signals: feature parity with the latest Signals paper. Porting logic from python repo (#903)

* signals: port to layered taxonomy with dual-emit OTel

Made-with: Cursor

* fix: silence collapsible_match clippy lint (rustc 1.95)

Made-with: Cursor

* test: parity harness for rust vs python signals analyzer

Validates the brightstaff signals port against the katanemo/signals Python
reference on lmsys/lmsys-chat-1m. Adds a signals_replay bin emitting python-
compatible JSON, a pyarrow-based driver (bypasses the datasets loader pickle
bug on python 3.14), a 3-tier comparator, and an on-demand workflow_dispatch
CI job.

Made-with: Cursor

* Remove signals test from the gitops flow

* style: format parity harness with black

Made-with: Cursor

* signals: group summary by taxonomy, factor misalignment_ratio

Addresses #903 review feedback from @nehcgs:

- generate_summary() now renders explicit Interaction / Execution /
  Environment headers so the paper taxonomy is visible at a glance,
  even when no signals fired in a given layer. Quality-driving callouts
  (high misalignment rate, looping detected, escalation requested) are
  appended after the layer summary as an alerts tail.

- repair_ratio (legacy taxonomy name) renamed to misalignment_ratio
  and factored into a single InteractionSignals::misalignment_ratio()
  helper so assess_quality and generate_summary share one source of
  truth instead of recomputing the same divide twice.

Two new unit tests pin the layer headers and the (sev N) severity
suffix. Parity with the python reference is preserved at the Tier-A
level (per-type counts + overall_quality); only the human-readable
summary string diverges, which the parity comparator already classifies
as Tier-C.

Made-with: Cursor

Author: syedhashmi

crates/Cargo.lock

@@ -3,6 +3,14 @@ name = "brightstaff"
 version = "0.1.0"
 edition = "2021"
 
+[[bin]]
+name = "brightstaff"
+path = "src/main.rs"
+
+[[bin]]
+name = "signals_replay"
+path = "src/bin/signals_replay.rs"
+
 [dependencies]
 async-openai = "0.30.1"
 async-trait = "0.1"
@@ -26,6 +34,7 @@ opentelemetry-stdout = "0.31"
 opentelemetry_sdk = { version = "0.31", features = ["rt-tokio"] }
 pretty_assertions = "1.4.1"
 rand = "0.9.2"
+regex = "1.10"
 lru = "0.12"
 metrics = "0.23"
 metrics-exporter-prometheus = { version = "0.15", default-features = false, features = ["http-listener"] }

crates/brightstaff/Cargo.toml

@@ -0,0 +1,175 @@
+//! `signals-replay` — batch driver for the `brightstaff` signal analyzer.
+//!
+//! Reads JSONL conversations from stdin (one per line) and emits matching
+//! JSONL reports on stdout, one per input conversation, in the same order.
+//!
+//! Input shape (per line):
+//! ```json
+//! {"id": "convo-42", "messages": [{"from": "human", "value": "..."}, ...]}
+//! ```
+//!
+//! Output shape (per line, success):
+//! ```json
+//! {"id": "convo-42", "report": { ...python-compatible SignalReport dict... }}
+//! ```
+//!
+//! On per-line failure (parse / analyzer error), emits:
+//! ```json
+//! {"id": "convo-42", "error": "..."}
+//! ```
+//!
+//! The output report dict is shaped to match the Python reference's
+//! `SignalReport.to_dict()` byte-for-byte so the parity comparator can do a
+//! direct structural diff.
+
+use std::io::{self, BufRead, BufWriter, Write};
+
+use serde::Deserialize;
+use serde_json::{json, Map, Value};
+
+use brightstaff::signals::{SignalAnalyzer, SignalGroup, SignalReport};
+
+#[derive(Debug, Deserialize)]
+struct InputLine {
+    id: Value,
+    messages: Vec<MessageRow>,
+}
+
+#[derive(Debug, Deserialize)]
+struct MessageRow {
+    #[serde(default)]
+    from: String,
+    #[serde(default)]
+    value: String,
+}
+
+fn main() {
+    let stdin = io::stdin();
+    let stdout = io::stdout();
+    let mut out = BufWriter::new(stdout.lock());
+    let analyzer = SignalAnalyzer::default();
+
+    for line in stdin.lock().lines() {
+        let line = match line {
+            Ok(l) => l,
+            Err(e) => {
+                eprintln!("read error: {e}");
+                std::process::exit(1);
+            }
+        };
+        let trimmed = line.trim();
+        if trimmed.is_empty() {
+            continue;
+        }
+        let result = process_line(&analyzer, trimmed);
+        // Always emit one line per input line so id ordering stays aligned.
+        if let Err(e) = writeln!(out, "{result}") {
+            eprintln!("write error: {e}");
+            std::process::exit(1);
+        }
+        // Flush periodically isn't strictly needed — BufWriter handles it,
+        // and the parent process reads the whole stream when we're done.
+    }
+    let _ = out.flush();
+}
+
+fn process_line(analyzer: &SignalAnalyzer, line: &str) -> Value {
+    let parsed: InputLine = match serde_json::from_str(line) {
+        Ok(p) => p,
+        Err(e) => {
+            return json!({
+                "id": Value::Null,
+                "error": format!("input parse: {e}"),
+            });
+        }
+    };
+
+    let id = parsed.id.clone();
+
+    let view: Vec<brightstaff::signals::analyzer::ShareGptMessage<'_>> = parsed
+        .messages
+        .iter()
+        .map(|m| brightstaff::signals::analyzer::ShareGptMessage {
+            from: m.from.as_str(),
+            value: m.value.as_str(),
+        })
+        .collect();
+
+    let report = analyzer.analyze_sharegpt(&view);
+    let report_dict = report_to_python_dict(&report);
+    json!({
+        "id": id,
+        "report": report_dict,
+    })
+}
+
+/// Convert a `SignalReport` into the Python reference's `to_dict()` shape.
+///
+/// Ordering of category keys in each layer dict follows the Python source
+/// exactly so even string-equality comparisons behave deterministically.
+fn report_to_python_dict(r: &SignalReport) -> Value {
+    let mut interaction = Map::new();
+    interaction.insert(
+        "misalignment".to_string(),
+        signal_group_to_python(&r.interaction.misalignment),
+    );
+    interaction.insert(
+        "stagnation".to_string(),
+        signal_group_to_python(&r.interaction.stagnation),
+    );
+    interaction.insert(
+        "disengagement".to_string(),
+        signal_group_to_python(&r.interaction.disengagement),
+    );
+    interaction.insert(
+        "satisfaction".to_string(),
+        signal_group_to_python(&r.interaction.satisfaction),
+    );
+
+    let mut execution = Map::new();
+    execution.insert(
+        "failure".to_string(),
+        signal_group_to_python(&r.execution.failure),
+    );
+    execution.insert(
+        "loops".to_string(),
+        signal_group_to_python(&r.execution.loops),
+    );
+
+    let mut environment = Map::new();
+    environment.insert(
+        "exhaustion".to_string(),
+        signal_group_to_python(&r.environment.exhaustion),
+    );
+
+    json!({
+        "interaction_signals": Value::Object(interaction),
+        "execution_signals": Value::Object(execution),
+        "environment_signals": Value::Object(environment),
+        "overall_quality": r.overall_quality.as_str(),
+        "summary": r.summary,
+    })
+}
+
+fn signal_group_to_python(g: &SignalGroup) -> Value {
+    let signals: Vec<Value> = g
+        .signals
+        .iter()
+        .map(|s| {
+            json!({
+                "signal_type": s.signal_type.as_str(),
+                "message_index": s.message_index,
+                "snippet": s.snippet,
+                "confidence": s.confidence,
+                "metadata": s.metadata,
+            })
+        })
+        .collect();
+
+    json!({
+        "category": g.category,
+        "count": g.count,
+        "severity": g.severity,
+        "signals": signals,
+    })
+}

crates/brightstaff/src/bin/signals_replay.rs

@@ -441,10 +441,8 @@ impl ArchFunctionHandler {
                 }
             }
             // Handle str/string conversions
-            "str" | "string" => {
-                if !value.is_string() {
-                    return Ok(json!(value.to_string()));
-                }
+            "str" | "string" if !value.is_string() => {
+                return Ok(json!(value.to_string()));
             }
             _ => {}
         }