CodexBar Issue #139 Performance/Energy Simulation Report

Date: 2026-02-19 Workspace: /Users/michalkrsik/windsurf_project_folder/CodexBar Issue: https://github.com/steipete/codexbar/issues/139

Purpose

Determine which suspected culprit(s) can produce the abnormal CPU/energy behavior reported by users, using short reproducible simulations and process-level sampling.

Host/Tooling

• macOS: Darwin 25.2.0 (arm64)
• Swift: 6.2.3
• Sampling tools: ps, top
• Note: powermetrics was unavailable (requires sudo password in this session), so energy was sampled via top POWER proxy.

Simulated Culprits

• Culprit A: CLI/PTTY-style heavy subprocess churn with polling loop behavior.
• Culprit B: Web dashboard scrape/retry loop with repeated parse work and 400-600ms waits.
• Culprit C: 75ms idle polling loop (blink-style wakeups).
• Combined: A + B + C at once.
• Baseline: near-idle control.

Test Pass 1 (Primary Mechanism Pass)

Artifacts:

• /tmp/codexbar_perf_sim/results_20260219_111607

Summary:

Scenario	Avg CPU	Max CPU	Avg RSS MB	Avg POWER	Avg IDLEW
Baseline	0.00	0.10	0.54	0.00	0.00
Culprit A	113.68	117.40	121.76	0.00	0.00
Culprit B	4.64	13.30	64.15	0.00	5.04
Culprit C	0.25	2.30	33.12	0.00	10.43
Combined	114.62	121.30	217.62	0.00	0.00

Interpretation:

• CPU ranking was clear (A dominates strongly).
• POWER field in this pass was unusable (stuck at 0.00 for several scenarios due top sampling mode).

Test Pass 2 (Calibrated Energy Pass)

Artifacts:

• /tmp/codexbar_perf_sim/energy2_results_20260219_112350

Sampling correction:

• Switched to top -l 2 and parsed the second sample for tracked PIDs to get non-zero POWER values.

Summary:

Scenario	Avg CPU	Max CPU	Avg RSS MB	Avg POWER	Max POWER	Avg IDLEW
Baseline	0.00	0.00	0.55	0.00	0.00	0.00
Culprit A	113.32	115.90	114.73	94.85	150.60	6106.70
Culprit B	4.30	10.10	62.09	2.94	4.20	2.18
Culprit C	0.35	2.60	34.09	0.23	0.60	14.27
Combined	115.67	118.90	218.48	93.29	129.60	3858.60

Validation Against Expected Pattern

Computed checks on pass 2: 10/10 passed.

• A dominates CPU vs B (>=10x): PASS
• A dominates CPU vs C (>=50x): PASS
• A dominates POWER vs B (>=10x): PASS
• A dominates POWER vs C (>=100x): PASS
• Combined close to A CPU (+/-15%): PASS
• Combined close to A POWER (+/-25%): PASS
• C is low CPU (<1%): PASS
• B is moderate CPU (<15%): PASS
• Baseline near zero CPU (<1%): PASS
• Baseline near zero POWER (<1): PASS

Final Finding

Primary root-cause class for the extreme behavior is Culprit A (heavy long-lived CLI/subprocess churn under bad/failure paths).

Secondary:

• Culprit B contributes moderate load.
• Culprit C contributes wakeups/noise but is not a major CPU/energy driver.

Human-level answer: A tiny toolbar app should never keep heavyweight background subprocess/UI loops alive in failure conditions. That behavior is what creates the abnormal battery/CPU footprint.

Limitations

• These were controlled simulations, not a full end-user UI replay of CodexBar.app with all real auth/cookie/account paths.
• powermetrics could not be used in this session due sudo restriction.

Recommended Next Validation (Before Closing Issue)

• Run one short real-app before/after validation after fixes:
  • baseline
  • culprit A-focused repro
  • optional combined
• Capture powermetrics if sudo is available, plus process CPU snapshots.
• Publish before/after table in issue #139.