v3.0.3 — R5 Performance + R6 Cross-Module Contract Fixes

What's New in v3.0.3

R5 — Performance Optimization (bit-identical, zero numerical change)

• magfit: New ned_to_body_batch() batch quaternion rotation + np.bincount bin weights vectorization
• AP step_response_fft: np.full scale + np.cumsum SNR gaussian vectorization

R6 — Cross-Module Contract Bug Fixes

• 🔴 A2 generic-key injection (AP/BF): parse() now injects generic keys (pid.roll.p etc.) into params. Previously, PIDReviewer._get_current_pid always returned 0.0 because it queried generic keys while only native names were stored — all PID parameter suggestions were silently discarded.
• 🟡 BF pid_bounds: Added Betaflight integer-scale bounds (P[10,150]/I[20,220]/D[0,100]/FF[0,300]). Default AP-scale bounds (0.01~0.5) would clamp BF 45×1.1 down to 0.5.
• 🟡 thresholds shape guard: New _is_axis_threshold_shape() validates knowledge-base thresholds are in the consumer-expected shape. Previously, all pid_rules.json thresholds were dead config — the analyzer always used hardcoded defaults.
• New tests/test_platform_contracts.py (8 tests): A2 injection, thresholds guard, BF bounds, BF gain not clipped.

167 tests passed

Full Changelog: v3.0.2...v3.0.3

SmartTune v3.0.2 — PX4 ULog + A1 Convergence + Label Unification

Changes since v3.0.0

v3.0.1 — 14 Architecture Review Fixes (2026-06-11)

14 critical/high severity fixes from Claude architecture review:

• C1: magfit 100% KeyError crash fixed (was silently skipped)
• C2: mGauss unit normalization across magfit pipeline
• C3: INS_HNTCH_REF semantic fix — prevents incorrect dynamic notch params
• C4: Skip current_val ≤ 0 recommendations (removes false "0.0 → 0.01" suggestions)
• C5/C6: ODI symmetric soft-iron matrix + PitchIn guard
• C7/C8: imu_dict actual construction + bidirectional overshoot + settling=-1 sentinel
• C10: step_response_fft aligned with WebTools/PID Toolbox reference implementations
• C12/C14/A3/A4: frequency band merge, ARX fallback markers, registry metadata snapshot, TYPE_CHECKING

v3.0.2 — PX4 ULog + A1 Convergence + Label Unification (2026-06-12)

PX4 Full Implementation (first time complete):

• ULog parser based on pyulog — all uORB topics mapped (vehicle_rates_setpoint, sensor_combined, ekf2_innovations, etc.)
• Param A2 dual-write contract
• sensor_combined fallback path for pre-v1.13 firmware
• PX4 knowledge base (pid_rules.json, filter_rules.json)
• validate_px4_ulog.py — one-command end-to-end validation
• `stune quality/pid/fft/sysid/analyze -i .ulg` all functional

A1 Convergence (services layer unification):

• `run_module()` unified core — CLI and services no longer have duplicate analyzer wiring
• B2 fix: fft_result now properly assigned to FullAnalysisResult
• sysid --na/--nb now actually passes through (was broken)
• filter and quality commands call services layer (D1 fix)
• t_synthetic_regression.py: 11 case golden-standard synthetic tests

Label Unification:

• vibration_level: SEVERE→POOR, CRITICAL→UNUSABLE (consistent Assessment enum)
• vibration_level_raw preserves original labels
• formatter/httm_report color mapping compatible with both old and new labels

v6.2 Synthetic Regression Fix:

• Excitation signals add 8% broadband noise (improves Wiener deconvolution conditioning)
• Steady-state tolerances widened; underdamped test uses relative assertion
• formatter `_has_vec()` guard for numpy array truthiness

Test Results

`159 passed in 2.16s` — AP/BF/PX4 real log verified

Files Changed

v3.0.1: 12 files, +743/-492
v3.0.2: 13 files, +1181/-510

SmartTune CLI v3.0.0 — Firmware-Parameter-Aware Tuning

SmartTune CLI v3.0.0 — Firmware-Parameter-Aware Tuning

New: stune params — Firmware Parameter Validation

SmartTune CLI can now query and validate firmware parameters before recommending changes, preventing agents from suggesting parameters that don't exist in the target firmware.

stune params ap                           # List 2,574 ArduPilot parameters
stune params bf                           # List 45 Betaflight parameters
stune params ATC_RAT_RLL_P                # Show parameter details
stune params --search notch                # Search across all platforms
stune params --validate ATC_RAT_RLL_P 0.15 -p ardupilot  # Validate before recommending

New: MCP Parameter Tools (13 total)

3 new MCP tools for parameter query and validation:

• smarttune_list_params(platform, category) — List all firmware parameters
• smarttune_search_params(keyword, platform) — Search by keyword
• smarttune_validate_param(name, value, platform) — Validate before recommending

MCP agents can now autonomously verify parameter existence and value ranges.

Knowledge Base Architecture

Parameter tables live in smarttune/knowledge/params/*.json — data scraped from official firmware repositories:

• ArduPilot: 2,574 parameters from @Param annotations (100% described, 1,941 with valid ranges)
• Betaflight: 45 parameters from settings.c (BF 4.5+ names: p_roll, gyro_lpf1_static_hz, d_max_roll)
• PX4: 20 parameters from official YAML definitions

Breaking Change

Betaflight adapter's map_param_to_platform() now returns BF 4.5+ parameter names:

• pid_roll_p → p_roll
• gyro_lowpass_hz → gyro_lpf1_static_hz
• d_min_roll → d_max_roll

Legacy name support maintained for reading older BBL logs.

Consistency Fixes

• ParamTable.validate() fixed for parameters without min/max range
• BF filter_transfer.py supports both old and new parameter names
• All 35 previously-undescribed AP parameters now have verified descriptions and ranges
• Count metadata corrected in knowledge base JSONs

Full Changelog

Previous v2.4.2 → v3.0.0: [compare]

SmartTune CLI v2.3.1

Patch release adding a dedicated MCP-only SmartTune skill for read-only customer-support agents.\n\n- Adds skill-mcp/SKILL.md for agents without exec/shell/write permission\n- Documents SmartTune MCP tool workflow and safety boundaries\n- Keeps v2.3.0 MCP server/runtime behavior unchanged\n\nVerification: pytest -q -> 141 passed

SmartTune CLI v2.3.0

SmartTune CLI v2.3.0

Added

• Read-only MCP server via smarttune-mcp for OpenClaw and other MCP clients.
• smarttune_list_platforms, smarttune_log_quality, and smarttune_analyze_log MCP tools.
• Pure service layer under smarttune/services/ for analysis and JSON-safe serialization.
• MCP security tests for extension allowlists, path traversal, symlink escape, and file size limits.

Security

• MCP tools do not expose subprocess, shell execution, arbitrary output paths, parameter writes, or filesystem mutation.
• Path validation resolves symlinks, enforces allowed roots, rejects unsafe extensions, and respects SMARTTUNE_MCP_MAX_FILE_MB.
• Betaflight .txt remains CLI-supported but is intentionally excluded from the MCP allowlist.

Fixed

• Fixed MagFit import indentation.
• Hardware analysis service uses the existing generate_hardware_report() path.

Verification

• pytest -q -> 141 passed
• MCP smoke-tested against a real ArduPilot .BIN log.

v2.2.0 — English Docs & CLI Help Improvements

What's Changed

• 🌐 All documentation converted to English (SKILL.md, ROADMAP.md, MIGRATION.md)
• 📖 Complete --help descriptions for all subcommands (pid, fft, sysid, hardware, magfit)
• 🤖 SKILL.md for OpenClaw agent integration
• 🧹 Deprecated ardupilot-tune skill/CLI replaced by stune

Platform Support

Platform	Format	Status
ArduPilot	.bin / .log	✅ Full
Betaflight	.bbl / .bfl	✅ Full (v2.0)
PX4	.ulg	🔲 Partial (v2.1)

Full changelog

9a760a0 chore: sync version to 2.2.0 (pyproject.toml + __init__.py)
9e26108 feat(cli): add complete --help descriptions for all subcommands
6303c3d docs: convert all docs and skill to English (SKILL.md, ROADMAP.md, MIGRATION.md)
32ace2d feat(skill): add SmartTune SKILL.md for OpenClaw agent - replace deprecated ardupilot-tune skill
49be79e docs(readme): convert Agent Analysis Report to Markdown tables
dfb6d81 docs(readme): switch example images to dark theme
cce6f8f fix(plots): subtle light theme - gray reference lines and invisible borders
d3063ab feat(plots): customize dark theme colors
0ffc943 feat(cli): add --theme light|dark option for plot generation
b416ef9 docs(readme): add agent analysis report example section
371c736 revert(assets): restore original light-themed APM example image
ba576db revert(plots): restore original light theme, remove all dark_background attempts
3250841 refactor(plots): use matplotlib dark_background style instead of manual rcParams
d719542 fix(plots): compact layout for GitHub README (formatter changes gitignored)
73024c5 docs(readme): dark theme plots, move examples to end section
349659d docs(readme): add PID step response example images for ArduPilot and Betaflight
dec397d docs(readme): update for v2.1.0

v2.1.0 — Platform-Specific Analyzers

Full Changelog: v2.0.2...v2.1.0

v2.0.2: BBL Parser Critical Fixes

BBL Parser Fixes

Fixed 8 critical bugs in Betaflight Blackbox log parser:

Bug	Impact
TAG8_4S16 nibble packing	Frame sync loss → all fields corrupted
TAG2_3S32 tag bit extraction	axisI decoded as zero
NEG_14BIT sign extension	vbatLatest and fields wrong
P-frame predictor history	Difference decoding errors
loopIteration segment detection	Parser stopped after ~3K frames per segment
0x48 byte false header	Parser hangs on valid data
P interval sample rate	Duration wrong (41s→165s)
Quality filter threshold	Empty charts (rad/s↔deg/s mismatch)

Results

• BLL-PD.bbl: 1s → 91s parsed, 158K PID samples/axis
• H5_51377_6S_F722.bbl: 1s → 166s parsed, quality 45→75 (POOR→GOOD)

Credits

• BBL parser loopIteration fix: internal
• Remaining 7 fixes: Claude Code via ACP session

Full Changelog: v2.0.1...v2.0.2

v2.0.1 — PID unit contract fix + FFT threshold + CLI docs

🐛 Bug Fixes

ArduPilot 单位契约修复

• rad/s → deg/s 自动转换：PIDR/ATC_RAT_* 消息 Tar 字段自动检测单位（max < 6.5 判定为 rad/s）
• PIDR 字段兼容：支持 Tar（新固件）和 Des（旧固件）
• modern/legacy 消息源正确分离：修复旧版 PIDR+RATE 混入 bug

FFT 阶跃响应精度修复

• min_target_amplitude 阈值从 0.349（rad/s）改为 3.0（deg/s）
• Roll 峰值从 0.52 恢复到 0.64（与旧版 0.65 一致）

CLI --help 文档完善

• 版本号同步至 2.0.1
• Betaflight 状态: Planned → Full support
• PX4 状态: Planned → Partial support
• 7 个子命令 --platform/--visual 选项添加帮助文本

包含 Commits

Commit	说明
3a3ae08	ArduPilot PID 单位契约修复 + PIDR 兼容 + CLI 增强
86fde58	FFT 阈值同步 deg/s
c2988ef	版本号 + 平台状态 + 子命令帮助文本完善

v2.0.0 — Multi-platform release

SmartTune CLI v2.0.0

Multi-platform flight log analysis & tuning advisor CLI — ArduPilot + Betaflight + PX4

What's New

• Multi-platform architecture — Unified FlightData representation across ArduPilot, Betaflight, and PX4
• Betaflight support — Pure Python BBL parser, FF/RPM/D-term noise analyzers
• Platform auto-detection — stune analyze detects platform from log headers
• ParamRef system — Generic parameter names mapped to each platform
• 6-layer Knowledge Base — common → platform → user → Pro layers
• Error code system — Standardized E10xx-E50xx error classification

Quick Install

pip install smarttune
stune analyze -i flight-log.BIN

96 tests · 3 platforms · fully offline