docs: add refactoring implementation plan

.gitignore

@@ -17,3 +17,4 @@ apps/ui/test-results
 apps/ui/vite.config.js
 apps/ui/vite.config.d.ts
 *.log
+__*.md

TASKS.md

@@ -2,21 +2,27 @@
 
 ## Current
 
-### `feat/runtime-env-simplification`
+### `feat/refactoring-plan`
 
 Status: in_progress
 
 DoD:
-- runtime uses component-based PostgreSQL config as the primary model
+- `__REFACTORING.md` documents the full remediation plan for registry modularization, sqlx compile-time validation, KDF correction, typed timestamps, ID display support, pool tuning, and error normalization
-- `CRANK_BASE_URL` replaces split public URL variables
+- `.gitignore` ignores local `__*.md` working notes and refactoring specs
-- deployment uses one GitHub secret per runtime env variable
+- `TASKS.md` points to the refactoring track as the active work item
-- deprecated env variables are removed from compose, examples, and docs
 
 ## Next
 
-- `feat/websocket-test-run-polish`
+- `feat/postgres-registry-modularization`
 
 ## Backlog
 
+- `feat/postgres-registry-modularization`
+- `feat/sqlx-compile-time-verification`
+- `feat/typed-timestamps`
+- `feat/secret-crypto-hkdf`
+- `feat/id-display-support`
+- `feat/postgres-pool-config`
+- `feat/error-structure-normalization`
 - `feat/websocket-test-run-polish`
 - `feat/live-authenticated-staging-entry`

__REFACTORING.md

@@ -0,0 +1,932 @@
+# Refactoring Plan
+
+## Purpose
+
+This document defines the refactoring track required to improve maintainability, safety, and production-readiness in the current codebase. It is intentionally implementation-oriented: each area is split into concrete slices, target modules, function ownership, migration order, and acceptance criteria.
+
+The refactoring scope covers:
+
+1. splitting `crates/crank-registry/src/postgres.rs` into domain modules;
+2. migrating static SQL to `sqlx::query!` / `sqlx::query_as!` where possible;
+3. replacing SHA-256 key derivation in `SecretCrypto` with HKDF;
+4. replacing string timestamps in domain models with typed timestamps;
+5. adding `Display` support to `define_id!`;
+6. introducing explicit PostgreSQL pool configuration;
+7. removing duplicated SQL timestamp formatting;
+8. making runtime and API error payloads more structured.
+
+## Guiding Rules
+
+- Keep domain logic separate from SQL storage concerns.
+- Preserve current behavior while shrinking file size and coupling.
+- Move in vertical slices: storage module split first, then SQL typing, then timestamp migration.
+- Every slice must be independently reviewable and mergeable.
+- Prefer compatibility shims during migration instead of breaking multiple crates at once.
+- Add or update tests in the same slice as the refactoring.
+
+## Refactoring Tracks
+
+### Track A: Registry Modularization
+
+#### Problem
+
+`crates/crank-registry/src/postgres.rs` currently concentrates almost all database logic:
+
+- bootstrap and auth;
+- workspaces and memberships;
+- invitations;
+- operations and versions;
+- agents and bindings;
+- samples and descriptors;
+- API keys;
+- secrets;
+- stream sessions and async jobs;
+- invocation logs and usage;
+- YAML import jobs.
+
+This creates:
+
+- a permanent git conflict hotspot;
+- slow code navigation;
+- oversized review diffs;
+- accidental coupling between unrelated domains.
+
+#### Target Layout
+
+`crates/crank-registry/src/`
+
+```text
+error.rs
+lib.rs
+migrations.rs
+model.rs
+postgres/
+  mod.rs
+  connect.rs
+  auth.rs
+  workspace.rs
+  invitation.rs
+  secret.rs
+  stream.rs
+  api_key.rs
+  operation.rs
+  agent.rs
+  observability.rs
+  yaml_import.rs
+  shared.rs
+```
+
+#### Ownership by Module
+
+##### `postgres/mod.rs`
+
+Responsibilities:
+
+- define `PostgresRegistry`;
+- wire submodules;
+- expose the public impl surface;
+- keep only high-level constructor entry points.
+
+Functions staying here:
+
+- `new`
+- `connect`
+- `connect_with_options`
+- `migrate`
+
+Functions moving out:
+
+- everything domain-specific.
+
+##### `postgres/connect.rs`
+
+Responsibilities:
+
+- connection setup;
+- pool initialization;
+- migration invocation;
+- pool option translation from config to `PgPoolOptions`.
+
+Functions:
+
+- `connect`
+- `connect_with_options`
+- `migrate`
+- internal helpers:
+  - `build_pool_options`
+  - `build_connect_options`
+
+##### `postgres/auth.rs`
+
+Responsibilities:
+
+- auth users;
+- user sessions;
+- membership access checks;
+- auth profiles.
+
+Functions:
+
+- `upsert_bootstrap_user`
+- `get_auth_user_by_email`
+- `get_auth_user_by_id`
+- `update_user_profile`
+- `update_user_password`
+- `create_user_session`
+- `get_user_session`
+- `touch_user_session`
+- `revoke_user_session`
+- `set_user_session_current_workspace`
+- `user_has_workspace_access`
+- `save_auth_profile`
+- `get_auth_profile`
+- `list_auth_profiles`
+- `list_auth_profiles_referencing_secret`
+
+Internal helpers to extract:
+
+- `map_auth_user_row`
+- `map_session_row`
+- `map_auth_profile_row`
+
+##### `postgres/workspace.rs`
+
+Responsibilities:
+
+- workspaces;
+- memberships;
+- workspace CRUD.
+
+Functions:
+
+- `list_workspaces`
+- `list_workspaces_for_user`
+- `ensure_membership`
+- `list_memberships`
+- `update_membership_role`
+- `delete_membership`
+- `get_workspace`
+- `create_workspace`
+- `update_workspace`
+- `delete_workspace`
+
+Internal helpers:
+
+- `workspace_exists`
+- `map_workspace_row`
+- `map_membership_row`
+
+##### `postgres/invitation.rs`
+
+Responsibilities:
+
+- invitations lifecycle.
+
+Functions:
+
+- `list_invitations`
+- `create_invitation`
+- `delete_invitation`
+
+##### `postgres/secret.rs`
+
+Responsibilities:
+
+- secrets;
+- secret versions;
+- secret touch/rotation.
+
+Functions:
+
+- `list_secrets`
+- `get_secret`
+- `get_current_secret_version`
+- `create_secret`
+- `rotate_secret`
+- `delete_secret`
+- `touch_secret`
+
+##### `postgres/stream.rs`
+
+Responsibilities:
+
+- stream sessions;
+- async jobs;
+- state transitions;
+- cleanup.
+
+Functions:
+
+- `create_stream_session`
+- `get_stream_session`
+- `update_stream_session_state`
+- `close_stream_session`
+- `list_stream_sessions`
+- `delete_expired_stream_sessions`
+- `create_async_job`
+- `get_async_job`
+- `update_async_job_status`
+- `cancel_async_job`
+- `list_async_jobs`
+- `delete_expired_async_jobs`
+
+Internal helpers:
+
+- `assert_stream_session_transition`
+- `assert_async_job_transition`
+- `map_stream_session_row`
+- `map_async_job_row`
+
+##### `postgres/api_key.rs`
+
+Responsibilities:
+
+- platform API key lifecycle.
+
+Functions:
+
+- `list_platform_api_keys`
+- `get_platform_api_key_by_secret_for_workspace_slug`
+- `create_platform_api_key`
+- `revoke_platform_api_key`
+- `delete_platform_api_key`
+- `touch_platform_api_key`
+
+##### `postgres/operation.rs`
+
+Responsibilities:
+
+- operations;
+- versions;
+- published operations;
+- samples;
+- descriptors.
+
+Functions:
+
+- `create_operation`
+- `create_version`
+- `update_operation_draft`
+- `archive_operation`
+- `delete_operation`
+- `has_published_agent_bindings_for_operation`
+- `list_operations`
+- `list_operation_usage_summaries`
+- `list_operation_agent_refs`
+- `get_operation_summary`
+- `get_operation_version`
+- `list_operation_versions`
+- `publish_operation`
+- `get_published_operation`
+- `list_published_operations`
+- `save_sample_metadata`
+- `list_sample_metadata`
+- `save_descriptor_metadata`
+- `list_descriptor_metadata`
+- `insert_version_row`
+
+Internal helpers:
+
+- `map_operation_summary_row`
+- `map_operation_version_row`
+- `map_published_operation_row`
+- `serialize_optional_json`
+
+##### `postgres/agent.rs`
+
+Responsibilities:
+
+- agents;
+- agent versions;
+- bindings;
+- publish lifecycle.
+
+Functions:
+
+- `list_agents`
+- `get_agent_summary`
+- `create_agent`
+- `get_agent_version`
+- `save_agent_bindings`
+- `update_agent_summary`
+- `delete_agent`
+- `publish_agent`
+- `unpublish_agent`
+- `archive_agent`
+- `get_published_agent_tools_by_slug`
+- `list_agent_bindings`
+- `insert_agent_version_row`
+- `replace_agent_bindings_rows`
+
+##### `postgres/observability.rs`
+
+Responsibilities:
+
+- invocation logs;
+- usage rollups and breakdowns.
+
+Functions:
+
+- `create_invocation_log`
+- `list_invocation_logs`
+- `get_invocation_log`
+- `summarize_usage`
+- `list_usage_timeline`
+- `list_usage_by_operation`
+- `get_usage_for_operation`
+- `list_usage_by_agent`
+- `get_usage_for_agent`
+
+##### `postgres/yaml_import.rs`
+
+Responsibilities:
+
+- YAML import jobs.
+
+Functions:
+
+- `create_yaml_import_job`
+- `finish_yaml_import_job`
+- `get_yaml_import_job`
+
+##### `postgres/shared.rs`
+
+Responsibilities:
+
+- SQL snippets reused by several modules;
+- common timestamp select fragments during the migration period;
+- small row-mapping helpers;
+- pagination helpers.
+
+Planned content:
+
+- `Page<T>` helpers if they remain storage-local;
+- constants for table-independent fragments;
+- transitional helpers until typed timestamps remove string formatting entirely.
+
+#### Slice Plan for Track A
+
+##### Slice A1: File layout and constructors
+
+- create `postgres/` directory;
+- move `connect`, `connect_with_options`, `migrate`, and struct wiring;
+- keep all public signatures unchanged.
+
+##### Slice A2: Workspace/Auth/API-key split
+
+- move `auth.rs`, `workspace.rs`, `invitation.rs`, `api_key.rs`;
+- leave behavior unchanged;
+- add targeted module tests if not present.
+
+##### Slice A3: Operation/Agent split
+
+- move `operation.rs`, `agent.rs`, `yaml_import.rs`;
+- keep SQL unchanged for this slice;
+- ensure published-operation and published-agent behavior is unchanged.
+
+##### Slice A4: Secret/Stream/Observability split
+
+- move `secret.rs`, `stream.rs`, `observability.rs`;
+- keep transition logic and cleanup behavior unchanged.
+
+##### Slice A5: Remove legacy monolith
+
+- remove remaining methods from root `postgres.rs`;
+- rename module structure so `postgres/mod.rs` is the only entry point.
+
+#### Done Criteria for Track A
+
+- no `postgres.rs` monolith remains;
+- no single registry source file exceeds roughly 800–1200 lines;
+- all existing registry tests pass unchanged;
+- `lib.rs` still re-exports `PostgresRegistry` without downstream breakage.
+
+## Track B: Compile-Time SQL Verification
+
+### Problem
+
+Most static SQL currently uses `sqlx::query(...)` / `query_as::<_, T>(...)` with runtime validation only. That leaves:
+
+- column name typos;
+- mismatched types;
+- bad nullability assumptions;
+- schema drift
+
+to runtime instead of compile time.
+
+### Goal
+
+Use `sqlx::query!` and `sqlx::query_as!` for every static query with stable shape.
+
+Keep runtime builders only where the query shape is truly dynamic.
+
+### Query Classification
+
+#### Must migrate to `query!` / `query_as!`
+
+Use macros for:
+
+- single-row fetches by ID;
+- list queries with static columns;
+- insert/update/delete with static statement shape;
+- existence checks;
+- count queries;
+- cleanup queries with fixed predicates.
+
+Expected modules:
+
+- `auth.rs`
+- `workspace.rs`
+- `invitation.rs`
+- `secret.rs`
+- `stream.rs`
+- `api_key.rs`
+- most of `agent.rs`
+- most of `yaml_import.rs`
+
+#### May stay runtime-based
+
+Keep dynamic APIs only when necessary:
+
+- optional filters that materially change SQL shape;
+- optional sort columns chosen from a whitelist;
+- bulk dynamic `IN` lists where macro usage becomes unreadable;
+- `QueryBuilder`-based multi-row inserts/updates.
+
+Expected modules:
+
+- `operation.rs` list queries with optional filters;
+- `observability.rs` usage aggregations and breakdowns;
+- some search/filter endpoints.
+
+### Required Infrastructure
+
+#### Add SQLx offline support
+
+Repository-level requirements:
+
+- use `sqlx prepare` to generate checked metadata;
+- keep checked metadata in version control if the team wants deterministic CI;
+- wire verification into CI or `just verify`.
+
+Concrete additions:
+
+- `sqlx-data.json` or the current SQLx metadata equivalent for the chosen version;
+- `DATABASE_URL` or test DB wiring for `cargo sqlx prepare`.
+
+#### Introduce row structs where needed
+
+For `query_as!`, add storage-local row structs for cases where domain structs should not be coupled directly to SQL column shape.
+
+Examples:
+
+- `WorkspaceRow`
+- `AuthUserRow`
+- `OperationSummaryRow`
+- `InvocationLogRow`
+- `UsageTimelineRow`
+
+Rule:
+
+- domain structs remain domain-oriented;
+- row structs remain SQL-oriented.
+
+### Function-Level Migration Rules
+
+#### For `query_as!`
+
+Use `query_as!` when:
+
+- the query result maps cleanly to a typed row struct;
+- all returned columns are explicit;
+- nullability is known.
+
+#### For `query!`
+
+Use `query!` when:
+
+- no reusable row struct is needed;
+- the query is a simple `insert`, `update`, `delete`, or `exists` fetch.
+
+#### For dynamic list queries
+
+Use:
+
+- `sqlx::QueryBuilder<Postgres>` with explicit allowed sort/filter sets;
+- storage-local mapper functions from row to domain object.
+
+### Migration Order
+
+##### Slice B1: `auth`, `workspace`, `invitation`
+
+Low-risk, fixed-shape queries.
+
+##### Slice B2: `secret`, `stream`, `api_key`, `yaml_import`
+
+Still mostly fixed-shape.
+
+##### Slice B3: `operation`, `agent`
+
+Introduce mixed strategy: macros for fixed queries, builder for dynamic lists.
+
+##### Slice B4: `observability`
+
+Likely the most dynamic area. Convert only static pieces first, leave complex aggregations to `QueryBuilder`.
+
+### Done Criteria for Track B
+
+- all fixed-shape SQL uses compile-time checked macros;
+- dynamic SQL is explicitly documented and isolated;
+- CI fails on SQL drift before runtime tests;
+- no new static query is accepted through `sqlx::query(...)` without justification.
+
+## Track C: SecretCrypto KDF Correction
+
+### Problem
+
+Current code in `crates/crank-runtime/src/secret_crypto.rs` derives the AES key by hashing the master key with SHA-256:
+
+- no KDF semantics;
+- no labeled derivation context;
+- no future rotation/versioning path.
+
+### Goal
+
+Replace direct hashing with HKDF-SHA256.
+
+### Target Design
+
+`SecretCrypto::new(master_key: &str)` will:
+
+1. validate non-empty master key;
+2. derive a 32-byte AES key using `Hkdf<Sha256>`;
+3. use explicit `info` context, for example:
+   - `b"crank.secret-envelope.v1"`
+4. initialize `Aes256Gcm` from the derived bytes.
+
+### Module Changes
+
+File:
+
+- `crates/crank-runtime/src/secret_crypto.rs`
+
+Changes:
+
+- replace `Sha256::digest(trimmed.as_bytes())` with HKDF expansion;
+- preserve the public constructor signature;
+- keep ciphertext format unchanged if possible;
+- if ciphertext format must change later, introduce envelope `version` before doing so.
+
+### Required Tests
+
+- `encrypt_decrypt_roundtrip_still_works`
+- `empty_master_key_is_rejected`
+- `same_master_key_derives_stable_key`
+- `different_master_keys_produce_different_ciphertexts`
+
+### Done Criteria for Track C
+
+- no direct SHA-256 KDF remains;
+- `SecretCrypto` uses HKDF-SHA256 with explicit context;
+- all existing secret encrypt/decrypt flows remain compatible.
+
+## Track D: Typed Timestamps
+
+### Problem
+
+Domain and storage structs use `String` timestamps. SQL converts timestamps via repeated `to_char(...)`.
+
+That causes:
+
+- repeated formatting logic in SQL;
+- weak type guarantees;
+- inability to do time arithmetic without reparsing;
+- broad surface for formatting drift.
+
+### Goal
+
+Use `time::OffsetDateTime` in domain and storage models, with RFC 3339 serialization at the transport boundary.
+
+### Target Type Strategy
+
+#### Domain layer
+
+Use:
+
+```rust
+#[serde(with = "time::serde::rfc3339")]
+pub created_at: OffsetDateTime
+```
+
+and the same for `updated_at`, `expires_at`, `used_at`, and similar fields.
+
+#### Storage row layer
+
+Also use `OffsetDateTime`.
+
+Do not format timestamps in SQL except for transitional compatibility shims.
+
+#### Transport layer
+
+JSON still emits RFC 3339 strings through Serde.
+
+### Files Affected
+
+#### Domain crates
+
+- `crates/crank-core/src/workspace.rs`
+- `crates/crank-core/src/access.rs`
+- `crates/crank-core/src/auth.rs`
+- `crates/crank-core/src/agent.rs`
+- `crates/crank-core/src/operation.rs`
+- `crates/crank-core/src/observability.rs`
+- `crates/crank-core/src/secret.rs`
+- `crates/crank-core/src/stream_session.rs`
+
+#### Registry models
+
+- `crates/crank-registry/src/model.rs`
+
+#### Admin API transport/service DTOs
+
+- `apps/admin-api/src/service.rs`
+- any route DTO files that mirror time fields directly
+
+### Migration Strategy
+
+##### Slice D1: storage row types
+
+- change storage row structs to `OffsetDateTime`;
+- remove `to_char(...)` from fixed-shape SQL;
+- keep service DTOs as strings temporarily if needed.
+
+##### Slice D2: domain models
+
+- convert `crank-core` models to typed timestamps with Serde helpers;
+- update constructors and tests.
+
+##### Slice D3: API/service transport
+
+- remove string timestamp shims from service DTOs where possible;
+- ensure JSON remains stable.
+
+##### Slice D4: cleanup
+
+- remove transitional formatting helpers;
+- remove `now_string()` helpers in favor of typed timestamps.
+
+### Acceptance Criteria
+
+- no `created_at: String` or `updated_at: String` remains in domain/storage models;
+- no `to_char(created_at...)` / `to_char(updated_at...)` remains in registry SQL except perhaps in legacy compatibility code slated for removal in the same slice;
+- JSON contract stays RFC 3339.
+
+## Track E: `Display` for ID Newtypes
+
+### Problem
+
+`define_id!` currently requires `.as_str()` everywhere for formatting and messages.
+
+### Goal
+
+Add `std::fmt::Display` support to every generated ID newtype.
+
+### File
+
+- `crates/crank-core/src/ids.rs`
+
+### Change
+
+Add inside `define_id!`:
+
+```rust
+impl std::fmt::Display for $name {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        write!(f, "{}", self.0)
+    }
+}
+```
+
+### Follow-Up Cleanup
+
+After macro support lands:
+
+- replace formatting sites like `workspace_id.as_str()` inside messages;
+- keep `.as_str()` only where an actual `&str` API requires it.
+
+### Acceptance Criteria
+
+- all ID newtypes implement `Display`;
+- formatting code no longer uses `.as_str()` inside `format!`, `println!`, or error messages unless `&str` is actually required.
+
+## Track F: PostgreSQL Pool Configuration
+
+### Problem
+
+The pool is currently created via default `PgPoolOptions`, which means the runtime behavior is implicit and under-configured.
+
+### Goal
+
+Make pool behavior explicit and configurable from environment.
+
+### Target Config Model
+
+Add a typed config struct, for example:
+
+```rust
+pub struct PostgresPoolConfig {
+    pub max_connections: u32,
+    pub min_connections: u32,
+    pub acquire_timeout_ms: u64,
+    pub idle_timeout_ms: u64,
+    pub max_lifetime_ms: u64,
+}
+```
+
+### Runtime Environment
+
+Add env variables:
+
+- `POSTGRES_MAX_CONNECTIONS`
+- `POSTGRES_MIN_CONNECTIONS`
+- `POSTGRES_ACQUIRE_TIMEOUT_MS`
+- `POSTGRES_IDLE_TIMEOUT_MS`
+- `POSTGRES_MAX_LIFETIME_MS`
+
+### Files
+
+- `crates/crank-registry/src/postgres/connect.rs`
+- `apps/admin-api/src/main.rs`
+- `apps/mcp-server/src/main.rs`
+- config docs and examples
+
+### Behavior
+
+`connect_with_options` should accept either:
+
+- direct `PgConnectOptions` plus explicit pool config;
+- or an internal default config built from env.
+
+Defaults should be conservative and explicit, for example:
+
+- `max_connections = 20`
+- `min_connections = 2`
+- `acquire_timeout = 5s`
+- `idle_timeout = 10m`
+- `max_lifetime = 30m`
+
+Exact values can be tuned later, but they must no longer be hidden defaults.
+
+### Acceptance Criteria
+
+- pool limits and timeouts are explicit;
+- docs reflect the new variables;
+- startup logs or config inspection make active pool settings visible.
+
+## Track G: Timestamp Formatting Deduplication
+
+### Problem
+
+The same `to_char(... at time zone 'UTC', 'YYYY-MM-DD"T"HH24:MI:SS"Z"')` fragment is copied across many queries.
+
+### Goal
+
+Eliminate duplicated formatting logic.
+
+### Strategy
+
+This track is mostly resolved by Track D, but there is a transition window.
+
+During transition:
+
+- centralize any remaining formatting in `postgres/shared.rs`;
+- define named SQL fragments only once;
+- remove inline duplicated snippets.
+
+After Track D:
+
+- remove formatting fragments entirely.
+
+### Acceptance Criteria
+
+- duplicated inline timestamp formatting is no longer scattered across query strings;
+- final state has typed timestamps instead of SQL formatting.
+
+## Track H: Error Structure Normalization
+
+### Problem
+
+Some runtime and API error variants carry only a `details: String`, which makes programmatic handling and consistent observability harder.
+
+### Goal
+
+Introduce structured error payloads where the current model is too string-heavy.
+
+### Scope
+
+Priority types:
+
+- `RuntimeError`
+- registry storage errors where context is stringly typed;
+- API error mapping around secret/runtime/streaming faults.
+
+### Design Rules
+
+- keep `thiserror`;
+- use structured fields in variants whenever stable context exists;
+- reserve free-form `details` only for opaque third-party messages.
+
+Examples:
+
+Instead of:
+
+```rust
+RuntimeError::SecretCrypto { details: String }
+```
+
+prefer:
+
+```rust
+RuntimeError::SecretCrypto {
+    operation: &'static str,
+    details: String,
+}
+```
+
+or dedicated variants such as:
+
+- `InvalidSecretEnvelope`
+- `SecretDecryptFailed`
+- `SecretSerializeFailed`
+
+### Slice Plan
+
+##### Slice H1
+
+- normalize `SecretCrypto` and streaming/session/job errors.
+
+##### Slice H2
+
+- normalize adapter errors for REST/gRPC/WebSocket/SOAP where stable categories exist.
+
+##### Slice H3
+
+- normalize API mapping and error codes.
+
+### Acceptance Criteria
+
+- core runtime error variants expose structured context;
+- API can map key failure classes deterministically;
+- logs and telemetry can group errors by category instead of by ad hoc message text.
+
+## Delivery Order
+
+Recommended order:
+
+1. `feat/refactoring-plan`
+2. `feat/id-display-support`
+3. `feat/postgres-pool-config`
+4. `feat/postgres-registry-modularization`
+5. `feat/sqlx-compile-time-verification`
+6. `feat/secret-crypto-hkdf`
+7. `feat/typed-timestamps`
+8. `feat/error-structure-normalization`
+9. `feat/post-refactor-cleanup`
+
+Reasoning:
+
+- `Display` is tiny and immediately reduces noise.
+- pool config is operational and isolated.
+- module split should happen before large query migrations, otherwise the macro migration lands into the monolith and becomes harder to review.
+- HKDF is isolated and low-risk.
+- typed timestamps should land after storage modules are split and query migration is underway.
+
+## Per-Slice Verification
+
+Every slice must run:
+
+- `just fmt-check`
+- `just check`
+- `just clippy`
+- targeted tests for the touched area
+
+Additionally:
+
+- SQL macro migration slices must run SQLx prepare/verification;
+- timestamp slices must run serialization regression tests;
+- secret crypto slice must run compatibility and roundtrip tests.
+
+## Success Definition
+
+The refactoring track is complete when:
+
+- registry storage code is split into domain modules;
+- static SQL is compile-time validated;
+- `SecretCrypto` uses HKDF instead of SHA-256 as a KDF;
+- domain/storage timestamps are typed;
+- ID newtypes implement `Display`;
+- PostgreSQL pool settings are explicit and configurable;
+- timestamp formatting duplication is removed;
+- error variants carry structured context where it matters.
+
+At that point, maintainability improves materially without changing the product model or feature scope.