bug: SonaEngine.forceLearn() always returns 'insufficient trajectories' after completed trajectories

[PogeystickJoe]

Summary

SonaEngine.forceLearn() in ruvector 0.2.12 always returns "skipped: insufficient trajectories" even after completing multiple trajectories with beginTrajectory → addStep (x3) → endTrajectory in the same session. The trajectories complete without error, but forceLearn() reports 0 trajectories available.

Environment

• ruvector: 0.2.12
• Node.js: 22.17.1
• Platform: Windows (MINGW64), win32-x64

Steps to Reproduce

const { SonaEngine } = require('ruvector');
const sona = new SonaEngine(384);

// Create proper 384-dim vectors
const mkEmb = (seed) => new Array(384).fill(0).map((_, i) => Math.sin(i * 0.1 + seed) * 0.1);
const attn = new Array(384).fill(0).map(() => Math.random() * 0.05);

// Complete 5 distinct trajectories
for (let t = 0; t < 5; t++) {
  const tid = sona.beginTrajectory(mkEmb(t));     // Returns 0, 1, 2, 3, 4
  sona.addStep(tid, mkEmb(t+0.1), attn, 0.85);   // No error
  sona.addStep(tid, mkEmb(t+0.2), attn, 0.70);   // No error
  sona.addStep(tid, mkEmb(t+0.3), attn, 0.60);   // No error
  sona.endTrajectory(tid, 0.75 + t * 0.02);       // No error
}

const patterns = sona.forceLearn();
console.log(patterns);
// Output: "Forced learning: 0 trajectories -> 0 patterns, status: skipped: insufficient trajectories"

Expected Behavior

After 5 completed trajectories with 3 steps each (15 total steps), forceLearn() should process them and extract patterns. Or at minimum, report 5 trajectories instead of 0 trajectories.

Actual Behavior

• beginTrajectory() returns incrementing IDs (0-4) — suggests they're being created
• addStep() and endTrajectory() return without error
• forceLearn() reports 0 trajectories as if none were stored
• The message says "insufficient trajectories" but doesn't specify the minimum threshold

Questions

1. Is there a minimum number of trajectories required? If so, what is it?
2. Does endTrajectory() actually persist the trajectory, or does it get dropped?
3. Is there a required minimum diversity threshold between trajectory embeddings?
4. Does the SONA coordinator buffer need to be flushed before forceLearn()?

Additional Context

The SONA getStats() method (see #257) shows trajectories_buffered: 0 after all 5 trajectories complete, which suggests trajectories may be getting dropped during endTrajectory() rather than buffered for learning.

The buffer_success_rate: 1.0 is also suspicious — if 0 trajectories were buffered but success rate is 100%, it may indicate the buffer was never attempted rather than trajectories failing.

Workaround

Currently using RuVector MCP hooks_batch_learn and hooks_trajectory_begin/step/end as the learning pathway instead of the native SONA engine. The MCP hooks do process and store learning data (confirmed 36 updates via batch_learn), but SONA pattern extraction specifically does not work.