补全指数与完整范围排名的只读事件计算

crates/fidc-core/src/daily_patterns.rs

@@ -33,7 +33,10 @@ pub struct PatternSpec {
     pub expression: Option<crate::factor_events::Expr>,
 }
 impl PatternSpec {
-    pub fn validate(mut self) -> Result<Self, String> {
+    pub fn validate(self) -> Result<Self, String> {
+        self.validate_with_context(&[])
+    }
+    fn validate_with_context(mut self, context_fields: &[&str]) -> Result<Self, String> {
         if (self.template == "expression") != self.expression.is_some() {
             return Err("expression_template_requires_expression_only".into());
         }
@@ -53,7 +56,7 @@ impl PatternSpec {
             ];
             let missing = crate::factor_events::field_dependencies(expr)
                 .into_iter()
-                .filter(|f| !supported.contains(&f.as_str()))
+                .filter(|f| !supported.contains(&f.as_str()) && !context_fields.contains(&f.as_str()))
                 .collect::<Vec<_>>();
             if !missing.is_empty() {
                 return Err(format!(
@@ -191,6 +194,16 @@ pub fn evaluate(
     spec: &PatternSpec,
     days: &[NaiveDate],
     series: &PatternSeries,
+) -> Result<PatternResult, String> {
+    evaluate_with_context(spec, days, series, &BTreeMap::new(), false)
+}
+
+fn evaluate_with_context(
+    spec: &PatternSpec,
+    days: &[NaiveDate],
+    series: &PatternSeries,
+    context: &BTreeMap<String, Vec<Option<f64>>>,
+    numeric_output: bool,
 ) -> Result<PatternResult, String> {
     if days.len() != spec.history_len() || days.windows(2).any(|w| w[0] >= w[1]) {
         return Err("pattern_calendar_incomplete: 需要完整、唯一且递增的真实交易日窗口".into());
@@ -369,6 +382,13 @@ pub fn evaluate(
                     .collect::<Result<Vec<_>, String>>()?;
                 fields.insert(name.into(), values);
             }
+            for (name, values) in context {
+                if fields.contains_key(name) || values.len() != days.len()
+                    || values.iter().flatten().any(|v| !v.is_finite()) {
+                    return Err(format!("research_context_invalid: {} {name}", series.symbol));
+                }
+                fields.insert(name.clone(), values.clone());
+            }
             let frame = crate::factor_events::Frame {
                 symbol: series.symbol.clone(),
                 frequency: "1d".into(),
@@ -383,6 +403,12 @@ pub fn evaluate(
             result.values["expression_contract"] = json!(crate::factor_events::CONTRACT);
             result.values["price_policy"] = json!("backward1_anchored_to_decision_close");
             result.score = latest;
+            if numeric_output {
+                if values.value_type != crate::factor_events::ValueType::Number {
+                    return Err("research_rank_input_requires_numeric_expression".into());
+                }
+                return Ok(result);
+            }
             if latest.is_none() {
                 result.exclusion = Some(
                     json!({"reason":"expression_undefined_or_warmup","signal_date":days.last()}),
@@ -651,6 +677,60 @@ pub fn evaluate_batch(
     )
 }
 
+/// Research transport only. Strategy PatternSpec validation still rejects these fields.
+#[derive(Debug, Clone, Default, Deserialize)]
+#[serde(deny_unknown_fields)]
+pub struct ResearchContext {
+    #[serde(default)]
+    pub common: BTreeMap<String, Vec<Option<f64>>>,
+    #[serde(default)]
+    pub by_symbol: BTreeMap<String, BTreeMap<String, Vec<Option<f64>>>>,
+}
+
+pub fn evaluate_research_batch(
+    spec: PatternSpec, days: &[NaiveDate], series: &[PatternSeries],
+    context: &ResearchContext, numeric_output: bool,
+) -> Result<Value, String> {
+    let common_fields = ["index_open", "index_high", "index_low", "index_close"];
+    let symbol_fields = ["scope_rank", "scope_percentile", "scope_size"];
+    if spec.template != "expression" || series.is_empty() || series.len() > 200
+        || series.iter().map(|s| &s.symbol).collect::<BTreeSet<_>>().len() != series.len()
+        || context.common.keys().any(|k| !common_fields.contains(&k.as_str()))
+        || context.by_symbol.iter().any(|(s, fields)| !series.iter().any(|row| &row.symbol == s)
+            || fields.keys().any(|k| !symbol_fields.contains(&k.as_str()))) {
+        return Err("research_context_scope_or_fields_invalid".into());
+    }
+    for (name, values) in &context.common {
+        if values.len() != days.len() || values.iter().any(|v| !v.is_some_and(|x| x.is_finite() && x > 0.0)) {
+            return Err(format!("research_index_window_incomplete: {name}"));
+        }
+    }
+    let allowed = common_fields.into_iter().chain(symbol_fields).collect::<Vec<_>>();
+    let spec = spec.validate_with_context(&allowed)?;
+    let dependencies = crate::factor_events::field_dependencies(spec.expression.as_ref().unwrap());
+    let mut rows = Vec::with_capacity(series.len());
+    for item in series {
+        let mut fields = context.common.clone();
+        fields.extend(context.by_symbol.get(&item.symbol).cloned().unwrap_or_default());
+        if allowed.iter().any(|f| dependencies.contains(*f) && !fields.contains_key(*f)) {
+            return Err(format!("research_context_missing: {}", item.symbol));
+        }
+        for (name, values) in &fields {
+            if values.len() != days.len() || values.iter().flatten().any(|v| !v.is_finite()
+                || (name == "scope_percentile" && !(0.0..=1.0).contains(v))
+                || (matches!(name.as_str(), "scope_rank" | "scope_size") && *v < 1.0)) {
+                return Err(format!("research_context_invalid: {} {name}", item.symbol));
+            }
+        }
+        let mut result = evaluate_with_context(&spec, days, item, &fields, numeric_output)?;
+        result.values["research_context_latest"] = json!(fields.iter().map(|(k,v)|(k,v.last().copied().flatten())).collect::<BTreeMap<_,_>>());
+        rows.push(result);
+    }
+    Ok(json!({"contract":CONTRACT,"context_contract":"fidc_research_event_context_v1","spec":spec,
+        "required_history":spec.history_len(),"rows":rows,"read_only":true,
+        "source_evidence_verified":false,"live_routing":false,"rule_backtest_supported":false}))
+}
+
 pub fn expression_specs(expression: &str) -> Result<Vec<PatternSpec>, String> {
     let mut specs = Vec::new();
     for helper in ["pattern_signal", "pattern_score"] {
@@ -687,6 +767,37 @@ pub fn expression_specs(expression: &str) -> Result<Vec<PatternSpec>, String> {
 mod tests {
     use super::*;
     #[test]
+    fn research_index_and_ranking_context_never_unlock_strategy_mapping() {
+        let days=["2026-09-04","2026-09-07","2026-09-08"].map(|s|s.parse::<NaiveDate>().unwrap());
+        let spec:PatternSpec=serde_json::from_value(json!({"template":"expression","parameters":{"history_window":3},
+            "expression":{"kind":"operator","name":"CROSS_ABOVE","args":[{"kind":"field","name":"close"},{"kind":"field","name":"index_close"}]}})).unwrap();
+        assert!(spec.clone().validate().unwrap_err().contains("mapping_required"));
+        let series:PatternSeries=serde_json::from_value(json!({"symbol":"TEST","bars":days.iter().zip([9.0,10.0,11.0]).map(|(d,c)|json!({"date":d,"open":c,"high":c,"low":c,"close":c,"volume":100.0,"adjustment_factor_backward1":1.0,"paused":false})).collect::<Vec<_>>()})).unwrap();
+        let mut context=ResearchContext{common:BTreeMap::from([("index_close".into(),vec![Some(10.0);3])]),..Default::default()};
+        let result=evaluate_research_batch(spec.clone(),&days,&[series.clone()],&context,false).unwrap();
+        assert_eq!(result["rows"][0]["matched"],true);
+        assert_eq!(result["source_evidence_verified"],false);
+        assert_eq!(result["rule_backtest_supported"],false);
+        context.common.get_mut("index_close").unwrap()[1]=None;
+        assert!(evaluate_research_batch(spec.clone(),&days,&[series.clone()],&context,false).unwrap_err().contains("index_window_incomplete"));
+        context.common=BTreeMap::from([("close".into(),vec![Some(10.0);3])]);
+        assert!(evaluate_research_batch(spec,&days,&[series],&context,false).is_err());
+    }
+
+    #[test]
+    fn research_numeric_output_keeps_warmup_unknown_without_a_false_signal() {
+        let days=["2026-09-04","2026-09-07","2026-09-08"].map(|s|s.parse::<NaiveDate>().unwrap());
+        let spec:PatternSpec=serde_json::from_value(json!({"template":"expression","parameters":{"history_window":3},
+            "expression":{"kind":"operator","name":"PCT_CHANGE","window":2,"args":[{"kind":"field","name":"close"}]}})).unwrap();
+        let series:PatternSeries=serde_json::from_value(json!({"symbol":"TEST","bars":days.iter().zip([10.0,10.5,11.0]).map(|(d,c)|json!({"date":d,"open":c,"high":c,"low":c,"close":c,"volume":100.0,"adjustment_factor_backward1":1.0,"paused":false})).collect::<Vec<_>>()})).unwrap();
+        assert!(evaluate_batch(spec.clone(),&days,&[series.clone()]).is_err());
+        let result=evaluate_research_batch(spec,&days,&[series],&ResearchContext::default(),true).unwrap();
+        let values=&result["rows"][0]["values"]["expression"]["values"];
+        assert!(values[0].is_null() && values[1].is_null());
+        assert!((values[2].as_f64().unwrap()-0.1).abs()<1e-12);
+        assert_eq!(result["rows"][0]["matched"],false);
+    }
+    #[test]
     fn expression_condition_preserves_native_types_and_rejects_numeric_as_signal() {
         let make = |expression: Value| {
             serde_json::from_value::<PatternSpec>(json!({"template":"expression","parameters":{"history_window":3},"expression":expression})).unwrap().validate().unwrap()

crates/fidc-core/src/factor_cross_section.rs

@@ -28,6 +28,36 @@ pub struct Output {
     pub value: f64,
 }
 
+/// Every date ranks the same frozen research universe; unknown inputs invalidate the whole date.
+pub fn rank_history(
+    dates: &[chrono::NaiveDate], universe: &[String], values: &BTreeMap<String, Vec<Option<f64>>>,
+) -> Result<serde_json::Value, String> {
+    use serde_json::json;
+    if dates.is_empty() || dates.windows(2).any(|w| w[0] >= w[1]) || universe.len() < 2
+        || universe.len() > 20_000 || dates.len().saturating_mul(universe.len()) > 2_000_000
+        || universe.iter().collect::<BTreeSet<_>>().len() != universe.len()
+        || values.keys().collect::<BTreeSet<_>>() != universe.iter().collect::<BTreeSet<_>>()
+        || values.values().any(|v| v.len() != dates.len() || v.iter().flatten().any(|v| !v.is_finite())) {
+        return Err("research_rank_history_incomplete_or_invalid_universe".into());
+    }
+    let mut rank = universe.iter().map(|s|(s.clone(),vec![None;dates.len()])).collect::<BTreeMap<_,_>>();
+    let mut percentile = rank.clone();
+    let mut unknown_dates = Vec::new();
+    for (i, date) in dates.iter().enumerate() {
+        let missing = universe.iter().filter(|s|values[*s][i].is_none()).collect::<Vec<_>>();
+        if !missing.is_empty() {
+            unknown_dates.push(json!({"date":date,"missing_count":missing.len(),"missing_symbol_sample":missing.iter().take(20).collect::<Vec<_>>(),"sample_limit":20}));
+            continue;
+        }
+        let observations = universe.iter().map(|s|Observation{symbol:s.clone(),value:values[s][i].unwrap(),industry:None,market_cap:None}).collect::<Vec<_>>();
+        for item in evaluate("RANK", universe, &observations, 0.0)? {rank.get_mut(&item.symbol).unwrap()[i]=Some(item.value);}
+        for item in evaluate("PERCENTILE", universe, &observations, 0.0)? {percentile.get_mut(&item.symbol).unwrap()[i]=Some(item.value);}
+    }
+    Ok(json!({"rank":rank,"percentile":percentile,"unknown_dates":unknown_dates,
+        "universe":universe,"dates":dates,"tie_policy":"average_rank_descending",
+        "membership_policy":"fixed_research_scope_not_historical_index_membership"}))
+}
+
 fn mean(values: &[f64]) -> f64 {
     let base = values[0];
     base + values
@@ -153,6 +183,20 @@ pub fn evaluate(
 #[cfg(test)]
 mod tests {
     use super::*;
+    #[test]
+    fn historical_ranks_keep_ties_and_unknown_full_cross_sections() {
+        let dates=["2026-09-07","2026-09-08","2026-09-09"].map(|d|d.parse().unwrap());
+        let universe=vec!["A".into(),"B".into(),"C".into()];
+        let values=BTreeMap::from([("A".into(),vec![None,Some(10.0),Some(20.0)]),("B".into(),vec![Some(10.0),Some(10.0),Some(10.0)]),("C".into(),vec![Some(20.0),Some(5.0),Some(15.0)])]);
+        let out=rank_history(&dates,&universe,&values).unwrap();
+        assert_eq!(out["rank"]["A"],serde_json::json!([null,1.5,1.0]));
+        assert_eq!(out["rank"]["C"],serde_json::json!([null,3.0,2.0]));
+        assert_eq!(out["unknown_dates"][0]["missing_count"],1);
+        let earlier=values.iter().map(|(s,v)|(s.clone(),v[..2].to_vec())).collect();
+        let first=rank_history(&dates[..2],&universe,&earlier).unwrap();
+        assert_eq!(&out["rank"]["A"].as_array().unwrap()[..2],first["rank"]["A"].as_array().unwrap());
+        assert!(rank_history(&dates,&universe[..2],&values).is_err());
+    }
     fn rows() -> Vec<Observation> {
         [1.0, 3.0, 3.0, 4.0]
             .iter()