docs(buck2): trajectory gate, devenv-eval budgets, livestore scope; strict-green main

context/buck2/.decisions/0030-external-cells-are-not-a-composition-mechanism.md

@@ -72,7 +72,7 @@ rules-distribution role.
 
 - No change to COMP-R08/R10, decision 0020, or mr's mount pipeline from this
   decision alone. What changes the mount pipeline's _scope_ is the separate
-  proposal [.proposed/artifact-composition.md](./.proposed/artifact-composition.md),
+  proposal the artifact-composition proposal (superseded 2026-09-14; see [05-composition/open-questions.md](../05-composition/open-questions.md)),
   which asks whether Buck2 should compose cells across repositories at all.
 - The root-owned capability cell is recorded as an open question in
   05-composition; it is the right ownership boundary in every option and

context/buck2/.decisions/0031-complexity-gate-and-authority-ledger.md

@@ -71,3 +71,15 @@ Buck target on its default branch.
   closes, nothing is deleted from the VRS.
 - Consumers with one slice add a row and a deletion entry in their PR and
   author no VRS node; dotfiles keeps its delta node.
+
+## Amendment 1 (2026-09-14)
+
+The per-close fold is unreachable for the hub until consumers migrate and the
+composition layer is deleted, so a gate that only fires at close would be silent
+for the whole period in which +28,363 lines accumulated. Johannes (q26,
+2026-09-14) added a trajectory obligation: the ledger stores a reconciliation
+snapshot (cumulative net at a revision); at every reconciliation the cumulative
+net must be lower than at the previous snapshot, otherwise admissions pause
+until a contraction slice lands. Reconciliations happen per phase or after ten
+rows change status, whichever is first. The check gains the snapshot list and
+the comparison; the per-close fold is unchanged.

context/buck2/.decisions/0032-devenv-verbs-over-buck-aggregate.md

@@ -58,3 +58,18 @@ five samples per workload before reconsidering the outer verb.
 - Speed remains unresolved. Failure elapsed times are not benchmark samples.
 
 Evidence: [2026-09-12 check-entry bakeoff](../.experiments/2026-09-12-check-entry-bakeoff.md).
+
+## Amendment 1 (2026-09-14)
+
+Evidence after acceptance: on a contended host the devenv _evaluation_ step —
+not any task — stalled 20–60 min for five workers while every producer invoked
+directly completed in seconds; dotfiles decision 0027 measured the same cost
+class as our own Nix configuration (shell entry 229 s → 17.8 s after four fixes;
+devenv's fixed overhead 254 ms). The verb decision stands (q25, 2026-09-14) with
+two budgets attached: warm shell entry ≤ 20 s and task-run overhead ≤ 2 s,
+measured per repository at n ≥ 3 on a host below load 16. Each repository
+carries a `devenv-eval` residual ledger row whose dissolution is meeting those
+budgets by removing the cost at its source (task/shell closure decoupling,
+filtered source coercion, a recursive-input budget gate). While that row is
+open, a change may be verified with the same producer the task wraps, invoked
+directly, and must say so in its PR body; this is not a second verb.

context/buck2/05-composition/open-questions.md

@@ -9,10 +9,25 @@ by attribution), paid today by effect-utils alone: no repository authors an
 `effect_utils//` label and Phase 6 has not started. The only Buck2-native
 alternative, git external cells, is rejected
 ([decision 0030](../.decisions/0030-external-cells-are-not-a-composition-mechanism.md)).
-[.proposed/artifact-composition.md](../.decisions/.proposed/artifact-composition.md)
-proposes artifact-granular reuse for the effect-utils library edges and pauses
-composed-by-default; it lists the VRS edits and falsification spikes. Blocked
-on: Johannes' decision on criterion 6, and spikes 1–2 of the proposal.
+The first artifact-composition proposal (merged by mistake as a PR-local
+record via #1271; removed 2026-09-14) ran its falsification spikes in PR #1282:
+URL tarball closure, digest refusal, and a root-owned capability cell passed;
+the consumer boundary failed (publisher rejected scoped names; a single URL
+artifact could not coexist with workspace consumers under the consumer's pnpm
+policy). The architecture bakeoff (PR #1287,
+`.experiments/2026-09-13-composition-bakeoff.md`, proposed decision
+`.decisions/.proposed/composition-architecture.md` on that branch) provisionally
+selects artifact-default composition. Johannes (q22/q23, 2026-09-13) adopted
+that direction with **no registry**: the durable origin is the existing
+content-addressed buck2-products release layout, the CAS is an accelerator, and
+consumers pin tarball URL + integrity in their lockfile. The one-edge proof
+(PR #1289: `@overeng/utils` published, dotfiles notion-scan consuming by URL,
+typecheck + 28 tests green) succeeded; its net (+482) is judged by the
+reconciliation trajectory (decision 0031 Amendment 1). Composed-by-default
+(decision 0027) stays paused until the proposal is promoted or rejected.
+Remaining blocker: the constitutional edits (vision criterion 6, BUCK-R05/R06,
+COMP-R01/R02) are Johannes', and pnpm's injected-workspace pruning must be
+shown to be a strict no-op on the second install.
 
 ## Open 2026-09-12: root-owned capability cell
 

context/buck2/open-questions.md

@@ -61,4 +61,8 @@ Subsystem questions live in their subsystem (`03-materialization`,
 - Resolution signal: a written contributor loop for livestore under each
   answer, with the tool set and cold-start time measured on a machine outside
   the fleet.
-- Blocker: not started; livestore-lead owns it once the ledger exists.
+- Resolved (q17, 2026-09-14): livestore is an artifact consumer only. It pins
+  `@overeng/*` release assets by tarball URL + integrity (decision 0034 once the
+  artifact-composition proposal is accepted); no Buck2 or `mr` in livestore; its
+  contributors install nothing new; livestore rows are excluded from the ledger.
+  livestore PR #1622 closes.

context/buck2/requirements.md

@@ -139,6 +139,9 @@ invariants named in its own document:
   (which later deletion pays for it); the per-change signal is advisory, the
   per-close gate is not
   ([decision 0031](./.decisions/0031-complexity-gate-and-authority-ledger.md)).
+  Between closes the gate has a trajectory: at every reconciliation the
+  cumulative sum must be lower than at the previous reconciliation, or
+  admissions pause until it is (decision 0031, Amendment 1).
   When BUCK-R15 conflicts with coverage (BUCK-R01) or the wall-clock budgets
   (BUCK-R07), BUCK-R15 wins: the others are constraints with tolerances.
 - **BUCK-R16 Benchmark evidence:** Efficiency claims are measured, never