Repository navigation
refactor(ui): Split FloatingPanel.swift into 10 organized files - #2
Merged
Merged
Conversation
Break up the monolithic 2,413-line FloatingPanel.swift into a clean
folder structure organized by concern:
FloatingPanel/
├── FloatingPanelController.swift - Window controller
├── FloatingPanelView.swift - Main SwiftUI view
├── PillStateManager.swift - State machine
├── Components/
│ ├── WaveformViews.swift - Audio visualizers
│ ├── PillViews.swift - Pill state views (Idle, Recording, Processing, Reviewing)
│ ├── CelebrationViews.swift - Success animations
│ ├── ErrorViews.swift - Error handling UI
│ ├── AttentionPromptView.swift - Notification prompts
│ └── ReviewTrayView.swift - Action item tray
└── Helpers/
└── LawsComponents.swift - UI primitives (buttons, status text)
Also removes ~250 lines of dead "retro" code (ScrewView, RetroButton,
LEDVisualizer, MarqueeTextView, etc.) that was no longer used.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Owner
Author
Code reviewNo issues found. Checked for bugs and CLAUDE.md compliance. Review Summary:
This is a clean refactoring that improves code organization without changing functionality. 🤖 Reviewed by Claude Code |
r3dbars
added a commit
that referenced
this pull request
Mar 18, 2026
refactor(ui): Split FloatingPanel.swift into 10 organized files
This was referenced Apr 25, 2026
Merged
6 tasks
This was referenced Jun 16, 2026
r3dbars
added a commit
that referenced
this pull request
Jun 17, 2026
…m (it is NOT a fix) Wired mean-centering into the matcher side (transform cached dump embeddings, replay through the REAL clusterer + DB matcher; measurement only, no app code) and tested 4 centering modes x 6 codec arms + an online alpha-sweep + frozen-warmup. Adversarially verified (3 independent probes recomputing from raw data; baseline==normonly control holds). Result OVERTURNS the prior report's optimistic "lever 0: mean-center embeddings" claim: - Global (oracle) centering does NOT reduce real downstream false positives. At profile- count-matched operating points, merge-contamination (fp_mass) is within +-0.02 of baseline (opus12k actually +0.011, worse); every fp drop is matched by an equal split- error (fn) rise. Baseline and global ride the SAME fp-vs-profiles frontier — global just slides rightward to more profiles. Clean 1:1 identity recovery never approaches 32. - The only genuine effect is un-collapsing mega-profiles on heavy codecs (opus8k 5->20 profiles) — a fragmentation/band-separation move, not precision; people-trapped stays ~30/32 through it. - WHY: ~28/30 trapped speakers and 75-88% of fused pairs are WITHIN-MEETING diarizer co-clusters (verified byte-for-byte: centering never alters the diarizer's clusters). Only ~2 speakers/arm are matcher-side. A matcher-free floor of ~28-32 trapped exists on every arm before matching. The diarizer is the dominant ceiling. - Deployability: no online/causal estimator approaches the oracle. Best running EMA alpha=0.005 recovers ~71%/4% of the oracle band-sep gain (opus12k/opus8k) and over- fragments; frozen-warmup matches oracle direction (cos ~0.8) but under-merges. Needs out-of-band per-codec calibration, not live-stream estimation. Report updated: AMI_CODEC_REID_REPORT.md gains §3a + a revised lever ranking (diarizer #1, codec-augmented embeddings #2; offline mean-centering demoted to a narrow anti-collapse remedy; online centering not deployable). Tooling: center_dumps.py (normonly/global/ per_meeting/running/frozen), run_centering_experiment.sh, compare_centering.py, center_purity_analysis.py, q3_{attribution,trapped,floor}.py, dev/run_online_centering_sweep.sh. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
r3dbars
added a commit
that referenced
this pull request
Jun 19, 2026
…al verdict + shippable gate) 5-agent parallel analysis of the quality matrix. Refinements: - Parameter ranking for cross-quality robustness: autoBar #1, marginMin #2 (VoxCeleb only; inert on AMI), suggestFloor is the prompts knob. margins 0.00/0.03/0.05 are no-ops (gate engages at 0.08); 0.12 cuts worst-channel false-auto ~10x on VoxCeleb at ~2% prompt cost. - Adversarial verdict: degradation->confusion is ROBUST via suggest-precision (Spearman -0.78/-0.71, tight CIs) and strongest in the >=30-auto tier (not small-N). Honest caveats: "any degradation" is false (noise/opus_8k/mp3-collapse driven); VoxCeleb autoBar>=0.92 "0% false-auto" is a denominator-collapse artifact. - Shippable gate: autoBar 0.83 + margin 0.12 + fixed cc>5 (VoxCeleb); AMI in-room should stay SUGGEST-ONLY (no policy clears fa<=0.5% with >=50 degraded autos — all "safe" gates are denominator-vacuous). AMI-full N=189 run will certify this. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This was referenced Jul 3, 2026
6 tasks
14 of 18 tasks
11 of 13 tasks
This was referenced Sep 25, 2026
This was referenced Oct 7, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
FloatingPanel.swiftinto 10 organized filesComponents/,Helpers/)New Structure
Dead Code Removed
ScrewView- Unused retro screw decorationRetroButton- Old button styleStatusTextView- Replaced byLawsStatusTextViewMarqueeTextView- Unused scrolling textLEDVisualizer- Old LED-style visualizerTest plan
🤖 Generated with Claude Code