@@ -4,6 +4,12 @@ All notable changes to AgentFlow are documented in this file.
 
 ## [Unreleased]
 
+### Docs — STATUS/Next after E4 + drain ≥100; P2-6 plan only
+
+- STATUS: E4 Checks 1–4 in Proven; Next is sustained ≥100 (with
+  `--pace-eps` protocol) and P2-6 packaging plan — not delivery topology.
+- Plan (no code): `docs/plans/p2-6-runtime-namespace-migration.md`.
+
 ### Perf — S10 stretch try: 107.3 eps bridge apply on 2000-event drain
 
 - Live Mac compose (1 Flink TM): 2000-event realpath drain cleared at
 
@@ -8,8 +8,8 @@
 AgentFlow's product axis — **event → live metric** on the real streaming path
 (Kafka → Flink → serving bridge → ClickHouse → API with Redis push
 invalidation) — is implemented, measured, and documented. Current work is
-raising the write-path throughput ceiling and turning documented limits into
-passed ones (endurance, scale, delivery topology).
+raising the write-path ceiling from **drain-window** numbers to **sustained**
+ones, and packaging hygiene for the next breaking release.
 
 ## Proven
 
@@ -19,6 +19,7 @@ passed ones (endurance, scale, delivery topology).
 | In-process demo freshness | 1.06 s p50 / 1.99 s p95 | [freshness-benchmark.md](freshness-benchmark.md) |
 | Real-path throughput measured | produce ~700 eps; bridge apply is the ceiling (see below) | [perf/throughput-realpath.md](perf/throughput-realpath.md) |
 | 2-pod control plane on kind | webhook registered on pod A visible on pod B; verify script PASS | [perf/e4-2pod-topology-2026-07-09.md](perf/e4-2pod-topology-2026-07-09.md) |
+| E4 Checks 1–4 (2 pods, delivery + alert single-page) | **PASS** on kind | [perf/e4-check4-alert-single-page-2026-07-17.md](perf/e4-check4-alert-single-page-2026-07-17.md) |
 | 4 h endurance soak (real path + API reads) | bounded lag (peak 2 915 → 0), bridge RSS/FD flat, one faulted batch replayed exactly-once by the journal guard, **zero cache drift** | [perf/soak-s11-2026-07-10.md](perf/soak-s11-2026-07-10.md) |
 | At-scale on own data (S13) | **51.2 M rows / 2.87 M orders / 4 years of legend history**, analyst queries 20–730 ms, all 17 §12 invariants pass incl. full-scan GTIN validation | [perf/scale-own-data-2026-07-11.md](perf/scale-own-data-2026-07-11.md) |
 | Security pass (offline/unit remainder) | closed; third-party pen-test **not** claimed | [security-s12-2026-07-09.md](security-s12-2026-07-09.md), [security-audit.md](security-audit.md) |
@@ -72,14 +73,13 @@ batched path (fold rules, idempotency, replay) are in
 
 ## Next
 
-1. **Delivery topology** — Checks 1–4 live on kind
-   ([perf/e4-replica-topology-2026-07-11.md](perf/e4-replica-topology-2026-07-11.md),
-   [perf/e4-check3-exactly-one-delivery-2026-07-16.md](perf/e4-check3-exactly-one-delivery-2026-07-16.md),
-   [perf/e4-check4-alert-single-page-2026-07-17.md](perf/e4-check4-alert-single-page-2026-07-17.md)).
-   Phase 3 automated topology proof is closed.
-2. **≥ 100 eps sustained** — multi-hour / paced produce at ≥ 100 eps with
-   bounded lag (numeric drain bar observed at 107.3 eps on a 2000-burst —
-   [perf/throughput-realpath-100eps-try-2026-07-17.md](perf/throughput-realpath-100eps-try-2026-07-17.md)).
+1. **≥ 100 eps sustained** — multi-hour / paced produce at ≥ 100 eps with
+   bounded lag. Drain-window ≥100 is already measured (107.3 eps @ 2000);
+   protocol and flags: [perf/throughput-realpath-100eps-try-2026-07-17.md](perf/throughput-realpath-100eps-try-2026-07-17.md)
+   (`--pace-eps 100 --count …`). Needs a healthy Flink hop for the whole window.
+2. **P2-6 packaging** (breaking) — rename wheel package `src` →
+   `agentflow_runtime` with deprecation shim; plan only until a release
+   window: [plans/p2-6-runtime-namespace-migration.md](plans/p2-6-runtime-namespace-migration.md).
 
 External gates remain unchanged and are listed in the README scope note:
 production CDC onboarding, a benchmark on production-grade hardware, and a
 
@@ -49,7 +49,7 @@ Mac compose stand, and the bridge is not trailing Flink on that window
   metrics `:9108`, `config/serving.yaml` (ClickHouse).
 - Co-tenant containers left stopped; no `docker system prune`.
 
-## Reproduce
+## Reproduce (drain window — this run)
 
 ```bash
 export PATH=$HOME/bin:/usr/local/bin:$PATH
@@ -65,6 +65,32 @@ docker compose -f docker-compose.yml -f docker-compose.flink.yml \
   --bridge-metrics http://127.0.0.1:9108/metrics --latency-samples 0 --no-md
 ```
 
+## How to measure *sustained* ≥100 (not yet run)
+
+The driver already supports paced produce (`--pace-eps`). Closing the sustained
+claim is a **longer produce window** with lag remaining bounded while the
+*target ingress* is ≥100 eps — not another unpaced backlog drain.
+
+```bash
+# Example: 15 min @ 100 eps target ≈ 90_000 events (wall ~900 s of produce)
+# then catch-up; raise --catchup-timeout-seconds if lag is high.
+# Pass criteria (proposed, not yet signed off as product bar):
+#   - produce pace ≈ 100 eps (driver)
+#   - bridge_apply_eps and flink_eps stay near the produce rate over the window
+#   - lag_peak bounded and lag_end → 0 after catch-up
+#   - apply_failures_delta == 0
+.venv/bin/python scripts/benchmark_throughput_realpath.py \
+  --bootstrap 127.0.0.1:19092 \
+  --count 90000 --pace-eps 100 \
+  --bridge-metrics http://127.0.0.1:9108/metrics \
+  --latency-samples 0 --catchup-timeout-seconds 600 --no-md
+```
+
+**Stand constraints (Mac 6 GiB):** keep `--scale flink-taskmanager=1`; stop
+co-tenants; do not claim multi-hour until a run of that length actually
+finishes with healthy Flink. A 15 min paced arm is a first step; multi-hour
+remains a separate soak-class exercise.
+
 ## Honest claim (what to put in SSOT)
 
 | Claim | Status | Evidence |
 
@@ -0,0 +1,101 @@
+# P2-6 — Runtime namespace: `src` → `agentflow_runtime`
+
+> Plan only. **Do not implement** until a breaking release window is
+> scheduled. Source of the finding: `audit_gpt_11_07_26.md` §P2-6.
+
+## Problem
+
+| Today | Risk |
+|-------|------|
+| Distribution name `agentflow-runtime` | Fine |
+| Wheel packages top-level **`src`** (`[tool.hatch.build.targets.wheel] packages = ["src"]`) | Namespace collision with any other project that also ships `src`; non-idiomatic public import (`from src.serving…`) |
+| Import surface used everywhere in-repo | `src.*` in app code, tests, Docker/helm command lines, Flink image layout, docs |
+
+## Target
+
+- Installable package name: **`agentflow_runtime`** (import path matches
+  distribution intent; aligns with wheel evidence paths in
+  `docs/runbooks/release-rollback.md`).
+- Deprecation window: keep a top-level **`src`** shim that re-exports or
+  warns, then remove in N+1.
+- No change to distribution name `agentflow-runtime` unless a separate
+  rename is decided.
+
+## Non-goals (this plan)
+
+- Renaming `sdk/agentflow` or `sdk-ts` packages.
+- Making `agentflow-integrations` publishable (separate P2-6 bullet —
+  either first-class CI/publish or explicit “internal”).
+- Full monorepo re-layout.
+
+## Migration phases
+
+### Phase 0 — inventory (no code move)
+
+1. List all runtime entrypoints that hardcode `src.`:
+   - process modules: `python -m src.processing.bridge_consumer`,
+     `src.serving.api.main:app` (uvicorn / helm / staging Dockerfile)
+   - tests / mypy `packages=["src"]` if any
+   - docs and scripts
+2. Confirm hatch config: today only `packages = ["src"]` under
+   `[tool.hatch.build.targets.wheel]`.
+3. Build a wheel and record `unzip -l dist/*.whl | head` so the before/after
+   tree is evidence-backed.
+
+### Phase 1 — physical package + dual import (breaking-prep)
+
+1. Move or package-dir map `src/` content so imports are
+   `agentflow_runtime.…` while **source tree stays reviewable** (prefer
+   `src/agentflow_runtime/` layout *or* hatch `packages` map — pick one in
+   implementation PR, not both ad hoc).
+2. Provide `src/__init__.py` shim:
+   - re-export / path-hook / `import agentflow_runtime as …` pattern that
+     keeps `import src.serving…` working for one minor line.
+   - optional `DeprecationWarning` on first import of `src` (env-gated so
+     tests stay quiet until the window starts).
+3. Update **first-party** entrypoints only when shim is green:
+   - helm chart `command` / uvicorn module path
+   - staging Dockerfile CMD
+   - `scripts/k8s_staging_up.sh` inline Dockerfile
+   - compose if any
+4. Gate: unit + contract + a wheel install smoke
+   (`python -c "import agentflow_runtime; import src"`).
+
+### Phase 2 — first-party cutover
+
+1. Change in-repo imports `src.` → `agentflow_runtime.` (codemod + review).
+2. Keep shim for external consumers / old docs.
+3. mypy/ruff package paths, pytest `pythonpath`, coverage sources.
+
+### Phase 3 — remove shim (breaking release)
+
+1. Drop top-level `src` from the wheel.
+2. Changelog + migration note in `docs/migration/`.
+3. Tag as minor/major per semver policy for the release line.
+
+## Compatibility matrix (related P2-6 tail)
+
+| Artifact | Versioning today | Note |
+|----------|------------------|------|
+| `agentflow-runtime` | `2.0.0` in pyproject | This plan |
+| Python SDK `sdk/` | separate | Pin matrix in contract tests later |
+| TS SDK `sdk-ts/` | separate | Same |
+| `integrations/` | `1.0.1`, not in main publish workflow | Declare internal **or** promote |
+
+Do not block the namespace rename on integrations publish.
+
+## Verification checklist (when implementing)
+
+- [ ] `python -m build` wheel contains `agentflow_runtime/…` and (phase 1–2) `src/…` shim only
+- [ ] `pip install dist/*.whl` → both import styles work in phase 1–2
+- [ ] helm/staging still boots API (`uvicorn …:app`)
+- [ ] bridge: `python -m agentflow_runtime.processing.bridge_consumer` (or shim module path)
+- [ ] CI unit + contract green; no accidental `packages = ["src"]` only
+
+## Decision needed before coding
+
+1. **Layout:** `src/agentflow_runtime/` (src-layout) vs rename tree root.
+2. **Shim lifetime:** one minor vs one major.
+3. **Warning default:** on vs off until phase 2 complete.
+
+Until those three are answered in a release ticket, leave code as-is.