diff --git a/CLAUDE.md b/CLAUDE.md
index f3be047..84aed0c 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -1,37 +1,115 @@
+# CLAUDE.md
+
+This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
+
 # pglifecycle
 
-A PostgreSQL schema management tool, being rewritten in Rust. See PLAN.md
-for the rewrite plan and phase gates.
+A PostgreSQL schema management tool (Rust rewrite of a former Python
+implementation). It keeps a database schema as a version-controlled
+project of YAML files — one file per object, validated against a
+JSON-Schema contract — and moves schema between the repository and live
+databases.
+
+## Commands
 
-## Development
+`just` (the Justfile) is the task runner; `just --list` shows all
+recipes. The common ones:
 
 ```bash
-cargo build          # Build the binary
-cargo test           # Run tests
-cargo fmt            # Format (max_width 79, see rustfmt.toml)
-cargo clippy --all-targets -- -D warnings
+just build          # cargo build
+just check          # fmt-check + lint + test (the CI trio)
+just fmt            # cargo fmt (max_width 79, see rustfmt.toml)
+just lint           # cargo clippy --all-targets -- -D warnings
+just test           # cargo test
+just run -- pull --help   # cargo run -- <args>
 ```
 
-The Python implementation this replaces is preserved at the
-`python-final` tag; reference it via:
+Run a single test directly with cargo:
+
+```bash
+cargo test --test pull              # one integration test file in tests/
+cargo test name_of_test             # a single test by name
+```
+
+## Architecture
+
+The CLI (`src/cli.rs`, clap) dispatches four subcommands from
+`src/main.rs`, each a top-level module:
+
+- **`pull`** (`src/pull/`) — database (or `pg_dump` archive) → project.
+  Runs `pg_dump`/`pg_dumpall` (`src/pgdump.rs`), parses each entry's DDL
+  through `src/ddl/`, merges child entries (indexes, constraints,
+  triggers, comments, ACLs) into their owning objects, and writes the
+  structured YAML project.
+- **`build`** (`src/build/`) — project → `pg_restore -Fc` archive.
+- **`create`** (`src/skeleton.rs`) — scaffold an empty project.
+- **`deploy`** (`src/deploy/`) — compare the project against a live
+  database/dump and emit the DDL to make the database match. **Output
+  first**: the script goes to stdout/`--output` to be applied
+  separately; destructive statements (DROPs, data-losing changes,
+  drop+recreate fallbacks) are excluded unless `--allow-drop`.
+
+Shared layers used by the commands:
+
+- `src/models/` — the object model. `Definition` is an untagged enum
+  over every supported object type; `Item` wraps one with its inventory
+  `id` and dependency set. Serde field order **is** the emitted YAML key
+  order — declaration order is load-bearing, and `deny_unknown_fields`
+  is intentional. A `Project` (`src/project/`) is `load`ed (`load.rs`)
+  and `validate`d against the schemata before use.
+- `src/ddl/` — tree-sitter-postgres CST → models. The grammar is
+  generated from PostgreSQL's `gram.y`, so node kinds mirror grammar
+  productions and carry no named fields; extraction walks the tree by
+  kind via `NodeExt` helpers.
+- `src/deploy/diff.rs` + `alter.rs` — compute the project↔database diff
+  and resolve each change to in-place ALTER or drop+recreate.
+- `src/pgdump.rs` — `pg_dump`/`pg_dumpall` subprocess wrappers.
+- `src/progress.rs` — indicatif bars; logging is bridged through them so
+  log records print above live bars and stdout stays clean (deploy
+  writes its script there).
+
+External crates do the heavy lifting: `libpgdump` (archive
+read/write), `libpgfmt` (SQL formatting; `--style` selects a house
+style, but `deploy` always compares with the `pg_dump` default),
+`tree-sitter-postgres` (DDL parsing).
+
+## Parity oracle
+
+`build` rendering is deliberately bug-for-bug faithful to the old Python
+implementation so archives compare entry-by-entry; intentional
+deviations (places the Python emitted broken SQL) are enumerated at the
+top of `src/build/mod.rs` and asserted in `tests/build_parity.rs`. The
+Python implementation is preserved at the `python-final` tag:
 
 ```bash
 git clone --depth 1 --branch python-final https://github.com/gmr/pglifecycle /tmp/pglifecycle-py
 ```
 
-It is the parity oracle until the round-trip gates in PLAN.md pass.
+Module headers reference the Python file each replaces ("ports X.py").
 
-## Testing
+## Testing & gates
+
+Integration tests live in `tests/`. The round-trip and deploy gates need
+PostgreSQL: they bring up `compose.yaml` (Postgres 17) on a
+dynamically-mapped host port and run the scripts in `bin/`.
+
+```bash
+just db-up          # start the container, wait until healthy
+just db-port        # print the mapped host port
+just round-trip     # schema → pull → build → restore → diff
+just deploy-gates   # deploy equivalence / convergence / safety
+just gates          # both
+```
 
-- Integration tests live in `tests/`
-- Round-trip tests require PostgreSQL (via Docker Compose)
-- `fixtures/` holds the test database schema
+The gate recipes discover the mapped port automatically. To run a gate
+script by hand, set `PGHOST`/`PGPORT`/`PGUSER` (the recipes use
+`PGUSER=postgres`).
 
-## Key Directories
+## Key directories
 
-- `src/` - Rust sources (module map in PLAN.md)
-- `schemata/` - JSON-Schema (YAML) definitions for PostgreSQL objects —
-  the on-disk project contract, carried over from Python unchanged
-- `test-project/` - Example project structure (parity contract)
-- `fixtures/` - Test database schema
-- `bin/` - Utility scripts (fixture data generation)
+- `schemata/` — JSON-Schema (YAML) definitions for PostgreSQL objects;
+  the on-disk project contract, carried over from Python unchanged.
+- `test-project/` — example project structure; a parity contract.
+- `fixtures/` — test database schema.
+- `bin/` — gate scripts and fixture-data generation.
+- `docs/` — mkdocs site (`just docs` builds with `--strict`).