fix: address autoresearch review follow-ups

docs/autoresearch.md

@@ -33,8 +33,17 @@ specification copied from an untrusted source. The workspace must be clean at
 startup. Rejected iterations are restored to their checkpoint; ignored
 configuration/state files are restored to their pre-iteration contents, and
 untracked files created by that iteration are removed. Build output under
-`target/` is treated as disposable process state. Run autoresearch in a
-dedicated worktree when experimenting with valuable local files.
+`target/`, dependency trees, and other volatile build roots are treated as
+disposable process state. The remaining ignored snapshot is disk-backed and
+bounded to 16 MiB per file and 64 MiB total. Run autoresearch in a dedicated
+worktree when experimenting with valuable local files.
+
+Baseline validation and metric commands must not modify tracked files, change
+the branch, or move `HEAD`; the run aborts and restores the baseline if they
+do. The controller scrubs secret-bearing environment variables from its
+subprocesses and makes its acceptance commit with repository hooks disabled.
+The restricted experiment agent receives repository instructions but not
+`USER.md` or `MEMORY.md`.
 
 If `ledger_path` is inside the workspace, it must be Git-ignored; an external
 ledger path is also supported. This keeps the durable ledger from becoming an

src/agent_http.rs

@@ -39,6 +39,10 @@ pub struct StateBody {
     pub engine: String,
     pub mode: String,
     pub cost_usd: f64,
+    /// False means the displayed cost excludes calls whose model price is
+    /// unknown.
+    pub pricing_complete: bool,
+    pub unpriced_call_count: usize,
     pub total_tokens: usize,
     pub call_count: usize,
     /// Input tokens of the most recent model call — what the agent actually
@@ -116,6 +120,8 @@ pub async fn build_state(runner: &AgentRunner, chat_id: &str) -> StateBody {
         engine: "rx4".into(),
         mode: mode_name(&runner.get_mode()).into(),
         cost_usd: summary.total_cost,
+        pricing_complete: summary.pricing_complete,
+        unpriced_call_count: summary.unpriced_call_count,
         total_tokens: summary.total_tokens,
         call_count: summary.call_count,
         context_tokens,
@@ -621,6 +627,8 @@ mod tests {
             engine: "rx4".into(),
             mode: "auto".into(),
             cost_usd: 0.25,
+            pricing_complete: true,
+            unpriced_call_count: 0,
             total_tokens: 1234,
             call_count: 3,
             context_tokens: 8000,
@@ -636,6 +644,8 @@ mod tests {
             "engine",
             "mode",
             "cost_usd",
+            "pricing_complete",
+            "unpriced_call_count",
             "total_tokens",
             "call_count",
             "context_tokens",