bug(sona): getStats().trajectoriesRecorded always returns 0

[dmoellenbeck]

Describe the bug

SonaEngine.getStats() reports trajectoriesRecorded: 0 even after successfully recording and processing trajectories via beginTrajectory → addTrajectoryStep → endTrajectory.

Environment

• @ruvector/sona: v0.1.5
• Node.js: v22.x (NAPI bindings)
• Platform: macOS Apple Silicon (darwin-arm64)

To Reproduce

const { SonaEngine } = require('@ruvector/sona');
const engine = new SonaEngine(256);

// Record 120 trajectories
for (let i = 0; i < 120; i++) {
  const tid = engine.beginTrajectory(new Array(256).fill(Math.random()));
  engine.addTrajectoryStep(tid, new Array(256).fill(0.5), new Array(256).fill(0.1), 0.8);
  engine.endTrajectory(tid, 0.9);
}

console.log(engine.getStats());
// CoordinatorStats { trajectories_buffered: 120, ... }
// BUT: trajectoriesRecorded is always 0 in the JSON stats

const result = engine.forceLearn();
console.log(result);
// "Forced learning: 120 trajectories -> 0 patterns, status: completed"
// ^^^ forceLearn WORKS — confirms trajectories are buffered and processed

console.log(engine.getStats());
// trajectories_buffered: 0 (cleared after forceLearn)
// trajectoriesRecorded: still 0

Expected behavior

getStats() should report the actual number of trajectories recorded (120 in this case), matching trajectories_buffered before processing.

Actual behavior

trajectoriesRecorded is always 0, even though:

• trajectories_buffered correctly shows 120
• forceLearn() successfully processes them (status: completed)

The counter appears to never be incremented in the Rust code when endTrajectory is called.

Additional context

Using SONA via ruvector MCP server (v0.2.16) for a coaching application. The IntelligenceEngine wrapper also shows 0 because it reads from getStats(). This makes it impossible to monitor SONA learning progress from the outside.

[ruvnet]

Fixed in PR #284 (merged).

Root cause: Rust CoordinatorStats serialized as trajectories_buffered (snake_case) but TypeScript SonaStats expected trajectoriesRecorded (camelCase). The JSON field name mismatch meant the value was always undefined → 0.

Fix:

• Added trajectories_recorded field to CoordinatorStats = trajectories_buffered + trajectories_dropped
• Added patterns_learned alias for patterns_stored
• Updated TypeScript getStats() to map snake_case → camelCase with fallbacks:

trajectoriesRecorded: raw.trajectories_recorded ?? raw.trajectoriesRecorded ?? (raw.trajectories_buffered ?? 0) + (raw.trajectories_dropped ?? 0)

Files: crates/sona/src/loops/coordinator.rs, npm/packages/ruvector/src/core/sona-wrapper.ts

[dmoellenbeck]

Partial confirmation

The native Rust SONA engine correctly tracks trajectories at the binary level
(trajectories_buffered increments properly). However, getStats() still
crashes in ruvector@0.2.18 because the @ruvector/sona native binary is
still on v0.1.5 — it returns Rust Debug format instead of JSON, causing
JSON.parse() to fail in the wrapper.

Filed #301 for the binary version mismatch. Once @ruvector/sona is rebuilt
with the current napi.rs code (which uses serde_json::to_string), this
fix should work end-to-end.

Our workaround: sona-train.mjs uses forceLearn() directly and never calls
getStats(), so SONA training works correctly despite this issue.