ruvnet/ruflo · error · Error

Trajectory buffer not initialized

Error message

Trajectory buffer not initialized

What it means

Thrown by recordTrajectory() when the module-level trajectoryBuffer is null. initializeTraining() always constructs a trajectory buffer on success (WASM WasmTrajectoryBuffer or the JsTrajectoryBuffer fallback), so a null buffer means initializeTraining() never completed in this process — or cleanup() freed and nulled it. Capacity defaults to 10000 entries and dim to 256 (max).

Source

Thrown at v3/@claude-flow/cli/src/services/ruvector-training.ts:545

  } else if (microLoRA) {
    microLoRA.adapt_with_reward(improvement);
  }

  totalAdaptations++;
}

/**
 * Record a learning trajectory
 */
export function recordTrajectory(
  embedding: Float32Array,
  operatorType: number,
  attentionType: number,
  executionMs: number,
  baselineMs: number
): void {
  if (!trajectoryBuffer) {
    throw new Error('Trajectory buffer not initialized');
  }

  trajectoryBuffer.record(
    embedding,
    operatorType,
    attentionType,
    executionMs,
    baselineMs
  );
}

/**
 * Get trajectory statistics
 */
export function getTrajectoryStats(): {
  successRate: number;
  meanImprovement: number;
  bestImprovement: number;

View on GitHub (pinned to fa13ee4ad6)

Solutions

1. Await initializeTraining() (optionally with trajectoryCapacity sized to your run) before arming any trajectory-recording hooks.
2. Gate recordTrajectory call sites behind the same ensureInitialized() promise used for training.
3. After cleanup(), either stop recording or re-initialize first.
4. Size trajectoryCapacity to your step count if you rely on the stats — recording beyond capacity evicts old entries, but that never causes this error.

Example fix

// before
instrumentation.on('step', (ev) => {
  recordTrajectory(ev.embedding, ev.op, ev.attn, ev.ms, ev.base); // pre-init → throws
});

// after
const ready = initializeTraining({ trajectoryCapacity: 50_000 });
instrumentation.on('step', async (ev) => {
  await ready;
  recordTrajectory(ev.embedding, ev.op, ev.attn, ev.ms, ev.base);
});

const ready = initializeTraining({ trajectoryCapacity: 50_000 }); // awaited later, started once
instrumentation.on('step', async (ev) => {
  await ready; // guarantees trajectoryBuffer exists (WASM or JS fallback)
  recordTrajectory(ev.embedding, ev.op, ev.attn, ev.executionMs, ev.baselineMs);
});

try {
  recordTrajectory(emb, op, attn, execMs, baseMs);
} catch (e) {
  if (e instanceof Error && e.message === 'Trajectory buffer not initialized') {
    await initializeTraining();
    recordTrajectory(emb, op, attn, execMs, baseMs);
    return;
  }
  throw e;
}

When it happens

Trigger: Recording execution-vs-baseline trajectories before awaiting initializeTraining(); calling recordTrajectory after cleanup() in test teardown; trajectory hooks registered in a worker process whose parent alone ran init.

Common situations: Instrumentation installed at module load (decorators, monkey-patched timers) that fires before async init resolves; long-running daemons that cleanup() on SIGHUP then keep serving trajectory events; benchmark harnesses that assume state persists from a previous run.

Related errors

• Training system not initialized
• SSRF guard: invalid URL — ${rawUrl}
• ruvLLM bridge not initialized. Call with config first.
• Flash attention not initialized
• MoE attention not initialized

AI-assisted analysis of ruvnet/ruflo@fa13ee4ad6 (2026-08-18). Data as JSON: /api/errors/d398922f132de2c9. Report an issue: GitHub.