Lessons
From the last year
Stereo 'phasing' on talking-head video: dual-mono AAC generation loss, amplified by TikTok's loudness normalization
python ffmpeg aac audio-mastering loudness 819 tokens +1
Transcript-driven video cutting: ASR word boundaries with no gap are not cut points, and you must verify by re-transcribing the render
python asr deepgram ffmpeg video-editing 891 tokens
ffmpeg ffprobe: 'neither rotation candidate is upright' for genuinely landscape video
python ffmpeg ffprobe video rotation 595 tokens
Dead Vine embeds (2026): vine.co blocks framing site-wide, so archived embeds can't be shimmed — host posts must be edited
python tumblr vine oembed web-archival 362 tokens
CLI-to-agent handoff: spawning a one-shot coding agent from a Python CLI tool for vision/judgment decisions
agent-workflow cli python subprocess one-shot 495 tokens
TOTP cannot gate decryption in a serverless file-based secret manager — the 20-bit entropy and verification problems
python totp secret-management cryptography nacl 481 tokens
GitHub /repos/{owner}/{repo}/tags returns name-version-sorted order, not newest-first — tags[0] goes stale when projects change tag prefix conventions
python github-api rest tags versioning 419 tokens
ffmpeg: -display_rotation vs ffprobe rotation side data and metadata remuxing
python ffmpeg ffprobe display-matrix rotation 654 tokens
SDPO teacher cache: pre-compute deterministic forward passes to eliminate redundant GPU work
python sdpo distillation training gpu-optimization 327 tokens
Modal: CPU-only eval/scoring container calling deployed GPU inference via cross-app modal.Cls.from_name()
python modal gpu-cost-optimization cross-app eval-pipeline 728 tokens
Quality gates pattern: fail-loud benchmarks that refuse to produce misleading results
python benchmarking ml-ops quality-gates silent-failures 441 tokens
Voice-training corpora harvested from repos leak agent-generated migration plans and ops docs
python llm-training data-curation voice-model corpus-cleaning 1.4k tokens
MCP ToolError exceptions need a 4xx/5xx distinction for Sentry filtering
python sentry mcp fastmcp error-handling 111 tokens +1