chore: close out architecture experiments

.github/workflows/ci.yml

@@ -95,27 +95,20 @@ jobs:
     steps:
       - name: Checkout
         uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
-        with:
-          # Full history so the back-edge-ceiling ratchet can read the baseline
-          # at the merge-base and reject a same-PR raise. Mirrors the Fallow job.
-          fetch-depth: 0
 
       - name: Setup toolchain
         uses: ./.github/actions/setup-node-pnpm
         with:
           install-deps: false
 
-      - name: Check import-direction DAG and ratchets
+      - name: Check import-direction DAG
         # Generalizes the former inline commands/-import grep into a structured
         # import-direction lint over the resolved graph. See scripts/layering/check.ts
         # and CONTEXT.md (Architecture: folder DAG + layering lint). Invoked directly
         # (no deps needed) so the job stays fast and does not require a pnpm install.
-        # LAYERING_BASE pins the baseline-monotonicity comparison to the PR base.
-        env:
-          LAYERING_BASE: ${{ github.event_name == 'pull_request' && github.event.pull_request.base.sha || github.event.before }}
         run: |
           node --experimental-strip-types --test scripts/layering/model.test.ts
-          node --experimental-strip-types scripts/layering/check.ts --base "$LAYERING_BASE"
+          node --experimental-strip-types scripts/layering/check.ts
 
   affected-selector:
     name: Affected-check Selector

.github/workflows/perf-nightly.yml

@@ -112,18 +112,12 @@ jobs:
           profile: pixel_7
           target: google_apis_playstore
           emulator-options: -no-window -gpu swiftshader_indirect -no-snapshot -noaudio -no-boot-anim -no-metrics
-          script: |
-            set -e
-            # Disable animations up front so accessibility dumps don't time out (the harness
-            # also runs `settings animations off`, this is belt-and-suspenders).
-            adb -s emulator-5554 shell settings put global window_animation_scale 0 || true
-            adb -s emulator-5554 shell settings put global transition_animation_scale 0 || true
-            adb -s emulator-5554 shell settings put global animator_duration_scale 0 || true
-            node --experimental-strip-types scripts/perf/run.ts \
-              --platform android \
-              --serial emulator-5554 \
-              --n "$PERF_ROUNDS" --warmup 1 \
-              --out-dir "$GITHUB_WORKSPACE/perf-results"
+          script: >-
+            set -e;
+            adb -s emulator-5554 shell settings put global window_animation_scale 0 || true;
+            adb -s emulator-5554 shell settings put global transition_animation_scale 0 || true;
+            adb -s emulator-5554 shell settings put global animator_duration_scale 0 || true;
+            node --experimental-strip-types scripts/perf/run.ts --platform android --serial emulator-5554 --n "$PERF_ROUNDS" --warmup 1 --out-dir "$GITHUB_WORKSPACE/perf-results"
 
       - name: Upload Android perf report
         if: always()

AGENTS.md

@@ -141,7 +141,7 @@ This repo encodes invariants as self-declaring gates. The correct response to a
 - every command declares a timeout policy on its descriptor: the timeout-policy completeness test
 - TS/Swift rule parity: golden tables under `contracts/fixtures/` consumed by vitest and the gated XCTest
 - cross-command apple-leak guard, folder DAG/import lint (including zero value-import cycles and
-  down-only exact target-spine back-edge identities), fallow (dead code, duplication, complexity)
+  zero target-spine back-edges), fallow (dead code, duplication, complexity)
 
 ## Hard Rules
 - Use process helpers from `src/utils/exec.ts` for TypeScript process execution: `runCmd`, `runCmdStreaming`, `runCmdSync`, `runCmdBackground`, and `runCmdDetached`. Do not import raw `spawn`/`spawnSync` outside `src/utils/exec.ts`; add or extend an exec helper instead. Plain `.mjs` packaging fixtures that cannot import TypeScript helpers should keep child-process usage local and prefer `execFile`/`execFileSync` over spawn.

CONTEXT.md

@@ -96,8 +96,7 @@ The perfect-shape refactor is complete and merged. Its end-state:
 - Folder DAG + layering lint. `kernel`/`remote`/`metro`/`client`/`snapshot`/`screenshot-diff`/
   `replay`/`cli-parser`/`daemon-client`+`server`/`sdk` are arranged as an import-direction DAG
   (imports point down toward the kernel sink), enforced in CI by `scripts/layering/check.ts`.
-  The gate rejects production value-import cycles and enforces the ranked target-spine's exact
-  back-edge identities as a down-only baseline in `scripts/layering/back-edge-baseline.json`.
+  The gate rejects production value-import cycles and every ranked target-spine back-edge.
 - Agent-cost. Responses carry a cost block and MCP `outputSchema`, rendered through a leveled
   `ResponseView`.
 

docs/agents/contract-projection-output-economy-spike.md

@@ -2,8 +2,9 @@
 
 Issues: #1185, #1186
 
-This report keeps the two experiments bounded: one typed command family for executable contract
-projection and one opt-in response family for digest expansion.
+This report records two completed, bounded experiments: one typed command family for executable
+contract projection and one opt-in response family for digest expansion. Neither result starts a
+repository-wide migration.
 
 ## Selection evidence
 
@@ -27,10 +28,10 @@ The baseline has three independent command-to-surface declarations per selected
 2. the public client method signature;
 3. the keyed MCP output-schema entry.
 
-That is 15 independent projection declarations around five executable definitions. The spike will
-make each executable definition the single command-to-client/MCP projection declaration and derive
-the client surface and MCP schema map from those five definitions. Success is five declarations
-instead of 15, without changing CLI, daemon, client, or MCP behavior.
+That was 15 independent projection declarations around five executable definitions. The experiment
+made each executable definition the single command-to-client/MCP projection declaration and
+derived the client surface and MCP schema map from those five definitions. It reached five
+declarations instead of 15 without changing CLI, daemon, client, or MCP behavior.
 
 The experiment stops at these five commands. It must not continue if it needs generated source, a
 schema DSL, a new dependency, a parallel registry, a new import back-edge, weaker parity checks,
@@ -63,7 +64,7 @@ seven commands, zero fallback observations, exactly one retry, and all in-sessio
 preserved. The network addition must keep those measurements unchanged and separately prove that a
 failed request remains identifiable and actionable from the digest.
 
-## Measurement plan
+## Measurement method
 
 - Declaration count: count independent selected-command projection declarations before and after.
 - Runtime: compare median cold imports of the MCP command-tool projection at the merge base and
@@ -92,8 +93,11 @@ from those projections, `AgentDeviceCommandClient` derives the five navigation m
 - Layering/Fallow: green; no new value back-edge (type-only import from client, value import only
   from the already-command-importing MCP schema map).
 
-Recommendation: #1185 should CONTINUE incrementally, one already-typed closed-contract family at a
-time, re-measuring declarations and bundle each step and stopping on any stop condition.
+The selected five-command experiment is complete. Other already-typed closed-contract families are
+candidates only, not a migration queue. Any follow-up must be proposed and measured independently,
+must reduce declarations without weakening ownership or parity checks, and must stop on any of the
+conditions above. Generated source, a schema DSL, a new dependency, a parallel registry, a new
+back-edge, or a material runtime/bundle regression remains a stop signal.
 
 ## Digest expansion results (#1186)
 
@@ -124,3 +128,71 @@ Defaults are unchanged; the digest is strictly opt-in via a non-default response
 risk: the dropped-field set is intentionally specific to the known verbose network payload fields;
 if a backend introduces a new large per-entry field it would remain until added to the drop list
 (safe by default — unknown fields are preserved, never silently narrowed).
+
+The selected network experiment is complete. Events, debug symbols, recording, and any new family
+remain candidates only. A follow-up requires paired evidence:
+
+1. deterministic byte/shape reduction on representative fixtures; and
+2. unchanged or better actionability, including follow-up command choice, fallback observation
+   count, retry count, identity, refs, warnings, artifacts, and in-session recovery.
+
+Byte reduction alone is not a successful digest.
+
+## Post-merge closeout evidence
+
+The deterministic suite was rerun after both implementations merged:
+
+- `pnpm test:output-economy`: 4 files and 26 tests passed.
+- Network default/digest: 26,371 / 1,665 bytes; all eight entries and failed-request identity
+  remain available.
+- MCP snapshot result: 1,173 bytes with four refs in the combined text + structured result.
+- Routine workflow: 2,276 bytes across seven commands, zero fallback observations, one retry, and
+  `recoveryPreservesSession=true` with code, session, refs generation, retriable flag, and hint.
+
+`pnpm bench:help-conformance --dry-run` generated eight prompts for the four output/recovery cases
+across the configured Haiku and GPT mini runners. This proves case selection and prompt plumbing
+only. No scored small-model result is recorded here because authenticated `claude` and `codex`
+runners were unavailable; future scored runs remain informational actionability oracles, not CI
+gates.
+
+No new live-device result is recorded in this closeout. The local environment had no simulator,
+emulator, or attached device. A one-round `perf-nightly` dispatch on the PR branch was also
+unavailable because the GitHub integration could not create workflow-dispatch events. Historical
+Android nightly logs did expose a deterministic workflow defect: the emulator action executes each
+`script` line independently, so shell continuation backslashes reached `scripts/perf/run.ts` as a
+literal `\` argument. The action now receives one folded shell script, so `set -e` governs the
+benchmark and no continuation tokens cross action boundaries. The next scheduled run after merge
+is the live validation point. Deterministic fixtures are not presented as live reliability
+evidence.
+
+The six-month churn × current-size refresh continues to identify bounded candidates rather than a
+split mandate:
+
+| File | Commits | LOC | Churn × LOC |
+|---|---:|---:|---:|
+| `src/cli/parser/cli-help.ts` | 34 | 1,227 | 41,718 |
+| `src/core/dispatch-interactions.ts` | 31 | 1,123 | 34,813 |
+| `src/client/client-types.ts` | 21 | 1,139 | 23,919 |
+
+Higher-churn files below 1,000 lines include `src/core/dispatch.ts`, `src/cli.ts`, and several
+daemon handlers. These are seam-stability signals, not automatic split targets. A future split must
+answer one ownership question, improve locality, preserve dependency direction, and mirror source
+and test topology. The refresh deliberately avoids a blanket LOC gate; path-based commit counts
+also undercount renamed files unless inspected with `git log --follow`.
+
+Repeat the broad ranking with:
+
+```sh
+SINCE="$(node -e "const d = new Date(); d.setUTCMonth(d.getUTCMonth() - 6); process.stdout.write(d.toISOString().slice(0, 10))")"
+git log --since="$SINCE" --name-only --pretty=format: -- 'src/**' \
+  | rg -v '^$|__tests__|\.test\.' | sort | uniq -c | sort -rn \
+  | while read -r commits file; do
+      test -f "$file" || continue
+      lines="$(wc -l < "$file")"
+      printf '%d\t%d\t%d\t%s\n' "$((commits * lines))" "$commits" "$lines" "$file"
+    done \
+  | sort -rn
+```
+
+`src/utils` contains 106 TypeScript files: 71 production files and 35 tests. That is a standing
+opportunistic ownership policy, not a mass relocation backlog.

package.json

@@ -118,7 +118,6 @@
     "check:affected": "node --experimental-strip-types scripts/check-affected/run.ts",
     "check:affected:test": "node --experimental-strip-types --test scripts/check-affected/model.test.ts scripts/check-affected/run.test.ts",
     "check:layering": "node --experimental-strip-types --test scripts/layering/model.test.ts && node --experimental-strip-types scripts/layering/check.ts",
-    "check:layering:baseline": "node --experimental-strip-types scripts/layering/check.ts --update-baseline",
     "check:production-exports": "fallow dead-code --config fallow-production-exports.json --production --unused-exports --baseline fallow-baselines/production-unused-exports.json --fail-on-issues",
     "check:production-exports:baseline": "fallow dead-code --config fallow-production-exports.json --production --unused-exports --save-baseline fallow-baselines/production-unused-exports.json --summary",
     "check:bundle-owner-files": "node --experimental-strip-types scripts/check-bundle-owner-files.ts",