World's first open-source enterprise world model.
Rust agent-memorybitemporalgraphragknowledge-baseknowledge-graphllmontologypgvectorpostgresqlragrustself-hostedsemantic-searchtemporal-knowledge-graphworld-model
A relation says whether it happens or holds (#593)
WaylandYang283829645593
diff --git a/crates/utopia-server/src/bootstrap_ontology.rs b/crates/utopia-server/src/bootstrap_ontology.rsindex 1c0922b0..b6b9ff92 100644--- a/crates/utopia-server/src/bootstrap_ontology.rs+++ b/crates/utopia-server/src/bootstrap_ontology.rs@@ -318,8 +318,19 @@ pub async fn bootstrap_ontology(state: &AppState, kb_id: Uuid) -> anyhow::Result// 指引,而这里没有可信的来源可写。编一句反而是往提示词里塞一个没人负责的断言,// 而 key 本身(`runs_on`、`available_on`)已经说清楚了。//- // temporal 一律 state:它只在 functional / inverse_functional 为真时驱动时态- // 引擎,而这条路**永不**自动设那两位(见文件头),所以此处它没有行为后果。+ /* **temporal 从提案里取,不再一律 state。**+ 这里原来写死 state,理由是「它只在 functional / inverse_functional 为真时+ 驱动时态引擎,而这条路永不自动设那两位」。那个理由**在 0031(#486)之后+ 已经不成立**:`Validity::under` 现在按谓词的 temporal 规整每一次写入,+ event 会把 `valid_to` 收到与 `valid_from` 同一刻。写死 state 的后果是+ `Meridian invested Kestrel 2025-05-09` 被记成「从那天起一直在投」,+ 时间轴上读出来就是这样——而它是一件发生过的事,不是一个持续的状态。++ 答案本来就在手边:提案那一步问模型要的 JSON 里就有+ `"temporal":"state|event|eternal"`,而下面的类、属性、map_to 都在用同一份+ 提案,唯独关系这一档把它丢了。**采纳与否仍旧只由票数决定**(0007),+ 模型只回答「这个关系是哪一种」——那是计数答不了的意义问题。 */+ let temporal_of = proposed_temporal(&proposals);for g in &counted {// 本体里已经有等价关系就不新建,直接把事实改写过去。// 被动形(`produced_by` 对上 `produces`)改写时主宾对调@@ -327,12 +338,13 @@ pub async fn bootstrap_ontology(state: &AppState, kb_id: Uuid) -> anyhow::ResultSome(hit) => hit,None => {let label = g.key.replace('_', " ");+ let temporal = temporal_of(&g.key, &g.forms);match utopia_store::ontology::create_relation_type(&state.pool,kb_id,&g.key,&label,- "state",+ temporal,// 冷启动不替人声明任何公理:推理机的判据必须是人写下来的Default::default(),"",@@ -371,6 +383,7 @@ pub async fn bootstrap_ontology(state: &AppState, kb_id: Uuid) -> anyhow::Resulttracing::info!(%kb_id, key = %g.key, forms = ?g.forms, docs = g.docs, facts = g.facts, moved, left_off,corrected, reused = g.existing.is_some(), swap,+ temporal = temporal_of(&g.key, &g.forms),"按票数采纳关系");moved_total += moved;@@ -571,6 +584,40 @@ pub async fn bootstrap_ontology(state: &AppState, kb_id: Uuid) -> anyhow::ResultOk(())}+/// 从提案里查一个说法的 temporal,查不到给 `state`。+///+/// 按 key 查,再按 forms 查——计数那条路的规范 key 取的是组里事实最多的那个说法,+/// 模型提案挑的未必是同一个(`acquires` 对 `acquired`),但它们同组,说的是一件事。+///+/// **查不到就 state**,与从前一致:这一档只在有明确答案时才偏离缺省,+/// 而 state 是三者里最保守的——它不会像 event 那样把 `valid_to` 收成一个点。+fn proposed_temporal(proposals: &serde_json::Value) -> impl Fn(&str, &[String]) -> &'static str {+ let mut by_word: HashMap<String, &'static str> = HashMap::new();+ if let Some(list) = proposals.get("relation_types").and_then(|v| v.as_array()) {+ for p in list {+ let t = match str_of(p, "temporal") {+ Some("event") => "event",+ Some("eternal") => "eternal",+ // 模型偶尔编第四个值;不认识的一律当没说+ _ => continue,+ };+ let mut words: Vec<String> = str_of(p, "key").map(str::to_string).into_iter().collect();+ if let Some(forms) = p.get("forms").and_then(|v| v.as_array()) {+ words.extend(forms.iter().filter_map(|f| f.as_str()).map(str::to_string));+ }+ for w in words {+ by_word.insert(w.to_lowercase(), t);+ }+ }+ }+ move |key: &str, forms: &[String]| {+ std::iter::once(key)+ .chain(forms.iter().map(String::as_str))+ .find_map(|w| by_word.get(&w.to_lowercase()).copied())+ .unwrap_or("state")+ }+}+fn str_of<'a>(v: &'a serde_json::Value, key: &str) -> Option<&'a str> {v.get(key).and_then(|x| x.as_str())@@ -583,6 +630,44 @@ mod tests {use super::*;use utopia_core::models::TypeCandidate;+ /// 一个关系是发生过一次,还是持续成立。+ ///+ /// 从前这里写死 `state`,于是「Meridian 2025-05-09 投了 Kestrel」被记成+ /// 从那天起一直在投。答案本来就在提案里,只是没人读。+ #[test]+ fn a_relation_takes_the_temporal_it_was_proposed_with() {+ let proposals = serde_json::json!({+ "relation_types": [+ { "key": "acquired", "temporal": "event", "forms": ["acquires", "acquisition_of"] },+ { "key": "capital_of", "temporal": "eternal", "forms": [] },+ { "key": "employs", "temporal": "state", "forms": ["employed_by"] },+ // 模型偶尔编一个第四值:当没说,落回缺省+ { "key": "sponsors", "temporal": "ongoing", "forms": [] },+ ]+ });+ let t = proposed_temporal(&proposals);++ assert_eq!(t("acquired", &[]), "event");+ // 规范 key 取的是组里事实最多的说法,未必是模型挑的那个——按 forms 也要查得到+ assert_eq!(t("acquisition_of", &[]), "event");+ assert_eq!(t("Acquires", &[]), "event", "大小写不该影响");+ assert_eq!(+ t("bought", &["acquires".into()]),+ "event",+ "本名查不到时看同组说法"+ );+ assert_eq!(t("capital_of", &[]), "eternal");+ assert_eq!(t("employs", &[]), "state");++ // **查不到一律 state**:三者里最保守的,不会像 event 那样把 valid_to 收成一个点+ assert_eq!(t("sponsors", &[]), "state", "不认识的值当没说");+ assert_eq!(t("never_proposed", &[]), "state");+ assert_eq!(+ proposed_temporal(&serde_json::json!({}))("anything", &[]),+ "state"+ );+ }+fn cand(key: &str, distance: f32) -> TypeCandidate {TypeCandidate {id: Uuid::now_v7(),diff --git a/docs/decisions/0007-who-decides-what-becomes-a-relation.md b/docs/decisions/0007-who-decides-what-becomes-a-relation.mdindex 804d5ad9..55d618be 100644--- a/docs/decisions/0007-who-decides-what-becomes-a-relation.md+++ b/docs/decisions/0007-who-decides-what-becomes-a-relation.md@@ -3,7 +3,12 @@- **Status**: Built · adoption by counting (`MIN_DOCS = 2`; an LLM run needs `MIN_SIGNALS= 3`); six defects fixed; proposals persist in `ontology_proposals` (#112); open:narrative verbs in the ontology, `merge_key` not folding `_by`; the starting point (10- seeds, the `related_to` share) no longer exists.+ seeds, the `related_to` share) no longer exists · 2026-09-10: the counting path now+ takes `temporal` from the proposal it already reads for classes and attributes, instead+ of writing `state` for every relation it adopts — the comment that justified `state`+ ("temporal has no consequence unless functional is set") stopped being true at+ [0031](0031-an-event-holds-at-the-moment-it-names.md), which normalises every write by+ it; counting still decides *whether*, the model only answers *which kind*.- **Written**: 2026-08-30 · condensed into English 2026-09-03- **Related**: [0006](0006-ontology-scale-and-the-prompt.md) for the bench and itscaveats; [0010](0010-no-relation-is-no-relation.md) and