code Quality #1 · Medium)	Fixed	read::list_selected applies status + type/container/label filters to the candidates before derived::compute; the rank scope stays every live item (§12 decision 2). Seam test: a goal/project type filter needs no now, person still does
Stale NextRank / n698 / "x0wf decides" lines (Code Quality #2 · Medium)	Fixed	docs/specs/next-finish-first-tiers.md rewritten; NextRank added to xtask/sweep-retired-vocabulary.sh; sweep run against origin/main: no file names NextRank any more
"Every paged read" prose leaves out list (Code Quality #3 · Low)	Fixed	next_cache.rs module doc, NextPage, Paging, page_through
Engine::list_query rustdoc silent on now (Code Quality #4 · Low)	Fixed	list_query / list_query_value document when now is needed and the malformed case
Guide says derived keys only in per-type orders (Code Quality #5 · Low)	Fixed	en + de plugins.md: the [ranking.list] default order takes them; the untyped rank-key refusal there; only list --sort rank reads the time
nxf list resolves now even when sorting by id (Code Quality #6 · Low)	Fixed	CLI and MCP resolve now only for --sort rank under a plugin with derived list keys
Ratio guard blind to uncounted reads (Test Quality #1 · Medium)	Fixed	EdgeReadStats gains per_item and bulk; every edge/label/custom-field read is counted (core test enumerates them); the guard asserts per_item == 0 and no growth across all four buckets
List rank scope ignoring --status not pinned (Test Quality #2 · Medium)	Fixed	a_list_rank_scope_ignores_the_status_filter (closed box still ranks); mutation list: Some(&items) goes red
Cross-section part of the paging fingerprint untested (Test Quality #3 · Medium)	Fixed	a_derived_list_token_restarts_when_only_the_next_order_changes; dropping +other goes red
No stale-token test for an outside op that reorders (Test Quality #4 · Medium)	Fixed	a_page_restarts_when_an_op_outside_the_snapshot_reorders_it (a goal's order moves the actions)
Derived key in [ranking.list] default order only tested at load (Test Quality #5 · Medium)	Fixed	a_derived_key_ranks_in_the_list_default_order_without_per_type_tables
Determinism test cannot fail / has no full tie (Test Quality #6 · Low)	Fixed (in part)	Added a real full tie, checked from both id orders. The insertion-order permutations stay as a guard against a future unordered read; the comment says so
count = "parent" has no positive case (Test Quality #7 · Low)	Fixed	count_reads_every_direction_and_kind
CLI/MCP never fill now in a test; malformed now untested (Test Quality #8 · Low)	Fixed / filed	a_malformed_now_is_refused at the seam. The CLI/MCP branch can't be reached by a black-box test: the binaries link only the bundled plugins, and neither declares a derived key. Filed as 6j6v.4qga (P3)
Smaller gaps (Test Quality #9 · Low)	Fixed	of = 3 arm, OfField unknown part, "one read" asserted for mentioners_of, CLI paginate refusals assert their message, archived mentioner at Engine::mentioned_by
Ratio guard via read::* not Engine::* (Test Quality #10 · Info)	No action	Engine::next_query / list_query are thin try_with_state wrappers over read::next_page / list_page, and the §9 chain tests run at the Engine seam
RankingToml lacks deny_unknown_fields (Integrity #2 · Low)	Fixed	[ranking.lsit] is refused at load; test
Untyped related … of = "rank" in [ranking.list] default order (Integrity #3 · Low)	Fixed	Refused at load with and without per-type tables; the typed form loads; test
mentioned_by_bulk duplicates across chunks (Integrity #4 · Info)	Fixed	targets are deduplicated before chunking; 602-target test
Integrity #5–#7 (Info)	No action	Confirmations (bounded growth, determinism, clean error paths)

Quality gates

• cargo fmt --all
• cargo test -p nexus-flow-facade --test derived_ranking --test list_query --test next_query --test mentioned_by --test next_rank_per_type
• cargo test -p nexus-flow-facade --lib plugin:: and --lib read
• cargo test -p nexus-flow-core --lib store::
• cargo test -p nexus-flow-cli --test list --test mention
• cargo xtask docs check, cargo xtask changelog check origin/main
• A mutation check reverted the three fixes the new tests guard (present_parent, rank scope, cross-section fingerprint). Each reversion turned its test red.

Clippy and the full suite are left to CI, by instruction.

🤖 Generated with Claude Code

https://claude.ai/code/session_016q5uMUkZJQW4fckJoA3FME