build: the lift matches cards by identity, not position

Sequences were walked side by side, so a rebuilt page whose cards sit
in a different order, or with a new card in front, gave every card
its neighbour's words: uhhm.no's front door got "Attention is earned"
on the wrong card that way. Items now match on the bucket they record
into, the page they lead to, a field's name, the group they invite
into or the bucket they read; information features with no identity
still match by position. Lifting live uhhm.no into the current plan
puts each of its three headings on the card with its bucket.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

README.md

@@ -45,9 +45,12 @@ default.
 
 `--words-from <repo>` lifts the words a repo already has back into the
 words map by structure, so a rebuild keeps hand-edited wording and
-adds only the new slots. The lift is by position, so it holds for a
-rebuild of the same shape and not across a restructure: on uhhm.no it
-put the old front door's three headings on three new cards. `--words
+adds only the new slots. Cards are matched by what they are (the
+bucket they record into, the page they lead to, a field's name, the
+group they invite into), not by position, so a rebuild that reorders
+or reshapes a page still hands each card its own words; on uhhm.no
+the live front door's three headings follow their buckets onto the
+new page. `--words
 <json>` overrides slots by page and may repeat, each file layered over
 the last (the words a build used, then a round of hand changes);
 `words.json` in the output is the words as used. See

src/build.rs

@@ -1693,6 +1693,39 @@ pub fn lift_words(plan: &Plan, repo: &Path) -> anyhow::Result<BTreeMap<String, V
     Ok(fallback.iter().map(|(path, v)| (path.clone(), untokenize(v, path, "", &found))).collect())
 }
 
+/// What identifies an item of a page regardless of its position: the
+/// structural strings portal itself keys on. Word slots (tokens on
+/// the rendered side) are never an identity.
+fn identity(v: &serde_yaml::Value) -> Vec<(&'static str, String)> {
+    const KEYS: &[&[&str]] = &[
+        &["record_as"],
+        &["action"],
+        &["name"],
+        &["invite", "group"],
+        &["resource", "source", "bucket"],
+        &["self_transition", "bucket"],
+    ];
+    let mut id = Vec::new();
+    for path in KEYS {
+        let mut at = v;
+        for key in path.iter() {
+            match at.get(*key) {
+                Some(next) => at = next,
+                None => {
+                    at = &serde_yaml::Value::Null;
+                    break;
+                }
+            }
+        }
+        if let Some(s) = at.as_str() {
+            if !s.starts_with('\u{1}') && !s.is_empty() {
+                id.push((path[0], s.to_string()));
+            }
+        }
+    }
+    id
+}
+
 fn token(path: &str, pointer: &str) -> String {
     format!("\u{1}{path}\u{2}{pointer}\u{3}")
 }
@@ -1725,9 +1758,27 @@ fn collect_words(a: &serde_yaml::Value, b: &serde_yaml::Value, found: &mut std::
                 }
             }
         }
+        // Items are matched by what they are, not where they sit: the
+        // bucket an alternative records into, the page it leads to, a
+        // field's name, the group it invites into. A rebuild that
+        // reorders or reshapes a page still hands each card its own
+        // words; on uhhm.no a positional match once put the old front
+        // door's three headings on three unrelated new cards. Items
+        // with no identity (an information feature) fall back to
+        // position.
         (serde_yaml::Value::Sequence(x), serde_yaml::Value::Sequence(y)) => {
-            for (va, vb) in x.iter().zip(y) {
-                collect_words(va, vb, found);
+            let mut used = vec![false; y.len()];
+            for (i, va) in x.iter().enumerate() {
+                let ida = identity(va);
+                let hit = if ida.is_empty() {
+                    (i < y.len() && !used[i] && identity(&y[i]).is_empty()).then_some(i)
+                } else {
+                    (0..y.len()).find(|&j| !used[j] && identity(&y[j]).iter().any(|k| ida.contains(k)))
+                };
+                if let Some(j) = hit {
+                    used[j] = true;
+                    collect_words(va, &y[j], found);
+                }
             }
         }
         (serde_yaml::Value::String(t), serde_yaml::Value::String(live)) if t.starts_with('\u{1}') => {
@@ -1786,4 +1837,25 @@ mod invite_tests {
         assert!(lifted2["questions/trafikk/index.yaml"]["invites"]["styret"]["heading"].as_str().unwrap().starts_with("Invite someone"));
         let _ = std::fs::remove_dir_all(&dir);
     }
+
+    #[test]
+    fn words_follow_their_card_when_the_live_page_is_ordered_differently() {
+        let plan: Plan = serde_yaml::from_str(PLAN).unwrap();
+        let mut words = fallback_words(&plan);
+        words.get_mut("questions/index.yaml").unwrap()["needs"]["melde"]["heading"] = json!("Meld fra om en fare");
+        let dir = std::env::temp_dir().join(format!("lift-order-{}", std::process::id()));
+        write_repo(&dir, render(&plan, &words).unwrap()).unwrap();
+        // The live front door has an extra first card the plan knows
+        // nothing about, so the hazard form is no longer first.
+        let front = dir.join("questions/index.yaml");
+        let mut page: serde_yaml::Value = serde_yaml::from_str(&std::fs::read_to_string(&front).unwrap()).unwrap();
+        let alternatives = page["alternatives"].as_sequence_mut().unwrap();
+        let stranger: serde_yaml::Value = serde_yaml::from_str("name: About us\ndescription: Founded in 1947.\nrecord_as: newsletter\nfeatures: []\n").unwrap();
+        alternatives.insert(0, stranger);
+        std::fs::write(&front, serde_yaml::to_string(&page).unwrap()).unwrap();
+
+        let lifted = lift_words(&plan, &dir).unwrap();
+        assert_eq!(lifted["questions/index.yaml"]["needs"]["melde"]["heading"], "Meld fra om en fare");
+        let _ = std::fs::remove_dir_all(&dir);
+    }
 }