From b5f5c46ff201654bcd8661f030301b41ac0051c8 Mon Sep 17 00:00:00 2001
From: Bendik Aagaard Lynghaug <bendik.lynghaug@gmail.com>
Date: Wed, 23 Sep 2026 23:00:37 +0200
Subject: [PATCH] build --llm-for <prefix>, and rate reads the mails

The words model can be pointed at a subset of pages (`aggregates.yaml`
for the mails alone) while everything else keeps the words it has.
`rate --repo` rates each mail's prose on its own, LIX and length,
since a person reads a mail with no card around it.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
---
 src/build.rs |  8 ++++++++
 src/rate.rs  | 50 ++++++++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 58 insertions(+)

diff --git a/src/build.rs b/src/build.rs
index 7a27983..6732a87 100644
--- a/src/build.rs
+++ b/src/build.rs
@@ -1473,7 +1473,15 @@ pub fn run(argv: &[String]) -> anyhow::Result<()> {
     }
     if let Some(cmd) = value("--llm") {
         std::fs::create_dir_all(out.join("llm"))?;
+        // --llm-for <prefix>: the model writes only the pages whose
+        // path starts with it (`aggregates.yaml` for the mails alone,
+        // `questions/received/` for the followups), and everything
+        // else keeps the words it has.
+        let only = value("--llm-for");
         for (n, (path, fallback)) in words.clone().iter().enumerate() {
+            if only.as_ref().is_some_and(|p| !path.starts_with(p.as_str())) {
+                continue;
+            }
             let prompt = words_prompt(&plan, path, fallback);
             let reply = crate::llm_with(&cmd, &prompt, &[("FORMAT", "json"), ("TEMPERATURE", "0.4")]).unwrap_or_default();
             std::fs::write(out.join(format!("llm/{:02}-{}.md", n + 1, path.replace('/', "_"))), format!("{prompt}\n\n=== REPLY ===\n{reply}"))?;
diff --git a/src/rate.rs b/src/rate.rs
index 0dc77c4..17beda5 100644
--- a/src/rate.rs
+++ b/src/rate.rs
@@ -158,6 +158,41 @@ fn followup_page(repo: Option<&Path>, step: &Value) -> Option<Value> {
 
 /// `/` is `questions/index.yaml`; `/a/b` is `questions/a/b.yaml` or
 /// `questions/a/b/index.yaml`.
+/// Every `mail:` on a state in the repo's aggregates.yaml, as
+/// (`bucket:state`, subject, body) with the `{placeholders}` taken
+/// out, so they read as the person will read them.
+pub(crate) fn mail_texts(repo: &Path) -> Option<Vec<(String, String, String)>> {
+    let raw = std::fs::read_to_string(repo.join("aggregates.yaml")).ok()?;
+    let doc: Value = serde_yaml::from_str(&raw).ok()?;
+    let strip = |s: &str| {
+        let mut out = String::new();
+        let mut skip = false;
+        for c in s.chars() {
+            match c {
+                '{' => skip = true,
+                '}' => skip = false,
+                _ if !skip => out.push(c),
+                _ => {}
+            }
+        }
+        out.split_whitespace().collect::<Vec<_>>().join(" ")
+    };
+    let mut out = Vec::new();
+    for aggregate in doc["aggregates"].as_array().into_iter().flatten() {
+        let bucket = aggregate["bucket"].as_str().unwrap_or_default();
+        for (state, spec) in aggregate["states"].as_object().into_iter().flatten() {
+            if let Some(mail) = spec.get("mail") {
+                out.push((
+                    format!("{bucket}:{state}"),
+                    strip(mail["subject"].as_str().unwrap_or_default()),
+                    strip(mail["body"].as_str().unwrap_or_default()),
+                ));
+            }
+        }
+    }
+    Some(out)
+}
+
 pub(crate) fn load_page(repo: &Path, route: &str) -> Option<Value> {
     let rel = route.trim_matches('/');
     let candidates = if rel.is_empty() {
@@ -213,6 +248,21 @@ pub fn rate(argv: &[String]) -> anyhow::Result<()> {
     }
 
     let mut report = BTreeMap::new();
+    // The mails first, when the repo carries any: a person reads a
+    // mail alone, with no card around it, so its prose is rated on
+    // its own - LIX and length.
+    if let Some(mails) = repo.as_deref().and_then(mail_texts) {
+        if !mails.is_empty() {
+            println!("{:<40} {:>6} {:<14} {:>6}", "mail", "LIX", "", "words");
+            for (name, subject, body) in &mails {
+                let text = format!("{subject}. {body}");
+                let score = lix(&text);
+                println!("{:<40} {:>6.0} {:<14} {:>6}", name, score, lix_band(score), text.split_whitespace().count());
+                report.insert(format!("mail {name}"), json!({ "lix": score, "words": text.split_whitespace().count() }));
+            }
+            println!();
+        }
+    }
     println!("{:<28} {:>6} {:<14} {:>8} {:>8} {:>8} {:>8}   {:<22}   {:<22}", "need", "LIX", "", "which", "where", "asked", "next", "gutted twin says", "page load");
     for need in model["needs"].as_array().into_iter().flatten() {
         let id = need["id"].as_str().unwrap_or_default();