diff --git a/README.md b/README.md
index 08d53c7..47d6436 100644
--- a/README.md
+++ b/README.md
@@ -94,6 +94,8 @@ claude mcp add ledger -- ledger mcp
 
 See [docs/mcp.md](docs/mcp.md) for the tool table, action-provenance
 contract, remote mode, and the official-registry listing.
+See the [Evaluator Guide](docs/EVALUATOR_GUIDE.md) for the three-tier architecture
+and the boundary between context, memory, and provenance.
 
 ```python
 from ledger_agent import Meter
diff --git a/docs/EVALUATOR_GUIDE.md b/docs/EVALUATOR_GUIDE.md
new file mode 100644
index 0000000..7958450
--- /dev/null
+++ b/docs/EVALUATOR_GUIDE.md
@@ -0,0 +1,75 @@
+# Perseus evaluator guide
+
+Perseus is a three-tier local-first platform. The tiers are composable, but they
+have different responsibilities and can be evaluated independently.
+
+## 1. Perseus Context Engine
+
+**Role:** workspace fact scanner and pre-session context renderer.
+
+It resolves operator-selected workspace sources and renders a bounded Markdown
+artifact, normally `AGENTS.md`, before an agent session starts. It describes the
+current workspace: project instructions, repository state, selected files, and
+other explicitly allowed facts. The output is an input artifact for an agent or
+MCP host; it is not an inference engine.
+
+Evaluate it by running `perseus quickstart` in a temporary workspace, then
+rendering `.perseus/context.md` to `AGENTS.md`, and inspecting the bounded output.
+
+## 2. Perseus Vault
+
+**Role:** durable local memory.
+
+Perseus Vault is an encrypted Rust memory engine backed by embedded SQLite and
+FTS5. It stores and recalls durable workspace facts, with optional local dense
+embeddings and hybrid retrieval. Its primary agent integration is MCP over stdio;
+it does not require a Perseus-hosted service for the local path.
+
+For an LLM host, use the lean advertisement profile to keep tool selection focused:
+
+```bash
+mkdir -p /tmp/verify_vault
+perseus-vault serve --profile lean --db /tmp/verify_vault/perseus-vault.db
+```
+
+The lean profile advertises the core memory operations (`remember`, `recall`,
+`forget`, `correct`, `context`, `perseus_vault_workspace_status`, and `health`).
+In lean mode, `perseus_vault_workspace_status` is scoped to the transport-stamped
+MCP `clientInfo.name` and does not disclose other profile/workspace bindings. The
+full registry remains available under the default/all profile. Evaluate the
+installed binary and the MCP `initialize`/`tools/list`/`tools/call` flow rather
+than relying on a registry count copied from another release.
+
+## 3. Perseus Ledger
+
+**Role:** tamper-evident event provenance.
+
+Perseus Ledger is a stdlib `http.server`-based threaded audit server and Python
+package. It records events in a hash-chained append-only history, supports
+verification and receipts, and exports OSCAL-compatible evidence. It is the
+provenance layer for actions and resource/accounting events; it is not the memory
+store and does not replace the Context Engine or Vault.
+
+Evaluate it with the repository's `uv run pytest` suite, then exercise a temporary
+SQLite database, append a record, verify the chain, and inspect an OSCAL export.
+
+## Boundary: what Perseus is not
+
+Perseus is **not** an LLM, model provider, inference API, or prompt-generation
+service. It does not select or host a model for the operator. Perseus is also
+**not a required cloud SaaS dependency**: the Context Engine, Vault's local stdio
+path, and Ledger's local deployment can run on operator-controlled machines.
+Optional remote transports, provider integrations, and hosted deployment choices
+are explicit configuration boundaries, not prerequisites for the local product.
+
+## Recommended evaluation order
+
+1. Run the Context Engine quickstart in `/tmp/verify_perseus`, render its context
+to `AGENTS.md`, and inspect the result.
+2. Start Vault with `--profile lean` and verify its advertised MCP tools and a
+   write/recall round trip using a temporary database.
+3. Run Ledger's tests and verify one append, chain check, and OSCAL export against
+   a temporary database.
+4. Keep results labeled by tier, repository revision, feature profile, and test
+   command; do not combine retrieval, context-rendering, and provenance results
+   into one product metric.
diff --git a/legacy/README.md b/legacy/README.md
new file mode 100644
index 0000000..da9038a
--- /dev/null
+++ b/legacy/README.md
@@ -0,0 +1,6 @@
+# Archived legacy assets
+
+`legacy/` contains historical artifacts retained for provenance only. They are
+not shipped as the current Perseus Ledger application or public product surface.
+The active implementation and package use the canonical `ledger_agent` namespace,
+the `ledger` CLI, and the **Perseus Ledger** product name.
diff --git a/pyproject.toml b/pyproject.toml
index eda2634..2742907 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -53,6 +53,20 @@ invarium = ["invarium>=0.3"]
 all = ["stripe>=7.0,<14", "reportlab>=4.0,<5", "boto3>=1.28", "sentry-sdk>=2.0,<3", "invarium>=0.3; python_version >= '3.10'", "jsonschema>=4.18,<5"]
 dev = ["stripe>=7.0,<14", "reportlab>=4.0,<5", "boto3>=1.28", "pytest>=7.0,<10", "sentry-sdk>=2.0,<3", "invarium>=0.3; python_version >= '3.10'", "jsonschema>=4.18,<5"]
 
+# ``uv run`` installs the default ``dev`` group, so the documented local test
+# gate is self-provisioning. Keep this separate from runtime dependencies and
+# mirror the optional ``dev`` extra above for pip consumers.
+[dependency-groups]
+dev = [
+    "stripe>=7.0,<14",
+    "reportlab>=4.0,<5",
+    "boto3>=1.28",
+    "pytest>=7.0,<10",
+    "sentry-sdk>=2.0,<3",
+    "invarium>=0.3; python_version >= '3.10'",
+    "jsonschema>=4.18,<5",
+]
+
 [project.urls]
 Homepage = "https://perseus.observer/ledger/"
 Repository = "https://github.com/Perseus-Computing-LLC/ledger"
diff --git a/tests/test_evaluator_guide.py b/tests/test_evaluator_guide.py
new file mode 100644
index 0000000..646f3ea
--- /dev/null
+++ b/tests/test_evaluator_guide.py
@@ -0,0 +1,14 @@
+"""Evaluator-guide facts stay aligned with the Ledger implementation."""
+from pathlib import Path
+
+
+ROOT = Path(__file__).resolve().parents[1]
+
+
+def test_evaluator_guide_describes_the_stdlib_ledger_server():
+    guide = (ROOT / "docs" / "EVALUATOR_GUIDE.md").read_text(encoding="utf-8")
+    app = (ROOT / "ledger_agent" / "server" / "app.py").read_text(encoding="utf-8")
+
+    assert "stdlib `http.server`" in guide
+    assert "FastAPI" not in guide
+    assert "http.server" in app
diff --git a/tests/test_uv_default_test_dependency.py b/tests/test_uv_default_test_dependency.py
new file mode 100644
index 0000000..a72df9c
--- /dev/null
+++ b/tests/test_uv_default_test_dependency.py
@@ -0,0 +1,18 @@
+"""The documented plain ``uv run pytest`` gate must be self-provisioning."""
+from __future__ import annotations
+
+import re
+from pathlib import Path
+
+
+ROOT = Path(__file__).resolve().parents[1]
+
+
+def test_plain_uv_run_provisions_pytest_from_the_default_dev_group():
+    text = (ROOT / "pyproject.toml").read_text(encoding="utf-8")
+    group = re.search(
+        r"(?ms)^\[dependency-groups\]\s+dev\s*=\s*\[(.*?)\]",
+        text,
+    )
+    assert group, "pyproject.toml must declare a default uv development group"
+    assert re.search(r"(?i)pytest(?:[<>=!~]|\s|\")", group.group(1))
diff --git a/uv.lock b/uv.lock
index 51c0bc4..8a7776a 100644
--- a/uv.lock
+++ b/uv.lock
@@ -275,6 +275,17 @@ stripe = [
     { name = "stripe" },
 ]
 
+[package.dev-dependencies]
+dev = [
+    { name = "boto3" },
+    { name = "invarium" },
+    { name = "jsonschema" },
+    { name = "pytest" },
+    { name = "reportlab" },
+    { name = "sentry-sdk" },
+    { name = "stripe" },
+]
+
 [package.metadata]
 requires-dist = [
     { name = "boto3", marker = "extra == 'all'", specifier = ">=1.28" },
@@ -300,6 +311,17 @@ requires-dist = [
 ]
 provides-extras = ["stripe", "sentry", "pdf", "fetchers", "invarium", "all", "dev"]
 
+[package.metadata.requires-dev]
+dev = [
+    { name = "boto3", specifier = ">=1.28" },
+    { name = "invarium", marker = "python_full_version >= '3.10'", specifier = ">=0.3" },
+    { name = "jsonschema", specifier = ">=4.18,<5" },
+    { name = "pytest", specifier = ">=7.0,<10" },
+    { name = "reportlab", specifier = ">=4.0,<5" },
+    { name = "sentry-sdk", specifier = ">=2.0,<3" },
+    { name = "stripe", specifier = ">=7.0,<14" },
+]
+
 [[package]]
 name = "pillow"
 version = "12.3.0"