docs: record native live append gate

.phrase/docs/CHANGE.md

@@ -5,6 +5,9 @@
 - 2026-08-20 plan and implement post-audit hardening (`task037`–`task042`,
   `task030` source-fresh evaluator continuation; `issue022`–`issue028`):
   `.phrase/phases/phase-kp-overhaul-20260726/change_post_audit_hardening_20260820.md`
+- 2026-08-20 select and implement native automatic live-append backend
+  (`task030`; `ekp-auto-justify-native-append`):
+  `.phrase/phases/phase-kp-overhaul-20260726/adr_native_live_append_backend_20260820.md`
 - 2026-08-13 close issues awaiting user-visible confirmation (`issue011`–
   `issue017`, `issue019`, `issue020`; `issue018`/`task030` remain open):
   `.phrase/phases/phase-kp-overhaul-20260726/change_close_user_confirmation_20260813.md`

.phrase/phases/phase-kp-overhaul-20260726/change_post_audit_hardening_20260820.md

@@ -78,16 +78,35 @@
   passes the same matrix without argument errors.
 - **Commit:** `01001b9`.
 
+## 2026-08-20 — Implement user-selected native live-append backend
+
+- **Modify** `ekp-buffer.el` and `ekp.el` to let prepared automatic live
+  append use the loaded C 1D DP when `ekp-auto-justify-native-append` is
+  non-nil, without changing ordinary full-layout dispatch or the 15-field C
+  ABI. Native-unavailable and option-disabled paths remain Elisp.
+- **Add** `adr_native_live_append_backend_20260820.md`, bilingual README
+  guidance, evaluator backend metadata, and a public buffer regression proving
+  the setting toggles the bridge.
+- **Verification:** native bridge/parity regression 1/1 and source-first ERT
+  296/296 pass. In the bounded all-width, 2/4/8/16-row, GC-excluded
+  source-fresh round, the candidate width-80 C p95/p99 was
+  14.190/16.495 ms and the Elisp-configured live path (native append backend)
+  was 13.484/14.334 ms. Layout parity, zero-work, GC, conflict, and width
+  non-regression checks passed; the C p99 remains just above the locked
+  16 ms target, so `issue018/task030` stays open.
+- **Commit:** `3d3dda6`.
+
 ## Verification
 
-- Source-first normal ERT: 295/295.
+- Source-first normal ERT: 296/296.
-- Source-first seeded random ERT: 295/295, including all nine GUI verifier
+- Source-first seeded random ERT: 296/296, including all nine GUI verifier
   tests previously omitted by the name filter.
 - Per-test isolated ERT: 294/294 process runs logged `0 unexpected`; the
   later batch-position test is covered by focused isolated C ERT.
 - C focused ERT: 15/15; C portable build warning-clean; source fuzz:
   300/300 with zero failures; checkdoc, release, dictionary, and shell gates
   pass.
-- Corrected narrowed source evaluator: parity, zero-work, GC, conflicts, and
+- Corrected source evaluator: parity, zero-work, GC, conflicts, and width
-  width non-regression pass, but source candidate p99 remains above 16 ms;
+  non-regression pass. The bounded all-width/row source round still reports
-  the performance issue is intentionally not closed.
+  C p99 16.495 ms at width 80, so the performance issue is intentionally not
+  closed.

.phrase/phases/phase-kp-overhaul-20260726/issue_narrow_live_append_latency_20260729.md

@@ -2,13 +2,16 @@
 
 ## issue018 [ ] Unique live appends miss the frame budget at very narrow widths
 
-- **Status:** Exact incremental implementation verified; the locked
+- **Status:** Native live-append backend implemented and verified; the locked
-  source-instrumented gate remains open.
+  source-fresh gate remains open for preparation/assembly latency.
 - **Summary:** Stable transactions removed per-key whole-hard-line planning,
-  and task030 now incrementally extends paragraph preparation, Elisp DP
+  and task030 incrementally extends paragraph preparation, DP state, and
-  state, and layout lines only at structural row crossings. Byte-compiled
+  layout lines only at structural row crossings. The user-selected
-  production paths are within the 16 ms frame budget; the deliberately
+  `ekp-auto-justify-native-append` backend now uses the loaded C 1D DP for
-  source-loaded, fully instrumented evaluator still exceeds it.
+  prepared live appends even when full layout is configured for Elisp.
+  Byte-compiled production paths are within the 16 ms frame budget; the
+  source-fresh evaluator still exceeds it in Elisp-owned preparation and plan
+  assembly.
 - **Environment:** Emacs 30.2 on macOS, C backend 1.6, benchmark width
   fixed at 80 px, `gc-cons-threshold` bound to
   `most-positive-fixnum` so mutator work is measured without GC pauses.
@@ -70,14 +73,25 @@
   pure-Elisp DP from a safe reachable state, reuses common layout lines, and
   reconstructs only the live dirty source island. Unsupported contexts take
   the unchanged full path. C int32 validation now performs one extraction.
+- **Native backend update (2026-08-20):** `ekp-auto-justify-native-append`
+  now routes prepared 1D live append DP through the loaded C module even when
+  full layout uses Elisp. The option can be disabled for a pure-Elisp live
+  path, and unavailable modules fall back exactly. In a bounded source-fresh
+  all-width, 2/4/8/16-row, GC-excluded round, width-80 candidate append-DP
+  p95/p99 was 2.051/3.127 ms for C and 2.495/3.406 ms for the Elisp-configured
+  live path. Total p95/p99 was 14.190/16.495 ms and 13.484/14.334 ms,
+  respectively; preparation and assembly remain the active budget owners and
+  the C p99 is still just above 16 ms.
 - **Verification:** Exact baseline/C/Elisp hashes and append-chain
   equivalence pass across 64/80/96/128/160 px, 2/4/8/16-row fixtures,
-  default/excluded GC, unsafe fallbacks, and randomized chains. Normal and
+  default/excluded GC, unsafe fallbacks, and randomized chains. Current
-  random-order ERT pass 199/199; 300 fuzz cases, warning-clean production/C
+  source-first normal and random-order ERT pass 296/296; the native bridge
-  compilation, 9/9 C tests, release checks, and reviewed temporal GUI
+  parity regression passes 1/1; 300 fuzz cases, warning-clean production/C
-  evidence pass.
+  compilation, C contract tests, release/dictionary checks, and reviewed
+  temporal GUI evidence pass. The bounded source-fresh performance round
+  remains red only on the C p99 absolute target.
 - **User Confirmation:** Pending; the formal source-instrumented performance
   target also remains open.
 - **Resolved At:** Unresolved.
 - **Resolved By:** Pending.
-- **Commit:** Pending.
+- **Commit:** `3d3dda6` (native backend); performance gate remains open.

.phrase/phases/phase-kp-overhaul-20260726/task_post_audit_hardening_20260820.md

@@ -63,3 +63,12 @@
     is 9.753/5.663 ms for both engines, while Elisp append-DP is 55.970 ms
     and C append-DP is 1.015 ms. The evaluator wrapper collision was fixed and
     the same source matrix passes the harness without argument errors.
+  - User-selected architecture option 2 is implemented as
+    `ekp-auto-justify-native-append`: loaded C DP is used only for prepared
+    auto live append, while full/string `ekp-use-c-module=nil` remains pure
+    Elisp and unavailable native modules fall back exactly. The source
+    append-DP p95 drops from roughly 56 ms to roughly 2.1 ms in the bounded
+    all-width/row run; preparation and assembly remain the active budget
+    owners. Width-80 C candidate p95/p99 is 14.190/16.495 ms, so the locked
+    16 ms gate remains open rather than being marked complete on a noisy
+    single-round result.