# Changelog

## Baseline

- First survey — no prior run to compare against. CAI 48.6.

## Lenses

- Code Health 68.3
- Architecture 76.3
- Maturity 63.3
- Readiness 22.9
- Security 96.2
- Performance 100.0

## Changes since last survey

- 300 commits — 188 feature/other, 112 fixes

## By area

- app/src — 246 commits
- (root) — 51 commits
- app/build.gradle.kts — 3 commits

## Notable commits

- fix: Fix bass bias bug, align cyber matrix cubes to controller nodes, add fake bloom layers
- fix: Fix image crash, enable persistent chat images, sync external prompts, and rename MemoryDatabase
- fix: Fix snare and stage light laser triggers with signed FFT byte extraction and responsive thresholding
- fix: Fix(Audio): Restore original '(operator sent a voice message - please respond to contents of message)' prompt
- fix: Fix(Audio): Update audio prompt to '[operator has sent a voice message - please listen and respond to contents of audio]', enhance listening cue, and use VOICE_RECOGNITION audio source
- fix: Fix(Multimodal): Enable maxNumImages in EngineConfig, ground vision/audio in system prompt, and fix voice message directive
- fix: Fix(Option B): Decouple background parallax (0.08x) from avatar (1.20x) and tame individual node popping into cohesive harmonic scaling
- fix: Fix(Option B): Decouple idle thickness, arm spread, and individual white flash + node pop
- fix: Fix(Option B): Restore original grand scales for core, flower lattice, node sizes, and halos across idle and active states
- fix: Fix(Option B): Restore sleek delicate crystalline dots (7px-13px idle, 11px-18px active) and wide arm spread matching original blue reference
- fix: Fix: Restore Option A star scale to 1250f to enclose iris rings and implement global TTSManager.stopAll() mic interruption
- fix: Visualizer: Fix Option A glow alignment and apply universal calibrated stereoscopic parallax across all presets
- fix: fix(agent): eliminate duplicate TTS speech and strip emojis from audio synthesis
- fix: fix(agent): enforce strict length bounds on session memory compaction to guarantee infinite continuous runtime
- fix: fix(agent): factor audio tokens (25t/sec, 750t/30s) and session media KV budget into auto-compaction guard
- fix: fix(agent): fix LiteRT single-session compaction crash and enforce proactive 3200-token KV auto-flush
- fix: fix(agent): harden alarm time parsing, unquoted parameter recovery, and inference loop guard
- fix: fix(agent): instant 3rd-token streaming TTS, ambient notification sensory grounding without history corruption, and image send trigger
- fix: fix(agent): prevent prompt swallowing on KV cache reset and ensure dialogue continuity
- fix: fix(agent): restore lean tool schema, targeted media ducking, and RAM-aware E2B model selection
- …and 280 more

## Architecture

- 0 containers · 1 bounded contexts · 0 dependency edges (baseline)
