plan(v0.7.0): split depth from breadth, and give the deferred timing triad a home

[avrabe]

Two committed artifacts disagreed about what v0.7.0 is

• docs/releases/v0.7.0-plan.md (2026-07-29) — depth on one claim: "new drivers, new capabilities and new targets all wait behind E1."
• rivet's v0.7.0 scope, typed in feat(rivet): type the v0.7.0 scope — CM drivers, three targets, ARINC component tenants #244 (2026-07-30) — a third target family, ARINC component tenants, driver componentization.

Both are in the repo and both cannot govern. Split, rather than let one silently overrule the other:

• v0.7.0 — depth. E1 execute the composite, E2 dissolve it, plus the timing triad E2 unblocks.
• v0.7.1 — breadth. RV32, partition tenants, driver componentization, plus E3/E4.

The machinery finding

v0.6.0 is tagged (2026-07-30), but six of its thirteen artifacts still pointed at it unverified — the T2/T3/T4 timing triad, deliberately deferred at that release and never re-pointed.

Two consequences, both bad: a shipped release queried as permanently unclosed, and six real work items were invisible to every later plan. The release-planning skill names this exactly — deferral is a scope decision (bump release:), surfaced and logged, not silence.

release	before	after
v0.6.0	7 verified, 5 proposed, 1 implemented	7 verified — CLOSED
v0.7.0	2 verified, 8 proposed	2 verified, 1 implemented, 9 proposed
v0.7.1	did not exist	6 proposed

Why the triad ships with the dissolve

E1  execute the composite ─────► one scheduler / one clock become OBSERVED
     │
E2  dissolve the composite ────► the first native object for the whole OS exists
     │                            ├──► T2  object-code verification has an object
     │                            └──► T4  per-function WCET has something to bound
     └────────────────────────────────► T3  schedulability consumes T4's numbers

T2 has nothing to verify and T4 has nothing to bound until E2 produces an object. Planning them apart would have scheduled the consumers ahead of the producer.

New artifact pair

The ladder's E2 rung had no rivet artifact at all — the scope did not represent the thing the triad depends on. Added REQ-OS-COMPOSITE-DISSOLVE-001 / VER-OS-COMPOSITE-DISSOLVE-001.

Its kill-criterion is the one that matters: a smaller object with no undefined symbols is the failure mode, not the win — it means the gust:hal seam was swallowed and the OS is no longer swappable.

Oracle

rivet validate: PASS, 341 warnings before and after — the new artifacts add none. (I did catch myself copying method: test from a neighbouring artifact; it is not an allowed value, so mine uses automated-test. Eight pre-existing artifacts still carry the invalid value — not fixed here, happy to do it as a separate change.)

What this PR does not do

No implementation. This is scope and plan only — E1 has still never been run, and v0.6.0's own notes remain the honest statement of that gap.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!