fix(docs): replace hardcoded local absolute paths with relative paths

Author: AIEraDev

docs/split-clip-audio-architecture-investigation.md

@@ -10,23 +10,23 @@ This report investigates the current implementation without changing the existin
 
 The native playback path is clip-ID-keyed decoded PCM, not a persistent decoder session keyed by source path.
 
-The frontend builds one immutable native timeline snapshot from active audio clips, starts the native clock, clears the existing native clip list, and loads every snapshot clip concurrently ([nativeAudioTimeline.ts](/Users/AIEraDev/Documents/clypra-family/clypra/src/core/audio/nativeAudioTimeline.ts:59)). The Tauri load command opens/decodes one source path and installs the resulting clip into the clock mixer ([native_audio.rs](/Users/AIEraDev/Documents/clypra-family/clypra/src-tauri/src/commands/native_audio.rs:100)).
+The frontend builds one immutable native timeline snapshot from active audio clips, starts the native clock, clears the existing native clip list, and loads every snapshot clip concurrently ([nativeAudioTimeline.ts](../src/core/audio/nativeAudioTimeline.ts:59)). The Tauri load command opens/decodes one source path and installs the resulting clip into the clock mixer ([native_audio.rs](../src-tauri/src/commands/native_audio.rs:100)).
 
-Each decode opens a fresh FFmpeg input and creates a `DecodedAudioClip` containing the clip ID, timeline position, source start, duration, and PCM samples ([decoder.rs](/Users/AIEraDev/Documents/clypra-family/clypra/src-tauri/src/audio/decoder.rs:71), [mixer.rs](/Users/AIEraDev/Documents/clypra-family/clypra/src-tauri/src/audio/mixer.rs:75)). The native clock stores `NativePcmClip` values in a mixer vector and replaces/removes them by clip ID ([native_audio.rs](/Users/AIEraDev/Documents/clypra-family/clypra/src-tauri/src/native_audio.rs:116)). There is no native source-path decoder pool or decoder-position LRU.
+Each decode opens a fresh FFmpeg input and creates a `DecodedAudioClip` containing the clip ID, timeline position, source start, duration, and PCM samples ([decoder.rs](../src-tauri/src/audio/decoder.rs:71), [mixer.rs](../src-tauri/src/audio/mixer.rs:75)). The native clock stores `NativePcmClip` values in a mixer vector and replaces/removes them by clip ID ([native_audio.rs](../src-tauri/src/native_audio.rs:116)). There is no native source-path decoder pool or decoder-position LRU.
 
 ### Boundary behavior
 
-`source_start_ticks` is applied during each clip's initial decode. The corrected global seek and preroll trimming are therefore exercised for every clip whose `trimIn` is nonzero, including every right-hand split piece ([decoder.rs](/Users/AIEraDev/Documents/clypra-family/clypra/src-tauri/src/audio/decoder.rs:118)).
+`source_start_ticks` is applied during each clip's initial decode. The corrected global seek and preroll trimming are therefore exercised for every clip whose `trimIn` is nonzero, including every right-hand split piece ([decoder.rs](../src-tauri/src/audio/decoder.rs:118)).
 
-Continuous playback does not reposition a shared decoder at a boundary. The mixer calculates each clip's source position from `timeline_ticks - timeline_start_ticks` and samples the already-decoded PCM ([native_audio.rs](/Users/AIEraDev/Documents/clypra-family/clypra/src-tauri/src/native_audio.rs:230)). Explicit seeks only move the native timeline clock ([native_audio.rs](/Users/AIEraDev/Documents/clypra-family/clypra/src-tauri/src/native_audio.rs:628)); they do not re-seek or mutate an FFmpeg decoder.
+Continuous playback does not reposition a shared decoder at a boundary. The mixer calculates each clip's source position from `timeline_ticks - timeline_start_ticks` and samples the already-decoded PCM ([native_audio.rs](../src-tauri/src/native_audio.rs:230)). Explicit seeks only move the native timeline clock ([native_audio.rs](../src-tauri/src/native_audio.rs:628)); they do not re-seek or mutate an FFmpeg decoder.
 
-When a split is present, both halves can be installed simultaneously. They cannot interfere through a shared decoder cursor, but each half independently decodes its source range. The native mixer bounds resource usage at 64 clips and 512 MiB of PCM ([native_audio.rs](/Users/AIEraDev/Documents/clypra-family/clypra/src-tauri/src/native_audio.rs:10), [native_audio.rs](/Users/AIEraDev/Documents/clypra-family/clypra/src-tauri/src/native_audio.rs:151)). Native audio has no independent 2–3 second decoder lookahead; the separate video prewarm path is bounded at 1.5 seconds ([PreviewMediaPool.ts](/Users/AIEraDev/Documents/clypra-family/clypra/src/core/resources/PreviewMediaPool.ts:240)).
+When a split is present, both halves can be installed simultaneously. They cannot interfere through a shared decoder cursor, but each half independently decodes its source range. The native mixer bounds resource usage at 64 clips and 512 MiB of PCM ([native_audio.rs](../src-tauri/src/native_audio.rs:10), [native_audio.rs](../src-tauri/src/native_audio.rs:151)). Native audio has no independent 2–3 second decoder lookahead; the separate video prewarm path is bounded at 1.5 seconds ([PreviewMediaPool.ts](../src/core/resources/PreviewMediaPool.ts:240)).
 
 The source-path/session question is therefore resolved explicitly: native audio is keyed by clip ID after independent decode, not by a shared source-path decoder session. There is also no native audio lookahead buffer with a second decode path to validate. The reproduction suite covers the closest equivalent by installing a future clip before its timeline boundary and asserting that it remains silent until the boundary, then starts with the correct source segment.
 
 ### Browser fallback
 
-The browser path is separate. Its decoded `AudioBuffer` is cached by media/source key, while active playback voices are keyed by clip ID ([AudioEngine.ts](/Users/AIEraDev/Documents/clypra-family/clypra/src/core/audio/AudioEngine.ts:142)). A shared full-source buffer is safe because each voice starts at `trimIn + timeIntoClip` ([AudioEngine.ts](/Users/AIEraDev/Documents/clypra-family/clypra/src/core/audio/AudioEngine.ts:158)). It does not reproduce the native shared-decoder-position bug, although it can share an incorrectly selected buffer if a clip's `audioPath` differs from the media asset while retaining the same `mediaId`.
+The browser path is separate. Its decoded `AudioBuffer` is cached by media/source key, while active playback voices are keyed by clip ID ([AudioEngine.ts](../src/core/audio/AudioEngine.ts:142)). A shared full-source buffer is safe because each voice starts at `trimIn + timeIntoClip` ([AudioEngine.ts](../src/core/audio/AudioEngine.ts:158)). It does not reproduce the native shared-decoder-position bug, although it can share an incorrectly selected buffer if a clip's `audioPath` differs from the media asset while retaining the same `mediaId`.
 
 ### Classification
 
@@ -36,36 +36,36 @@ The native architecture is orthogonal to the original wrong-time-base bug. The f
 
 ### Call sites and range behavior
 
-The `extract_waveform_data` Tauri command is invoked from one frontend hook ([useWaveformData.ts](/Users/AIEraDev/Documents/clypra-family/clypra/src/components/editor/timeline/useWaveformData.ts:90)); it is registered as a Tauri command ([lib.rs](/Users/AIEraDev/Documents/clypra-family/clypra/src-tauri/src/lib.rs:198)). Timeline video clips use `VolumeWaveform`, and audio clips use `TimelineWaveform`, with one waveform hook invocation per rendered clip ([Clip.tsx](/Users/AIEraDev/Documents/clypra-family/clypra/src/components/editor/timeline/Clip.tsx:815)).
+The `extract_waveform_data` Tauri command is invoked from one frontend hook ([useWaveformData.ts](../src/components/editor/timeline/useWaveformData.ts:90)); it is registered as a Tauri command ([lib.rs](../src-tauri/src/lib.rs:198)). Timeline video clips use `VolumeWaveform`, and audio clips use `TimelineWaveform`, with one waveform hook invocation per rendered clip ([Clip.tsx](../src/components/editor/timeline/Clip.tsx:815)).
 
-The native command passes FFmpeg `-ss` and `-t` from the requested `start_time` and `duration` ([media.rs](/Users/AIEraDev/Documents/clypra-family/clypra/src-tauri/src/commands/media.rs:499)). Therefore the native waveform decode is scoped to the clip's visible trimmed range rather than intentionally decoding the entire source file.
+The native command passes FFmpeg `-ss` and `-t` from the requested `start_time` and `duration` ([media.rs](../src-tauri/src/commands/media.rs:499)). Therefore the native waveform decode is scoped to the clip's visible trimmed range rather than intentionally decoding the entire source file.
 
-The browser fallback is different: `decodeAudioData` loads the complete source buffer, then JavaScript slices `startSample..endSample` for bucket computation ([useWaveformData.ts](/Users/AIEraDev/Documents/clypra-family/clypra/src/components/editor/timeline/useWaveformData.ts:118)). Thus the fallback performs full-file decode work even though the displayed waveform is trimmed.
+The browser fallback is different: `decodeAudioData` loads the complete source buffer, then JavaScript slices `startSample..endSample` for bucket computation ([useWaveformData.ts](../src/components/editor/timeline/useWaveformData.ts:118)). Thus the fallback performs full-file decode work even though the displayed waveform is trimmed.
 
 ### Split and caching behavior
 
-The split command creates two new clips, preserving the source path and assigning left/right trim ranges ([SplitClipCommand.ts](/Users/AIEraDev/Documents/clypra-family/clypra/src/core/history/commands/SplitClipCommand.ts:55)). The waveform cache key includes resolved path, source start, source duration, and bucket count ([useWaveformData.ts](/Users/AIEraDev/Documents/clypra-family/clypra/src/components/editor/timeline/useWaveformData.ts:79)). Consequently, a cold-cache split creates two waveform requests, one per resulting range. An N-way split creates N requests for N distinct ranges:
+The split command creates two new clips, preserving the source path and assigning left/right trim ranges ([SplitClipCommand.ts](../src/core/history/commands/SplitClipCommand.ts:55)). The waveform cache key includes resolved path, source start, source duration, and bucket count ([useWaveformData.ts](../src/components/editor/timeline/useWaveformData.ts:79)). Consequently, a cold-cache split creates two waveform requests, one per resulting range. An N-way split creates N requests for N distinct ranges:
 
 - Native/Tauri: N bounded-range FFmpeg decodes.
 - Browser fallback: N full-source `decodeAudioData` operations followed by N range slices.
 
 Exact duplicate requests can share the existing 50-entry LRU cache, but there is no coarse source-wide waveform cache analogous to the filmstrip L0 cache. This is an efficiency finding, not a playback-correctness defect.
 
-The separate media-card waveform implementation also decodes through Web Audio, but it is not a call site of `extract_waveform_data` and is unrelated to split-clip timeline rendering ([MediaCardWaveform.tsx](/Users/AIEraDev/Documents/clypra-family/clypra/src/components/ui/cards/MediaCardWaveform.tsx:32)).
+The separate media-card waveform implementation also decodes through Web Audio, but it is not a call site of `extract_waveform_data` and is unrelated to split-clip timeline rendering ([MediaCardWaveform.tsx](../src/components/ui/cards/MediaCardWaveform.tsx:32)).
 
 ## Part 3 — Detached audio isolation
 
-`DetachAudioCommand` creates an audio-kind clip, sets `audioPath` to the selected source asset path, retains `detachedFromClipId`, and mutes the original video clip ([DetachAudioCommand.ts](/Users/AIEraDev/Documents/clypra-family/clypra/src/core/history/commands/DetachAudioCommand.ts:51)). Native timeline audio prioritizes `clip.audioPath` over the media asset path ([audioClips.ts](/Users/AIEraDev/Documents/clypra-family/clypra/src/core/timeline/audioClips.ts:82)). Waveform rendering also prioritizes the explicit path ([Clip.tsx](/Users/AIEraDev/Documents/clypra-family/clypra/src/components/editor/timeline/Clip.tsx:849)). For the built-in detach command, the explicit path currently equals the original video asset path by design.
+`DetachAudioCommand` creates an audio-kind clip, sets `audioPath` to the selected source asset path, retains `detachedFromClipId`, and mutes the original video clip ([DetachAudioCommand.ts](../src/core/history/commands/DetachAudioCommand.ts:51)). Native timeline audio prioritizes `clip.audioPath` over the media asset path ([audioClips.ts](../src/core/timeline/audioClips.ts:82)). Waveform rendering also prioritizes the explicit path ([Clip.tsx](../src/components/editor/timeline/Clip.tsx:849)). For the built-in detach command, the explicit path currently equals the original video asset path by design.
 
-Detached audio does not enter the visual compositor: the evaluator explicitly skips `clip.kind === "audio"` before creating media layers ([evaluator.ts](/Users/AIEraDev/Documents/clypra-family/clypra/src/core/evaluation/evaluator.ts:225)), and the preview media pool similarly excludes audio-kind clips from video bindings ([PreviewMediaPool.ts](/Users/AIEraDev/Documents/clypra-family/clypra/src/core/resources/PreviewMediaPool.ts:409)). Existing evaluator and preview-pool regression tests pass.
+Detached audio does not enter the visual compositor: the evaluator explicitly skips `clip.kind === "audio"` before creating media layers ([evaluator.ts](../src/core/evaluation/evaluator.ts:225)), and the preview media pool similarly excludes audio-kind clips from video bindings ([PreviewMediaPool.ts](../src/core/resources/PreviewMediaPool.ts:409)). Existing evaluator and preview-pool regression tests pass.
 
 ### Latent divergence
 
-The canonical evaluator's audio-layer construction requires an asset and resolves `sourcePath` from `asset.path`, ignoring `clip.audioPath` ([evaluator.ts](/Users/AIEraDev/Documents/clypra-family/clypra/src/core/evaluation/evaluator.ts:313)). This does not affect the current native preview path, which uses `getActiveAudioClips`, nor the current detach command, because both paths are equal today. It is a latent correctness gap for any audio-kind clip backed by a video asset whose explicit `audioPath` differs from that asset. No fix was applied during this investigation because the reproduced detach scenarios did not demonstrate wrong audible content in the active playback path.
+The canonical evaluator's audio-layer construction requires an asset and resolves `sourcePath` from `asset.path`, ignoring `clip.audioPath` ([evaluator.ts](../src/core/evaluation/evaluator.ts:313)). This does not affect the current native preview path, which uses `getActiveAudioClips`, nor the current detach command, because both paths are equal today. It is a latent correctness gap for any audio-kind clip backed by a video asset whose explicit `audioPath` differs from that asset. No fix was applied during this investigation because the reproduced detach scenarios did not demonstrate wrong audible content in the active playback path.
 
 ## Part 4 — Reproduction matrix
 
-The deterministic headless reproduction is implemented in [split_investigation_tests.rs](/Users/AIEraDev/Documents/clypra-family/clypra/src-tauri/src/audio/split_clip_investigation_tests.rs:1). It generates a temporary six-second A/V fixture with 440 Hz, 880 Hz, and 1760 Hz source regions, then validates source-frequency identity and amplitude.
+The deterministic headless reproduction is implemented in [split_investigation_tests.rs](../src-tauri/src/audio/split_clip_investigation_tests.rs:1). It generates a temporary six-second A/V fixture with 440 Hz, 880 Hz, and 1760 Hz source regions, then validates source-frequency identity and amplitude.
 
 | Scenario | Headless result | Classification |
 | --- | --- | --- |

docs/text-effects-architecture.md

@@ -75,7 +75,7 @@ The interpreter must be structurally incapable of executing anything beyond a wh
 
 ### 1.2 Offline, Caching & Export Policy
 
-The desktop app and web studio must provide a reliable offline experience while strictly enforcing Clypra's [Architecture-First ADR](file:///Users/AIEraDev/Documents/clypra-family/clypra/docs/architecture-first-delivery-adr.md) (*"Silent fallback is not an acceptable production failure policy"*):
+The desktop app and web studio must provide a reliable offline experience while strictly enforcing Clypra's [Architecture-First ADR](./architecture-first-delivery-adr.md) (*"Silent fallback is not an acceptable production failure policy"*):
 
 - **Storage Location**:
   - Desktop: Content-addressed cache directory (`~/.clypra/cache/text-effects/{id}_v{version}.json`).
@@ -290,7 +290,7 @@ To eliminate typography discrepancy between Web (WASM) and Desktop (Tauri):
 
 ## 7. Parity Verification & Golden-Pixel Harness
 
-In accordance with Clypra's [Golden-Frame Protocol](file:///Users/AIEraDev/Documents/clypra-family/clypra/docs/golden-frame-protocol.md):
+In accordance with Clypra's [Golden-Frame Protocol](./golden-frame-protocol.md):
 
 1. **Primitive Fixture Suite**:
    - Every primitive from §3 is evaluated with a canonical test string ("Clypra FX 4K") and deterministic uniforms.