diff --git a/docs/README.md b/docs/README.md
index bf7f8ccc..06b92746 100644
--- a/docs/README.md
+++ b/docs/README.md
@@ -58,7 +58,8 @@ back here.
 - **[contributing/release-process.md](contributing/release-process.md)** — versioning, tag procedure, Maven Central publication.
 
 ### Roadmaps & migrations
-- **[roadmaps/v1.6-roadmap.md](roadmaps/v1.6-roadmap.md)** — current development roadmap.
+- **[roadmaps/post-2.0-engineering.md](roadmaps/post-2.0-engineering.md)** — forward-looking engineering roadmap: internal refactors, scale, and tooling deferred past the 2.0 line.
+- **[roadmaps/v1.6-roadmap.md](roadmaps/v1.6-roadmap.md)** — the v1.6 "expressive" roadmap (shipped).
 - **[roadmaps/migration-v1-8-to-v1-9.md](roadmaps/migration-v1-8-to-v1-9.md)** — upgrade guide for v1.8 → v1.9.
 - **[roadmaps/migration-v1-7-to-v1-8.md](roadmaps/migration-v1-7-to-v1-8.md)** — upgrade guide for v1.7 → v1.8.
 - **[roadmaps/migration-v1-6-to-v1-7.md](roadmaps/migration-v1-6-to-v1-7.md)** — upgrade guide for v1.6 → v1.7.
diff --git a/docs/roadmaps/post-2.0-engineering.md b/docs/roadmaps/post-2.0-engineering.md
new file mode 100644
index 00000000..1e7ed257
--- /dev/null
+++ b/docs/roadmaps/post-2.0-engineering.md
@@ -0,0 +1,87 @@
+# Post-2.0 engineering roadmap
+
+The 2.0 line focused on **packaging and internal hygiene**: splitting the
+monolithic jar into per-concern modules, removing the dead legacy
+Entity-Component-System code, retiring the deprecated API surface, and
+tightening CI. This document tracks the **engineering** work deliberately left
+for after 2.0 — internal refactors, scale work, and tooling. None of it changes
+the public authoring API (`GraphCompose.document(...)` → `DocumentSession` →
+`DocumentDsl`); it is about the health of the engine and the build.
+
+## Status legend
+
+| Status | Meaning |
+| --- | --- |
+| Planned | Shape agreed; not started. |
+| Investigating | Being scoped; the approach is not yet fixed. |
+| Deferred | Intentionally postponed; captured here so it is not lost. |
+
+## Engine internals
+
+### Decompose the layout hot files
+
+`LayoutCompiler` (~1690 LOC) and `TextFlowSupport` (~1900 LOC) are the two
+largest files in `com.demcha.compose.document.layout`, and each carries several
+distinct responsibilities. The plan is to split them along their natural seams —
+pagination, row distribution, stack / overlay placement, and decoration — into
+focused, individually-tested collaborators (the package-private `RowSlots`
+extraction is the pattern to follow), with layout output unchanged and covered
+by the existing snapshot suite. **Status: Planned.**
+
+### Retire the internal `Entity` model
+
+The engine still resolves layout on a legacy `Entity` / `EntityManager` object
+model — the live layout coordinate, geometry (`EntityBounds`), and guide helpers
+are built on it. The dead ECS *execution* layer around it has been removed; what
+remains is genuinely live infrastructure. Fully retiring `Entity` /
+`EntityManager` means rebuilding the coordinate / geometry / guide helpers on a
+non-`Entity` representation and removing the legacy `engine.debug` snapshot
+overloads in `graph-compose-testing` that still reference it. This is a real
+engine refactor, sequenced after the module line stabilises. **Status: Deferred.**
+
+### Fail loudly on non-converged layout
+
+Some layout passes iterate toward a fixed point. Today a pass that does not
+converge silently uses its last iteration. A `FAIL_ON_UNCONVERGED` policy flag
+would let a build opt into failing loudly instead, surfacing the rare
+non-converging document in tests rather than shipping a subtly-wrong render.
+**Status: Planned.**
+
+## Scale & memory
+
+### Streaming / bounded-memory rendering
+
+Rendering currently holds the full layout graph and all page fragments in
+memory. Very large documents (thousands of pages) would benefit from a streaming
+path that paginates and flushes finalised pages incrementally, bounding peak
+memory. This needs a backend seam that can emit pages as they are completed
+rather than at the end of the pass. **Status: Investigating.**
+
+## Tooling & guardrails
+
+### ArchUnit module-boundary guards
+
+The canonical / engine / render layering and the module split are guarded today
+by targeted tests plus path-based greps that can pass vacuously after a move.
+ArchUnit rules would enforce the boundaries structurally at test time (for
+example: no `com.demcha.compose.engine.*` import inside the canonical
+`document.*` public API, and no cross-module back-edges), replacing the brittle
+checks with ones that cannot silently rot. **Status: Planned.**
+
+### Cross-module coverage aggregation
+
+The canonical core packages (`document.layout` / `document.dsl` /
+`document.backend.fixed`) are exercised mostly by the cross-module `qa` suites,
+which depend on the engine at **test** scope — so a single-module coverage
+report undercounts, and JaCoCo's `report-aggregate` does not traverse test-scope
+dependencies. An accurate report needs a small, dedicated, non-published
+aggregation module that compile-depends on the tested modules. The first step is
+report-only; thresholds follow after a baseline read. **Status: Planned.**
+
+### Per-module binary-compatibility baselines
+
+`japicmp` runs report-only on the 2.0 line — the major intentionally breaks
+binary compatibility. Once the 2.0 GA artifacts are published, the gate should
+switch to per-module baselines pinned at the GA release and break-on-incompatible
+mode, so each published module's public surface is protected from that point on.
+**Status: Deferred (post-GA).**