You are a constitutional council ranking individual git commits for ownership allocation.

Compare these two commits. Decide which contributed more lasting value to the project.

Judge substance, not spectacle:
- Prefer correct, lasting design and real bugfixes over churn, formatting, renames, or generated noise.
- Prefer clarity and necessity over sheer line count. A small precise change can beat a large diffuse one.
- Do not favor a side merely because its patch is longer or noisier.
- Weight what the change does for the project, not the contributor's name.

Return ONLY a JSON object: {"winner": "A" or "B", "ratio": "N:M", "explanation": "..."}
The explanation must cite concrete differences in the patches (1-3 sentences).

Side A — contributor: tommy-mor
Side A — commit message:
[1c914c6e] stage set

Side A — unified diff (full patch):
diff --git a/plan.md b/plan.md
new file mode 100644
index 0000000000000000000000000000000000000000..00d6867a1e0ed144a16a020ea037f685ce646c73
--- /dev/null
+++ b/plan.md
@@ -0,0 +1,155 @@
+# Plan: `ItemId` + `RouteContext` (identity vs hrefs)
+
+This document is for **the next agent** to continue the refactor without re-deriving context from chat. It supersedes ad-hoc notes: treat it as the checklist of record until the work lands and this file is deleted or trimmed.
+
+## Goal
+
+- **Identity** (what lives in the reducer graph, votes, indexes) becomes a **structural `ItemId` enum** in `slug-types`, not a canonical `String` / `CanonicalItemUrl` newtype.
+- **Presentation** (tilde / dash display, breadcrumbs) derives from `ItemId` via explicit methods, not string stripping.
+- **Routing** (browser `href`s for public vs room) goes through **`RouteContext`** (started in `server/src/html/routing.rs`) so Maud/handlers do not stitch `/r/…` vs `/~` ad hoc.
+
+**Non-goals for v1 of the migration:** backward-compatible JSONL or dual-read of old canonical strings in the event log (project has accepted breaking changes). If you reintroduce compat, document it here.
+
+## Current state (as of this plan)
+
+- **`CanonicalItemUrl`** (`types/src/paths.rs`): newtype around `String`; `parse` / `parent` / `display_path` / `tilde_tail` / etc. Reducer `ContentState`, `VoteData`, ranking, RPC, search, garden, breadcrumbs all use it or `String` keys derived from it.
+- **`ThreadNav`** (`server/src/html/forum/nav.rs`): encodes scope prefixes for threads and garden URLs; **`RouteContext`** now wraps `ThreadNav` (`server/src/html/routing.rs`, re-exported from `server/src/html/mod.rs`) but **most HTML still takes `&ThreadNav` directly** — migration incomplete.
+- **URL normalization** lives in `types/src/url_normalize.rs` + `canonicalize_item` / `finalize_external_identity_url` in `paths.rs` (YouTube, sorted query params, room path `room_route_segment` in `paths.rs`).
+- **Room HTTP paths** are `/r/{short}{slug}` (fused segment); wire **`room_id`** remains `short/slug` for RPC/events.
+
+## Target architecture
+
+### `ItemId` (types)
+
+Suggested shape (adjust after profiling `Ord` / `Hash` / serde size):
+
+```text
+ItemId::Root                      — tilde ontology root (today `SLUG_TILDE_ONTOLOGY_ROOT`)
+ItemId::Local { segments }        — slug.social ~/… path as Vec<String> (lowercase segments, non-empty for non-root)
+ItemId::External { url: Url }    — normalized `url::Url` (crate `url` already in `slug-types`)
+```
+
+**API surface (minimum):**
+
+- `ItemId::parse(&str) -> Option<ItemId>` — single entry from DSL / user input / legacy wire (internally may call `canonicalize_item` + structured split).
+- `ItemId::to_wire_url(&self) -> String` — only for **external** boundaries if needed (HTTP fetch, rare assertions); avoid using as the primary key once maps use `ItemId`.
+- `parent`, `display_path`, `tilde_tail` / `tilde_http_tail`, `tilde_segments`, `last_segment`, `normalized_storage` — port from `CanonicalItemUrl`.
+- **`Ord` + `Hash` + `Eq`** stable for `BTreeSet` / `HashMap` (see `write_actor` scope-rank snapshots).
+- **`Serialize` / `Deserialize`** — decide **tagged JSON** for any persisted or API-carried structs (e.g. `VoteData` in tests). If RPC must stay stringy for clients, use a **DTO layer** that converts `ItemId` ↔ wire at the boundary only.
+
+**Remove:** `CanonicalItemUrl` type and all `path_types::CanonicalItemUrl` / `slug_types::paths::CanonicalItemUrl` exports once call sites are migrated. **`Borrow<str>`** on the old newtype goes away; update `nav!` / any code that assumed map keys borrowed as `str`.
+
+### `RouteContext` (server HTML)
+
+- **File:** `server/src/html/routing.rs` — **`RouteContext(ThreadNav)`** with `item_href`, `item_href_raw`, `thread_url`, `garden_root_url`, `room_url`, `From`/`Into` `ThreadNav`.
+- **Direction:** new code and refactored Maud should take **`&RouteContext`** (or owned where appropriate) instead of `&ThreadNav` when building links. Long term, **`item_href(&ItemId)`** should not parse strings — it should pattern-match `ItemId` and append tilde tail or `/-/…` external tail using the same rules as today’s `ThreadNav::garden_item_url`.
+
+### Axum / garden routes
+
+- **No** single catch-all route (explicit decision): keep the existing router layout in `server/src/lib.rs`.
+- Room routes stay **`/r/:room_key/...`** with `room_key` fused; parsing via `slug_types::room_id_from_route_segment` / `room_route_segment` in `paths.rs`.
+
+## Phased execution (recommended order)
+
+### Phase 0 — Preconditions (quick)
+
+1. Read **`AGENTS.md`** (UI contract, durability matrix, `RpcCommand` vs `HtmlUiAction`).
+2. Run **`cargo test --workspace`** and **`./scripts/clj-test.sh`** on clean `main` before large diffs; repeat after each phase.
+
+### Phase 1 — `ItemId` in `slug-types` (no server yet)
+
+1. Add **`ItemId`** (new file e.g. `types/src/item_id.rs` **or** inline at bottom of `paths.rs` — see **Module cycle** below).
+2. Implement **`ItemId::parse`** using existing **`canonicalize_item`** + normalization; port **`CanonicalItemUrl`** methods to **`ItemId`** with tests ported from `paths.rs` `#[cfg(test)] mod tests`.
+3. **`GardenItemUrl::from_stored(&ItemId, room_wire)`** (and thread helpers) — build absolute hrefs from structure, not from re-parsing a canonical string.
+4. **`TildeHttpPathTail::to_item_id`** (rename from `to_canonical`) / **`tilde_http_path_to_item_id`**.
+5. **`TildeOntologyPath::from_stored(&ItemId)`**.
+6. Export **`ItemId`** from **`types/src/lib.rs`**; update **`server/src/path_types.rs`** re-exports.
+7. **Delete `CanonicalItemUrl`** and fix all **in-crate** references in `types` only until `cargo test` passes for `slug-types`.
+
+**Module cycle trap:** `item_id.rs` must not `use crate::paths::{...}` if `paths.rs` also imports `ItemId` for `GardenItemUrl` in the same module. **Fix one of:**
+
+- **A)** Put `ItemId` **inside `paths.rs`** below `canonicalize_item` / helpers (simplest, large file), or  
+- **B)** Split **`canonicalize_item`** (+ dash host helpers + `finalize_external_identity_url`) into **`types/src/item_wire.rs`**, then `paths.rs` + `item_id.rs` both depend on `item_wire` only (cleaner, more files).
+
+### Phase 2 — Reducer + ranking (server core)
+
+1. **`server/src/reducer.rs`**: `ContentState` / `GroupState` / **`VoteData`** — replace **`CanonicalItemUrl`** with **`ItemId`** on all maps, sets, deques, vectors.
+2. **`apply_vote`**: normalize `a`/`b` via **`ItemId::parse`** or **`ItemId`**-aware logic (remove string round-trip).
+3. **`apply_ingest_to_content`**: **`dsl`** still yields strings for item titles in statements; normalize to **`ItemId`** at ingest boundary via **`ItemId::parse`** once per item.
+4. **`server/src/ranking.rs`**, **`server/src/scope_rank.rs`**, **`server/src/api/write_actor.rs`** (including **`BTreeSet`** ordering), **`server/src/api/validate.rs`**, **`server/src/api/helpers.rs`** — propagate **`ItemId`**.
+5. **`server/tests/basic.rs`** and any reducer tests constructing **`VoteData`** — use **`ItemId::parse(...).unwrap()`** or helpers.
+
+### Phase 3 — RPC + search + external resolver
+
+1. **`server/src/api/rpc.rs`**: rank/pair/matchup/search payloads; today many paths use **`GardenItemUrl::from_storage_str(item.as_str(), …)`** — switch to **`ItemId`** + **`GardenItemUrl::from_stored(&item_id, …)`** (or equivalent).
+2. **`server/src/html/search.rs`**: scoring uses item path strings — derive from **`ItemId::display_path`** / **`to_wire_url`** only at the scoring boundary if needed.
+3. **`server/src/external_resolver.rs`**: take **`&ItemId`** or **`ItemId::external_url()`** instead of **`&CanonicalItemUrl`**.
+
+### Phase 4 — HTML / Maud
+
+1. **`ThreadNav::garden_item_url`**: overload or replace with **`garden_item_href(&self, item: &ItemId)`** (no `CanonicalItemUrl::parse` inside).
+2. **`RouteContext`**: extend **`item_href(&ItemId)`**; migrate call sites from **`ThreadNav`** to **`RouteContext`** where only link-building is needed (keep **`ThreadNav`** where scope / auth helpers need the full struct).
+3. **`server/src/html/garden.rs`**, **`breadcrumb_path.rs`**, **`forum/*`**, **`editor.rs`**: replace **`CanonicalItemUrl`** with **`ItemId`**; breadcrumbs should walk **`ItemId::parent`** without string `rsplit`.
+4. **`types` JSON types** (`RankRow`, etc.): decide whether **`GardenItemUrl`** stays string for JSON or becomes a structured field; keep **one** wire format for the public API.
+
+### Phase 5 — Cleanup + docs
+
+1. Remove dead **`canonical_path`** / **`breadcrumb_path`** string logic if fully superseded.
+2. Update **`AGENTS.md`** if durability, `POST /ui`, or command surfaces change.
+3. Delete or shrink **`plan.md`** when done.
+
+## File / symbol checklist (non-exhaustive — grep-driven)
+
+Run periodically:
+
+```bash
+rg "CanonicalItemUrl" -g'*.rs'
+rg "path_types::CanonicalItemUrl" -g'*.rs'
+rg "tilde_http_path_to_canonical" -g'*.rs'
+```
+
+**High-touch files (from prior exploration):**
+
+| Area | Files |
+|------|--------|
+| Types | `types/src/paths.rs`, `types/src/lib.rs`, `types/src/url_normalize.rs`, (optional) `types/src/item_id.rs`, `types/src/item_wire.rs` |
+| Server re-exports | `server/src/path_types.rs`, `server/src/canonical_path.rs` |
+| Reducer / ingest | `server/src/reducer.rs`, `server/src/dsl.rs` (parse output types if changed) |
+| Ranking | `server/src/ranking.rs`, `server/src/scope_rank.rs` |
+| Writer / RPC | `server/src/api/write_actor.rs`, `server/src/api/rpc.rs`, `server/src/api/helpers.rs`, `server/src/api/validate.rs` |
+| HTML | `server/src/html/garden.rs`, `server/src/html/breadcrumb_path.rs`, `server/src/html/forum/nav.rs`, `server/src/html/routing.rs`, `server/src/html/search.rs`, `server/src/html/editor.rs`, `server/src/html/forum/ingest.rs`, … |
+| Tests | `server/tests/basic.rs`, `server/tests/integration.rs`, `types/src/paths.rs` tests, Clojure under `test/` if URLs/assertions mention canonical shapes |
+
+## Events / JSONL
+
+- **`Ingest`** events store **`raw` DSL** only — no change required for item identity inside the event.
+- If any future event type stores item ids as strings, migrate to **structured `ItemId` serde** or accept string only at the event boundary with immediate parse into **`ItemId`** on `apply_event`.
+
+## `nav!` macro (`server/src/paths.rs`)
+
+- Macros use **`keypath($key)`** with **`.clone()`** — **`ItemId`** must be **`Clone`** (already for enums). Remove any reliance on **`Borrow<str>`** for map keys.
+
+## Testing gate
+
+After each phase:
+
+```bash
+cargo test --workspace
+./scripts/clj-test.sh
+```
+
+## Risks / gotchas
+
+1. **`Ord` on `ItemId`**: must match prior **`CanonicalItemUrl`** / `String` ordering wherever **`BTreeSet`** is used (e.g. deterministic scope-rank snapshots in **`write_actor`**).
+2. **External `ItemId`**: **`Url`** equality / hashing — normalization is already centralized in **`url_normalize`**; ensure **`ItemId::parse`** always inserts normalized **`Url`** into **`External`**.
+3. **Fake parent URLs** in garden (e.g. **`https://.`** for external root ranking): find all **`parse("https://.")`** style hacks and express as **`ItemId`** or a dedicated sentinel.
+4. **Serde**: tests and any RPC clients that snapshot JSON may need expectation updates if **`VoteData`** shape changes.
+
+## Optional follow-ups (not blocking `ItemId`)
+
+- More **domain normalizers** in **`url_normalize.rs`** (e.g. `music.youtube.com`, Spotify, etc.).
+- **Room wire** vs **HTTP segment** helpers already in **`paths.rs`** (`ROOM_SHORT_ID_LEN`, `room_route_segment`, `room_id_from_route_segment`).
+
+---
+
+**End state criteria:** `rg CanonicalItemUrl` returns nothing; reducer maps use **`ItemId`**; HTML link generation for items goes through **`RouteContext` + `ItemId`**; tests and Kaocha green.
diff --git a/server/src/html/mod.rs b/server/src/html/mod.rs
index c3dccd03dc6da2a2f6f6fa828657e33a951884b1..668403a35c415e6f091362b28c63ac294057fb6d 100644
--- a/server/src/html/mod.rs
+++ b/server/src/html/mod.rs
@@ -15,6 +15,7 @@ mod breadcrumb_path;
 mod editor;
 mod forum;
 mod garden;
+pub mod routing;
 mod search;
 pub mod ui_action;
 use breadcrumb_path::{ExternalOntologyPath, OntologyPath};
@@ -35,6 +36,7 @@ pub use garden::{
     external_garden_index, external_ontology_path, garden_index, ontology_path, room_external_garden_index,
     room_external_ontology_path, room_garden_index, room_ontology_path,
 };
+pub use routing::RouteContext;
 pub use search::{search_page, search_results_fragment};
 pub use forum::user_profile_page;
 pub use ui_action::{parse_html_ui_from_form, HtmlUiAction, HtmlUiParseError, UI_RPC_FIELD};
diff --git a/server/src/html/routing.rs b/server/src/html/routing.rs
new file mode 100644
index 0000000000000000000000000000000000000000..13e9ae0cbe32d454e34a7737edf8234d2a558502
--- /dev/null
+++ b/server/src/html/routing.rs
@@ -0,0 +1,71 @@
+//! Scoped browser paths for HTML. **RouteContext** is the intended single place to build `href`s
+//! given public vs room scope (the original blueprint name); today it wraps [`ThreadNav`].
+//!
+//! Prefer `RouteContext::item_href` / [`RouteContext::thread_url`] in new Maud over stitching
+//! `/r/…` vs `/~` manually. Call sites can migrate incrementally from passing `&ThreadNav`.
+
+use crate::path_types::CanonicalItemUrl;
+
+use super::forum::ThreadNav;
+
+#[derive(Clone)]
+pub struct RouteContext(ThreadNav);
+
+impl RouteContext {
+    #[inline]
+    pub fn public() -> Self {
+        Self(ThreadNav::public())
+    }
+
+    #[inline]
+    pub fn from_room_id(room_id: &str) -> Option<Self> {
+        ThreadNav::from_room_id(room_id).map(Self)
+    }
+
+    #[inline]
+    pub fn thread_nav(&self) -> &ThreadNav {
+        &self.0
+    }
+
+    #[inline]
+    pub fn into_thread_nav(self) -> ThreadNav {
+        self.0
+    }
+
+    /// Relative path for a stored canonical item in this scope’s garden.
+    pub fn item_href(&self, item: &CanonicalItemUrl) -> String {
+        self.0.garden_item_url(item.as_str())
+    }
+
+    /// Same as [`Self::item_href`] but parses `item` first (raw DSL / user paste).
+    pub fn item_href_raw(&self, item: &str) -> String {
+        self.0.garden_item_url(item)
+    }
+
+    #[inline]
+    pub fn thread_url(&self, tag: &str) -> String {
+        self.0.thread_url(tag)
+    }
+
+    #[inline]
+    pub fn garden_root_url(&self) -> &str {
+        self.0.garden_root_url()
+    }
+
+    #[inline]
+    pub fn room_url(&self) -> &str {
+        self.0.room_url()
+    }
+}
+
+impl From<ThreadNav> for RouteContext {
+    fn from(nav: ThreadNav) -> Self {
+        Self(nav)
+    }
+}
+
+impl From<RouteContext> for ThreadNav {
+    fn from(ctx: RouteContext) -> Self {
+        ctx.0
+    }
+}


Side B — contributor: tommy-mor
Side B — commit message:
[9ecc4e2e] nice

Side B — unified diff (full patch):
diff --git a/server/src/api/rpc.rs b/server/src/api/rpc.rs
index 5675dcd2e28062acbe3c037b94b77c33adf32474..a18241f3ed7b4a248748fcefc799f9801ca62461 100644
--- a/server/src/api/rpc.rs
+++ b/server/src/api/rpc.rs
@@ -936,7 +936,15 @@ pub async fn handle_rpc_batch(
                 line_ok(RpcResult::ForumThreads(rpc_list_forum_threads(&reduced, &room)))
             }
             RpcCommand::RoomCreate { slug, visibility } => {
-                match verify_bearer_principal(&headers, &*state.reduced.read().await) {
+                // Scope the first read so its guard drops before any nested `read().await` / `write().await`.
+                // A guard from `match verify(..., &*state.reduced.read().await)` would otherwise live for the
+                // whole `match` and deadlock here (tokio::sync::RwLock is not reentrant).
+                let principal = {
+                    let reduced = state.reduced.read().await;
+                    verify_bearer_principal(&headers, &*reduced)
+                };
+                match principal {
+                    Err((_, m)) => line_err(m, None),
                     Ok(principal) => {
                         let slug = slug.trim().to_lowercase();
                         if slug.is_empty() || slug.len() > 64 {
@@ -994,7 +1002,6 @@ pub async fn handle_rpc_batch(
                             }
                         }
                     }
-                    Err((_, m)) => line_err(m, None),
                 }
             }
             RpcCommand::RoomGrant {
@@ -1002,17 +1009,28 @@ pub async fn handle_rpc_batch(
                 username,
                 capability,
             } => {
-                match verify_bearer_principal(&headers, &*state.reduced.read().await) {
+                let principal = {
+                    let reduced = state.reduced.read().await;
+                    verify_bearer_principal(&headers, &*reduced)
+                };
+                match principal {
                     Err((_, m)) => line_err(m, None),
                     Ok(principal) => {
-                        let reduced = state.reduced.read().await;
-                        if !reduced.user_has_cap(&room, &principal, ThreadCapability::Manage) {
+                        let can_manage = {
+                            let reduced = state.reduced.read().await;
+                            reduced.user_has_cap(&room, &principal, ThreadCapability::Manage)
+                        };
+                        if !can_manage {
                             line_err("requires Manage capability", None)
                         } else {
                             match parse_username(&username) {
                                 Err(msg) => line_err("invalid username", Some(msg)),
                                 Ok(target) => {
-                                    if !reduced.users_by_provider.values().any(|u| u == &target) {
+                                    let user_exists = {
+                                        let reduced = state.reduced.read().await;
+                                        reduced.users_by_provider.values().any(|u| u == &target)
+                                    };
+                                    if !user_exists {
                                         line_err(format!("user @{target} not found"), None)
                                     } else {
                                         match parse_capability(&capability) {
diff --git a/server/tests/integration.rs b/server/tests/integration.rs
index 9162bd5bee84035e3908bfb9c8e201b7878ca339..b930120da09fe7d307f0411b84fb639fb8bd0b15 100644
--- a/server/tests/integration.rs
+++ b/server/tests/integration.rs
@@ -95,6 +95,23 @@ async fn test_healthz() {
     assert_eq!(response.text().await.unwrap(), "ok");
 }
 
+#[tokio::test]
+async fn test_room_create_private_rpc() {
+    let (addr, _tmp, _log, _handle) = create_test_server().await;
+    let client = reqwest::Client::new();
+    let batch = serde_json::json!([{
+        "RoomCreate": { "slug": "secret-project", "visibility": "private" }
+    }]);
+    let body = rpc_batch(&client, addr, Some(&test_bearer()), batch).await;
+    let line = &body["results"][0];
+    assert_eq!(line["ok"], true, "room create: {:?}", line);
+    let room_id = line["result"]["RoomCreated"]["room_id"].as_str().unwrap();
+    assert!(
+        room_id.contains("/secret-project"),
+        "expected room_id to contain slug, got {room_id}"
+    );
+}
+
 #[tokio::test]
 async fn test_index_page() {
     // HTML routes are offline during the auth-v3 refactor.
diff --git a/test/auth.bb b/test/auth.bb
index 611e04f1806ef81d678a91f499597fe55f691dd7..a67cc1de763434176d2f68e419dd37a8cd328395 100644
--- a/test/auth.bb
+++ b/test/auth.bb
@@ -176,7 +176,7 @@
          (assert! (= 200 (:status poll)) "pending-session poll returns 200")
          (let [poll-json (json/parse-string (:body poll) true)]
            (assert! (:complete poll-json) "pending session complete=true")
-           (assert! (= "@bbuser" (:user poll-json)) "poll returns @bbuser")
+           (assert! (= "bbuser" (:user poll-json)) "poll returns stored username bbuser")
            (assert! (clojure.string/starts-with? (:token poll-json) "slug_") "poll returns bearer token")
 
            (println "\nwhoami…")
@@ -184,7 +184,7 @@
                                :headers {"Authorization" (str "Bearer " (:token poll-json))})]
              (assert! (= 200 (:status who)) "whoami returns 200")
              (let [who-json (json/parse-string (:body who) true)]
-               (assert! (= "@bbuser" (:user who-json)) "whoami user is @bbuser"))))))
+               (assert! (= "bbuser" (:user who-json)) "whoami user is bbuser (stored form)"))))))
 
      (println "\nCLI: identity start → OAuth → identity poll → whoami…")
      (let [cli-home (str tmp-dir "/cli-home")
@@ -208,7 +208,7 @@
                       (str "identity poll exits 0 (stderr: " (:err poll-proc) ")"))
              (let [poll-cli (json/parse-string (:out poll-proc) true)]
                (assert! (= "complete" (:phase poll-cli)) "identity poll --json phase")
-               (assert! (= "@cliuser" (:user poll-cli)) "CLI poll user")
+               (assert! (= "cliuser" (:user poll-cli)) "CLI poll user (stored form)")
                (assert! (clojure.string/starts-with? (:token poll-cli) "slug_") "CLI poll token")
                (let [token-path (str cli-home "/.config/slugsocial/token")]
                  (assert! (fs/exists? token-path) "token written under isolated HOME")
@@ -219,7 +219,7 @@
                  (assert! (zero? (:exit who-proc))
                           (str "whoami exits 0 (stderr: " (:err who-proc) ")"))
                  (let [who-cli (json/parse-string (:out who-proc) true)]
-                   (assert! (= "@cliuser" (:user who-cli)) "CLI whoami uses saved token"))))))))
+                   (assert! (= "cliuser" (:user who-cli)) "CLI whoami uses saved token"))))))))
 
      (finally
        (when-some [s @!server] (common/kill-server s))
diff --git a/test/common.bb b/test/common.bb
index ebb16c615a8b40e8830b5d5765d1e935a45538d0..4ed450377cdbb020f5ff164a812dfbbfc23c0f47 100644
--- a/test/common.bb
+++ b/test/common.bb
@@ -78,9 +78,20 @@
 
 (defn start-server
   "Start the slugsocial-server binary with the given env map.
-   Returns the babashka.process map."
-  [server-bin env-map]
-  (p/process [server-bin] {:out :inherit :err :inherit :env env-map}))
+   Returns the babashka.process map.
+
+   When `log-file` (string path) is provided, stdout and stderr are appended there
+   instead of inheriting the parent descriptors. Inheriting shared pipes while the
+   parent blocks on HTTP I/O can fill the pipe buffer and deadlock the server on log writes."
+  ([server-bin env-map]
+   (start-server server-bin env-map nil))
+  ([server-bin env-map log-file]
+   (p/process [server-bin]
+              (if log-file
+                ;; Two string paths (same file): babashka.process can deref the process cleanly.
+                ;; :err :out + ProcessBuilder$Redirect breaks stream copying in deref/kill-server.
+                {:env env-map :out log-file :err log-file}
+                {:out :inherit :err :inherit :env env-map}))))
 
 (defn kill-server
   "Forcibly kill a server process (babashka.process map) and wait for it to exit."
diff --git a/test/grants.bb b/test/grants.bb
index 7066c1f2a57f39a362052f8524206dccf37bb7b1..793ba216f2de76a77eb20a76d08403db07ff411b 100644
--- a/test/grants.bb
+++ b/test/grants.bb
@@ -35,13 +35,14 @@
 (defn- http-client []
   (-> (java.net.http.HttpClient/newBuilder)
       (.followRedirects java.net.http.HttpClient$Redirect/ALWAYS)
+      (.connectTimeout (java.time.Duration/ofSeconds 15))
       (.build)))
 
 (defn- http-get [url & {:keys [headers]}]
   (let [b (java.net.http.HttpRequest/newBuilder (java.net.URI/create url))]
     (doseq [[k v] (or headers {})]
       (.header b k v))
-    (let [req (-> b (.GET) (.build))
+    (let [req (-> b (.timeout (java.time.Duration/ofSeconds 60)) (.GET) (.build))
           resp (.send (http-client) req (java.net.http.HttpResponse$BodyHandlers/ofString))]
       {:status (.statusCode resp) :body (.body resp)})))
 
@@ -52,6 +53,7 @@
     (doseq [[k v] (or headers {})]
       (.header b k v))
     (let [req (-> b
+                  (.timeout (java.time.Duration/ofSeconds 60))
                   (.POST (java.net.http.HttpRequest$BodyPublishers/ofString body))
                   (.build))
           resp (.send (http-client) req (java.net.http.HttpResponse$BodyHandlers/ofString))]
@@ -67,6 +69,7 @@
         b (java.net.http.HttpRequest/newBuilder (java.net.URI/create url))]
     (.header b "Content-Type" "application/x-www-form-urlencoded")
     (let [req (-> b
+                  (.timeout (java.time.Duration/ofSeconds 60))
                   (.POST (java.net.http.HttpRequest$BodyPublishers/ofString pairs))
                   (.build))
           resp (.send (http-client) req (java.net.http.HttpResponse$BodyHandlers/ofString))]