diff --git a/README.md b/README.md
index 62b60f53..bba6a1c5 100644
--- a/README.md
+++ b/README.md
@@ -10,7 +10,10 @@ This repo merges three previously-separate projects into a single workspace with
 
 ## Status
 
-**Early implementation.** The workspace builds; the front-end crates are landing per the migration plan below. Present today: `crates/core` (shared planner core) and `crates/lower` (SQL/PromQL → L2 lowering). The two docs below are the design + migration plan that the code is following.
+**Early implementation.** The workspace builds and is organized into a
+layer-named crate stack (the query→sketch pipeline reads top-to-bottom). The
+two docs below are the broader design + migration plan the code is following;
+the deployment-model / runtime / bin crates they describe are still planned.
 
 - [`docs/design.md`](docs/design.md) — target architecture, repo layout, extension points, wire contracts.
 - [`docs/migration-plan.md`](docs/migration-plan.md) — phase-by-phase plan to land the merger without disrupting any running deployment.
@@ -19,38 +22,52 @@ This repo merges three previously-separate projects into a single workspace with
 
 **5-layer query→sketch pipeline**:
 
-| Layer | What | Where it lives |
+| Layer | What | Where it lives (today) |
 |---|---|---|
-| 1 | Query-language parsing (PromQL, SQL, DataFusion, ElasticDSL) | `core::query_language` |
-| 2 | Per-language logical plan (relational algebra tree) | `core::logical_plan` |
-| 3 | Intent algebra — language-, deployment-, AND data-model-independent IR (`QueryExpr` + `AggIntent`, intent only — no sketch type, no params). Supports both time-series and tabular data via a `Source` sum inside `Scan`. | `core::intent_algebra` |
-| 4 | Cost-aware optimizer — rule engine + shared rule library; deployment models pick rules. Produces the **sketch algebra** IR `SketchExpr` (sketch-bound: kind + params committed). | **framework in `core::optimizer`**; sketch-bound IR in `core::sketch_algebra`; rule choices in each deployment model |
-| 5 | Physical plan — stage allocation + emit to wire format | **framework in `core::physical`**; topology + emitter in each deployment model |
+| 1 | Query-language parsing (PromQL, SQL; DataFusion, ElasticDSL planned) | `asap-frontend-promql`, `asap-frontend-sql` |
+| 2 | Per-language relational algebra tree | `asap-ir::intent_algebra::relational` (emitted by the front ends) |
+| 3 | Intent algebra — language-, deployment-, AND data-model-independent IR (`QueryExpr` + `AggIntent`, intent only — no sketch type, no params). Supports both time-series and tabular data via a `Source` sum inside `Scan`. | `asap-ir::intent_algebra` |
+| 4 | Cost-aware optimizer — CSE + cost model + sketch-vs-exact boundary + canonicalization. Produces the **sketch algebra** IR (sketch-bound: kind + params committed). | optimizer passes in `asap-plan`; sketch-bound IR in `asap-sketch` |
+| 5 | Physical plan — stage allocation + emit to wire format | *planned* (per-deployment-model) |
 
-**Core owns the shared infrastructure across all 5 layers**, not just L1-3. Deployment models are thin: they pick which rules fire (L4), declare their deployment topology (L5), and provide an emitter for their output format. Typical deployment model crate: **500-2500 LOC**.
+### Crate DAG
+
+The workspace splits by pipeline role; the dependency arrows only ever point
+**up** — `asap-ir` never depends on a front end, an optimizer, or a runtime.
+
+```
+asap-ir            L2 relational + L3 canonical IR + converter + binder +
+                   resolution + schema + expr + names + workload types
+   ▲    ▲    ▲
+   │    │    └── asap-sketch          L4 sketch algebra
+   │    │
+   │    └─────── asap-plan            optimizer layer:
+   │                                  CSE + cost model + sketch-vs-exact
+   │                                  boundary + canonicalize
+   │
+   ├─ asap-frontend-promql   PromQL L1→L2   (dep: promql-parser only)
+   ├─ asap-frontend-sql      SQL   L1→L2   (dep: datafusion only)
+   │        └── asap-lower   thin facade re-exporting lower_promql / lower_sql
+   ▲
+asap-e2e           integration tests
+```
+
+Splitting the front ends quarantines their parsers: a caller that needs only
+PromQL depends on `asap-frontend-promql` and never compiles DataFusion, and
+vice-versa. `asap-lower` is the convenience facade for callers that want both.
 
 ```
 crates/
-  core/                 # all 5 layers of shared infrastructure; no I/O
-    query_language/     # L1 — parsers
-    logical_plan/       # L2 — per-language algebra trees
-    intent_algebra/     # L3 IR — QueryExpr + AggIntent (intent only, ~25 variants superset)
-    sketch_algebra/     # L4 IR — SketchExpr (sketch-bound: kind + params committed)
-    lower/              # L1→L2→L3 passes
-    optimizer/          # L4 framework — rule engine + shared rule library + cost traits; produces SketchExpr
-    physical/           # L5 framework — PhysicalPlanner + stage allocator + topology + sketch catalogue
-    pipeline/           # L1→…→L5 driver, parameterized on deployment model
-  runtime/              # HTTP / OpAMP / replanner / store — service skeleton
-  deployment-model-asaplifecycle    # DC-specific rules + 3-stage topology + OTel/backend emitters
-  deployment-model-asapquery        # precompute-engine rules + 1-stage + StreamingConfig/InferenceConfig YAML emitters
-  deployment-model-asapfusion       # DataFusion rewrites + 0-stage (in-process) + LogicalPlan emit
-  control-proto         # OpAMP proto + internal proto (tonic/prost)
-  testing               # shared fixtures
-bin/
-  asap-controller       # long-running service, all deployment models
-  asap-query             # one-shot CLI (deployment-model-asapquery only)
-  asap-lifecycle        # OPTIONAL: standalone service with only deployment-model-asaplifecycle
-  asap-fusion-bench     # OPTIONAL: benchmark harness over deployment-model-asapfusion
+  ir/                   # L2 + L3 IR, converter, binder, schema — no query-language deps
+  sketch/               # L4 sketch-bound IR (asap-sketch)
+  plan/                 # optimizer layer (asap-plan): CSE now; cost model / boundary / canonicalize are stubs
+  frontend-promql/      # PromQL L1→L2 (promql-parser)
+  frontend-sql/         # SQL L1→L2 (datafusion)
+  lower/                # asap-lower — facade re-exporting both front ends
+  e2e/                  # cross-language integration tests
+
+# planned (per docs/design.md): L5 physical framework, runtime service,
+# deployment-model-* crates, control-proto, and the bin/ entrypoints.
 ```
 
 A new deployment model lands by adding one crate with `rules.rs` (pick L4 rules) + `topology.rs` + an emitter, plus one line in `bin/asap-controller/main.rs`. No changes to core or runtime.
@@ -61,7 +78,7 @@ Three ways downstream can use ASAPController — mix as needed:
 
 | Mode | Use case | How |
 |---|---|---|
-| **Rust library** | In-process use of the IR or a specific deployment model (e.g. asap-fusion benchmarks) | `Cargo.toml`: `asap-control-core = { git = "...", tag = "v0.1.0" }` or any individual `deployment-model-*` crate. Per-crate dep isolation keeps dep trees small (deployment-model-asapfusion pulls DataFusion; deployment-model-asapquery pulls PromQL/SQL parsers; neither pulls axum/OpAMP). |
+| **Rust library** | In-process use of the IR or a specific deployment model (e.g. asap-fusion benchmarks) | `Cargo.toml`: `asap-ir = { git = "...", tag = "v0.1.0" }` or any individual `deployment-model-*` crate. Per-crate dep isolation keeps dep trees small (deployment-model-asapfusion pulls DataFusion; deployment-model-asapquery pulls PromQL/SQL parsers; neither pulls axum/OpAMP). |
 | **HTTP service sidecar** | Production control plane — e.g. ASAPQuery-backend POSTing QuerySpec on capability-miss | Run `asap-controller` binary, POST to `/api/v1/plan`. Same contract DC controller speaks today. |
 | **CLI / Docker image** | One-shot init container (e.g. docker-compose init job that writes `streaming_config.yaml`) | `asap-controller plan --workload ... --output-dir ...` or the dedicated `asap-query` binary |
 
@@ -74,7 +91,7 @@ ASAPController is the **control plane only**. The data plane — OTel collectors
 Because core now owns the L4/L5 infrastructure (not just L1-3), deployment model crates are small and largely self-contained. A deployment model can live either:
 
 - **Inside ASAPController workspace** (lockstep release with core, one-PR cross-deployment-model changes)
-- **In its own downstream repo** (independent release cadence, depends on published `asap-control-core`)
+- **In its own downstream repo** (independent release cadence, depends on published `asap-ir`)
 
 Both produce functionally identical artifacts. The placement is a deployment/team-ownership decision, not an architectural fork. Default:
 - `deployment-model-asaplifecycle` + `deployment-model-asapquery` in ASAPController workspace (share a YAML emitter).
@@ -82,7 +99,7 @@ Both produce functionally identical artifacts. The placement is a deployment/tea
 
 ## Building
 
-The build is a normal `cargo build`. The only setup is GitHub access: `crates/lower`
+The build is a normal `cargo build`. The only setup is GitHub access: `crates/frontend-promql`
 depends on our private PromQL parser fork (`ProjectASAP/promql-parser`, branch `asap`)
 as a git dependency, so Cargo has to be able to clone a private repo.