CHANGELOG.md

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[0.2.32] - 2026-04-04

Added

• Added scripts/package_refresh_bundle.py to package a completed .ssc-inclusive refresh work directory into a cleaner deployment bundle layout.
• Added docs/REFRESH_DEPLOYMENT_AND_ARROWVORTEX_VERIFICATION_2026-04-04.md documenting runtime compatibility, bundle-export strategy, and ArrowVortex-oriented usage for the refreshed artifacts.

Changed

• Documented the next post-refresh milestone for the completed refresh: verified AutoChart initialization against data/ssc_refresh_work/models + data/ssc_refresh_work/ffr_models, validated bundle packaging in dry-run mode, and updated usage docs to reference the refreshed local artifact paths.

[0.2.31] - 2026-04-04

Added

• Added docs/TRAINING_REFRESH_COMPLETION_2026-04-04.md capturing the first full completion-state inventory for the .ssc-inclusive refresh run.

Changed

• Documented the key completion milestone for the active refresh: all practical single/double placement buckets now show complete checkpoint sets, and refreshed floating-point FFR artifacts were produced for both dance-single and dance-double.

[0.2.30] - 2026-04-04

Added

• Added docs/TRAINING_REFRESH_PROGRESS_15_2026-04-04.md capturing continued double-mode progress with dance-double_Easy reaching at least its ninth checkpoint.

Changed

• Documented the next monitoring milestone for the active refresh: dance-double_Easy advanced to at least model_09.pth, the active log progressed into Epoch 10/10 for the active double-mode bucket, and the refresh remained busy with multiple Python processes visible.

[0.2.29] - 2026-04-04

Added

• Added docs/TRAINING_REFRESH_PROGRESS_14_2026-04-04.md capturing the refresh advancing beyond all practical single-mode buckets and into double-mode practical training.

Changed

• Documented the next major monitoring milestone for the active refresh: all practical single-mode buckets appear complete (dance-single_Easy, Medium, Hard, Challenge at 10 checkpoints each) and dance-double_Easy became the active frontier at at least model_08.pth.

[0.2.28] - 2026-04-04

Added

• Added docs/TRAINING_REFRESH_PROGRESS_13_2026-04-04.md capturing the active refresh advancing into Epoch 10/10 during the second practical bucket.

Changed

• Documented the next monitoring milestone for the active refresh: dance-single_Medium advanced to at least model_09.pth, the active log progressed into Epoch 10/10, and two active Python processes were visible while the run continued without interruption.

[0.2.27] - 2026-04-04

Added

• Added docs/TRAINING_REFRESH_PROGRESS_12_2026-04-04.md capturing the active refresh crossing from Epoch 8/10 into Epoch 9/10 during the second practical bucket.

Changed

• Documented the next monitoring milestone for the active refresh: dance-single_Medium advanced to at least model_08.pth, Epoch 8/10 completed with validation output, and the active log progressed into Epoch 9/10.

[0.2.26] - 2026-04-04

Added

• Added docs/TRAINING_REFRESH_PROGRESS_11_2026-04-04.md capturing continued in-flight progress deeper into the active Epoch 8/10 window of the second practical bucket.

Changed

• Documented the next monitoring milestone for the active refresh: the run remained alive, dance-single_Medium remained the active artifact frontier at model_07.pth, and the log progressed substantially deeper into the late portion of Epoch 8/10.

[0.2.25] - 2026-04-04

Added

• Added docs/TRAINING_REFRESH_PROGRESS_10_2026-04-04.md capturing continued in-flight progress deeper into the second practical bucket of the .ssc-inclusive refresh run.

Changed

• Documented the next monitoring milestone for the active refresh: dance-single_Medium advanced to at least model_07.pth and the active log progressed into Epoch 8/10.

[0.2.24] - 2026-04-04

Added

• Added docs/TRAINING_REFRESH_PROGRESS_9_2026-04-04.md capturing continued in-flight progress after the refresh moved into the second practical bucket.

Changed

• Documented the next monitoring milestone for the active refresh: the run remained alive, the active log progressed substantially deeper into Epoch 7/10, and artifact recency still indicated dance-single_Medium as the current active practical bucket frontier.

[0.2.23] - 2026-04-04

Added

• Added docs/TRAINING_REFRESH_PROGRESS_8_2026-04-04.md capturing the first practical bucket appearing complete and the second practical bucket becoming active during the .ssc-inclusive refresh run.

Changed

• Documented the next major in-flight milestone for the active refresh: dance-single_Easy reached a full observed 10-checkpoint set, dance-single_Medium appeared and advanced to at least model_06.pth, and later-epoch (7/10) log output was observed for the active bucket.

[0.2.22] - 2026-04-04

Added

• Added docs/TRAINING_REFRESH_PROGRESS_7_2026-04-04.md capturing continued in-flight progression of the first practical placement bucket during the active .ssc-inclusive refresh run.

Changed

• Documented the next in-flight milestone for the active refresh: dance-single_Easy advanced to at least model_05.pth and later-epoch (6/10) log output was observed.

[0.2.21] - 2026-04-04

Added

• Added docs/TRAINING_REFRESH_PROGRESS_6_2026-04-04.md capturing continued in-flight progression of the first practical placement bucket during the active .ssc-inclusive refresh run.

Changed

• Documented the next in-flight milestone for the active refresh: dance-single_Easy advanced to at least model_04.pth and later-epoch (5/10) log output was observed.

[0.2.20] - 2026-04-04

Added

• Added docs/TRAINING_REFRESH_PROGRESS_5_2026-04-04.md capturing continued in-flight progression of the first practical placement bucket during the active .ssc-inclusive refresh run.

Changed

• Documented the next in-flight milestone for the active refresh: dance-single_Easy advanced to at least model_03.pth and later-epoch (4/10) log output was observed.

[0.2.19] - 2026-04-04

Added

• Added docs/TRAINING_REFRESH_PROGRESS_4_2026-04-04.md capturing continued checkpoint accumulation inside the first practical SymNet bucket of the active .ssc-inclusive refresh run.

Changed

• Documented the next in-flight milestone for the active refresh: dance-single_Easy advanced from 1 to 2 observed checkpoints and later-epoch log output was observed for that bucket.

[0.2.18] - 2026-04-04

Added

• Added docs/TRAINING_REFRESH_PROGRESS_3_2026-04-04.md capturing the first observed practical SymNet bucket checkpoint during the active .ssc-inclusive refresh run.

Changed

• Documented the next in-flight milestone for the active refresh: dance-single_Easy/model_01.pth was observed, confirming that practical bucket-model regeneration is underway.

[0.2.17] - 2026-04-04

Added

• Added docs/TRAINING_REFRESH_PROGRESS_2_2026-04-04.md capturing a later in-flight progress snapshot for the active .ssc-inclusive refresh run.

Changed

• Documented that the active refresh progressed from onset training into the first practical SymNet bucket stage, with a full 5-checkpoint onset set observed and dance-single_Easy/ created for the next phase.

[0.2.16] - 2026-04-04

Added

• Added docs/TRAINING_REFRESH_PROGRESS_2026-04-04.md documenting an in-flight progress snapshot for the active .ssc-inclusive refresh run.

Changed

• Recorded the first observed runtime milestone for the active refresh: onset training had advanced into epoch 4/5 and produced initial checkpoints under data/ssc_refresh_work/models/onset/.

[0.2.15] - 2026-04-04

Added

• Added docs/TRAINING_REFRESH_LAUNCH_2026-04-04.md documenting the actual launch of the .ssc-inclusive refresh training run.

Changed

• Kicked off the resume-friendly .ssc-inclusive training refresh against data/ssc_refresh_work and documented its initial runtime state/log path.

[0.2.14] - 2026-04-04

Added

• Added scripts/audit_refresh_readiness.py and docs/SSC_REFRESH_READINESS_2026-04-04.md to document the exact state of the .ssc-inclusive refresh work directory.

Changed

• Added resume-friendly skip flags to scripts/train_all.py so the .ssc-inclusive refresh can be resumed without redoing completed prep/feature steps or retraining buckets that already have artifacts.

[0.2.13] - 2026-04-04

Added

• Added docs/LEGACY_SUBTREE_QUARANTINE_2026-04-04.md documenting the decision to treat the remaining ddc_stepmania/ conflict-marker files as quarantined legacy-subtree content.

Changed

• Clarified that the remaining unresolved merge-conflict markers are now isolated to legacy subtree content rather than the active main-path modernization surface.

[0.2.12] - 2026-04-04

Changed

• Resolved merge-conflict markers in autochart.py, learn/beatcalc.py, learn/data_gen.py, learn/models_v2.py, scripts/train_v2.py, scripts/smd_1_extract.sh, and scripts/smd_4_analyze.sh.
• Refreshed the repository health audit and reduced unresolved merge-conflict-marker files from 9 to 2.

[0.2.11] - 2026-04-04

Changed

• Resolved top-level merge-conflict-marker files in AGENTS.md, CLAUDE.md, GEMINI.md, GPT.md, LLM_INSTRUCTIONS.md, and setup.py.
• Refreshed the repository health audit and reduced unresolved merge-conflict-marker files from 15 to 9.

[0.2.10] - 2026-04-04

Added

• Added scripts/audit_repo_health.py for repeatable repository-health auditing.
• Added docs/REPO_HEALTH_AUDIT_2026-04-04.md documenting unresolved merge conflicts, TensorFlow hotspots, and legacy .h5 / train_v2.py / models_v2 references.

Changed

• Documented the remaining blocker surface that should be cleaned up before considering the repo fully normalized after the PyTorch-oriented training refresh work.

[0.2.9] - 2026-04-04

Added

• Added scripts/compare_bucket_counts.py for reproducible before/after split-count comparison between prepared work directories.
• Added docs/BUCKET_SPLIT_DELTA_2026-04-04.md documenting the exact train/valid/test deltas in every bucket after the .ssc-inclusive refresh.

Changed

• Documented the exact downstream DDC training-input expansion for the practical 8-bucket plan.

[0.2.8] - 2026-04-04

Added

• Added docs/RETRAINING_REFRESH_PLAN_2026-04-04.md capturing the exact next-phase .ssc-inclusive retraining workflow.

Changed

• Cleaned and replaced the conflicted root README.md with a concise current-state project overview and documentation index.

[0.2.7] - 2026-04-04

Added

• Added scripts/audit_note_objects.py for repeatable note-object semantic audits against extracted JSON charts.
• Added docs/NOTE_OBJECT_SEMANTICS_2026-04-04.md documenting observed note-object semantics in the refreshed official DDR corpus.

Changed

• Documented that the refreshed corpus contains taps, hold-heads, tails, and mines, while no roll-head, attack, fake, keysound, or lift symbols were observed in the audited official-pack extraction.
• Documented per-bucket semantic coverage so future difficulty-model work can target hold/mine-aware features more precisely.

[0.2.6] - 2026-04-04

Added

• Documented FFR .ssc-inclusive loader expansion and validation counts in docs/SSC_EXPANSION_ANALYSIS_2026-04-04.md.

Changed

• Updated the FFR difficulty-data loader to prefer .ssc over .sm within each song directory.
• Cleaned learn/extract_feats_v2.py so the refreshed feature-extraction path is syntactically valid again after prior merge-conflict residue.
• Validated the refreshed FFR preprocessing path at approximately 1255 simfiles and 9407 serialized charts.

[0.2.5] - 2026-04-04

Added

• Added docs/SSC_EXPANSION_ANALYSIS_2026-04-04.md documenting the measured corpus delta unlocked by .ssc extraction support.

Changed

• Fixed scripts/prepare_data.py so the downstream data-prep path is syntactically valid again after prior merge-conflict residue.
• Documented the precise .ssc-driven corpus expansion:
  • songs: 1234 -> 1254
  • charts: 9241 -> 9403
  • major practical DDC buckets all increased

[0.2.4] - 2026-04-04

Added

• Added .ssc extraction support to dataset/extract_json.py using simfile for .ssc-only songs.
• Added validation/audit updates confirming the refreshed extracted corpus grows from 1234 to 1254 songs and from 9241 to 9403 charts after including .ssc content.

Changed

• Updated corpus-audit reporting to reflect the new .ssc-inclusive extraction state.
• Updated training analysis to distinguish between extractor support being present and full retraining against the refreshed corpus still being pending.

[0.2.3] - 2026-04-04

Added

• Added a reproducible corpus-audit script at scripts/audit_corpus.py.
• Added docs/CORPUS_AUDIT_2026-04-04.md documenting raw .sm/.ssc inventory, note-symbol distribution, and per-bucket special-symbol coverage.

Changed

• Documented that 20 .ssc files exist in the official DDR corpus but are not yet ingested by the current extractor.
• Documented that the extracted DDC symbolic corpus contains substantial non-binary note symbols (2, 3, M) and is therefore not tap-only.

[0.2.2] - 2026-04-04

Added

• Added a comprehensive training/data-coverage report in docs/TRAINING_ANALYSIS_2026-04-04.md.
• Added local export/documentation flow for the newly trained DDC v2 model set.

Changed

• Switched the DDC training orchestrator to use the PyTorch trainer for onset and SymNet training.
• Trained the practical 8-bucket placement configuration for ArrowVortex-oriented DDR use:
  • single Easy / Medium / Hard / Challenge
  • double Easy / Medium / Hard / Challenge
• Updated inference scaffolding in infer/autochart_lib.py toward PyTorch checkpoint loading.
• Corrected the FFR difficulty-model training path so dance-single and dance-double both survive data cleaning and train successfully.
• Preserved floating-point difficulty regression behavior for the difficulty evaluator.
• Added ignore rules for large local training artifacts and generated exports.

Fixed

• Fixed target-shape/model-output issues in the PyTorch DDC training path.
• Fixed mode-specific NaN handling in ffr-difficulty-model/scripts/train_model.py.
• Resolved repository documentation inconsistency around current training status and artifact handling.

[0.2.1] - 2023-10-27

Added

• Versioning: Centralized version number in VERSION file.
• Documentation: Added DASHBOARD.md, HANDOFF.md, and LLM instruction files.
• Submodule: Ensured ffr-difficulty-model is fully integrated.

[0.2.0] - 2023-10-27

Added

• Modernization: Ported entire codebase from Python 2.7 / TensorFlow 0.12 to Python 3.8+ / TensorFlow 2.x.
• Git Submodules: Added ddc_onset for beat detection and ffr-difficulty-model for difficulty rating.
• AutoChart CLI: New autochart.py tool for end-to-end chart generation, difficulty rating, and metadata handling.
• Training Pipeline: scripts/train_all.py allows full retraining of DDC models from raw audio and .sm files.
• Feature Extraction: Replaced essentia dependency with librosa.
• Packaging: Added setup.py for standard pip installation.

Changed

• Server: infer/ddc_server.py updated to use the new AutoChart library logic.
• Dependencies: Updated requirements.txt to reflect modern ecosystem (TF 2.x, librosa, simfile).
• Structure: Reorganized legacy scripts; removed obsolete bash scripts.

Removed

• Legacy Code: Removed infer/onset_net.py and infer/sym_net.py (logic moved to ddc_onset submodule and learn/models_v2.py).