Dual-Stream Checking: Expected vs Observed Correlation

Correlation model

Dual-stream checking compares predicted intent with monitor-observed truth. Correlate via transaction IDs and bounded timing windows.

[ENV][UVM][CHECK] dual-stream pipeline

expected stream:
  predictor/model -> exp_q[id]

observed stream:
  monitor -> obs_q[id]

correlator:
  if id in both queues -> compare
  if timeout on one side -> missing counterpart error

task correlate_loop();
  forever begin
    if (exp_q.exists(id) && obs_q.exists(id)) begin
      compare_pair(exp_q[id], obs_q[id]);
      exp_q.delete(id);
      obs_q.delete(id);
    end
    check_timeouts();
    @(posedge vif.clk);
  end
endtask

• IDs should be protocol-meaningful and stable across transformations.

• Timeout policy must be explicit and reviewed.

• Diagnostics should show both expected and observed snapshots.

Key takeaways

• Reliable correlation is the heart of scoreboard trust.

• Timeouts are correctness logic, not optional debug aids.

Common pitfalls

• Comparing in arrival order when protocol allows reordering.

• Using weak IDs that collide under stress traffic.

Applied Patterns

Keep dual-stream state machine explicit so out-of-order and retry flows remain debuggable.

[CHECK] correlation states

IDLE
  -> WAIT_EXP / WAIT_OBS
  -> BOTH_READY
  -> COMPARE
  -> DONE or MISMATCH

retry path:
  mark provisional
  re-correlate on final response

function void compare_pair(txn exp, txn obs);
  compare_hits++;
  if (!exp.compare(obs)) begin
    mismatch_hits++;
    `uvm_error("CMP",
      $sformatf("id=%0d exp=%s obs=%s", exp.id, exp.sprint(), obs.sprint()))
  end
endfunction

• State labels make triage simpler than ad-hoc queue logic.

• Print ID and key fields on every mismatch.

• Retry handling should be visible, not implicit.

Integration Drilldown

Inject reordered responses and confirm correlation remains stable without false mismatches.

[CHECK] reorder validation

sequence:
  issue IDs 10,11,12
  observe responses 11,10,12

expect:
  compare_hits == 3
  mismatch_hits == 0
  timeout_hits == 0

Key takeaways

• Reorder validation proves correlation correctness under realistic behavior.

Common pitfalls

• Assuming in-order behavior in protocols that permit reordering.