sota-management

name	sota-management
description	4-level SOTA performance tracking for VibeCodeHPC. Use when checking or updating best performance records.

SOTA Management

Hierarchy

Level	Scope	File	Manager
Local	PG directory best	sota_local.txt	PG
Family	Same middleware parent-child	Virtual (from PG_visible_dir.md)	Auto
Hardware	Same hardware config	sota_hardware.txt	SE
Project	Project-wide best	sota_project.txt	PM

Check Timing

• After PG records test performance in ChangeLog.md
• During SE/PM statistical analysis

Python API

${CLAUDE_SKILL_DIR}/scripts/sota_checker.py — checks and updates all 4 levels. Run with --help for usage.

Benefits

• Fast comparison via dedicated files (no ChangeLog scanning)
• Hierarchical isolation
• Family SOTA auto-computed from visibility scope

Details: references/file_formats.md, references/sota_api.md

CLI Script

python3 ${CLAUDE_SKILL_DIR}/scripts/sota_checker.py --help
python3 ${CLAUDE_SKILL_DIR}/scripts/sota_checker.py /path/to/PG1.1 check "350.0 GFLOPS"
python3 ${CLAUDE_SKILL_DIR}/scripts/sota_checker.py /path/to/PG1.1 update "350.0 GFLOPS" --version 1.2.0 --agent PG1.1
python3 ${CLAUDE_SKILL_DIR}/scripts/sota_checker.py /path/to/PG1.1 info --json

Visualizer Script

scripts/sota_visualizer.py — self-contained SOTA visualization pipeline. Generates 4-level graphs (local/family/hardware/project) from ChangeLog.md data.

Requires: matplotlib, numpy

# Pipeline mode (default, periodic execution)
python3 ${CLAUDE_SKILL_DIR}/scripts/sota_visualizer.py

# Debug mode (low DPI)
python3 ${CLAUDE_SKILL_DIR}/scripts/sota_visualizer.py --debug

# Summary only (no graph generation)
python3 ${CLAUDE_SKILL_DIR}/scripts/sota_visualizer.py --summary

# Data export for multi-project analysis
python3 ${CLAUDE_SKILL_DIR}/scripts/sota_visualizer.py --export

# Custom levels and DPI
python3 ${CLAUDE_SKILL_DIR}/scripts/sota_visualizer.py --levels local,project --dpi 80