Research digest titles¶
Every research digest, one heading each (799), so the site search finds a digest by its title; digest bodies are not in the search index (ADR-1512).
Research-0001: Cache shape for bisect-model-quality nightly¶
0001-bisect-model-quality-cache
Research-0002: Automating process-ADR enforcement¶
0002-automated-rule-enforcement
Research-0003: SSIMULACRA 2 port sourcing and upstream-drift strategy¶
0003-ssimulacra2-port-sourcing
Research-0004: Vulkan compute backend — toolchain, loader, memory model, DMABUF import¶
Research-0005: Embedded MCP in libvmaf — threading, JSON library, SSE server, Power-of-10 fit¶
Research-0006: Tiny-AI PTQ — accuracy regression targets, ORT API comparison, calibration-set sourcing¶
0006-tinyai-ptq-accuracy-targets
Research-0007: SSIMULACRA 2 scalar port — YUV handling, blur deviation, snapshot tooling¶
Research-0008: MS-SSIM decimate SIMD — FLOP accounting, summation order, bit-exactness¶
Research-0009: Batch-A upstream small-fix sweep — porting strategy for open PRs¶
0009-batch-a-upstream-port-strategy
Research-0010: Is Netflix about to ship a SpEED-driven VMAF successor?¶
0010-speed-netflix-upstream-direction
Research-0011: _iqa_convolve AVX2 — bit-exactness via __m256d, kernel invariants, Amdahl¶
Research-0012: SSIM SIMD bit-exactness to scalar — where the ULP drifted¶
Research-0013: SIMD DX framework — audit + NEON bit-exactness port¶
Research digest 0014 — psnr_hvs NEON sister port (T3-5-neon)¶
Research digest 0015 — SSIMULACRA 2 SIMD ports (T3-1 + T3-2)¶
Research digest 0016 — SSIMULACRA 2 IIR blur SIMD (T3-1 phase 2)¶
0016-ssimulacra2-iir-blur-simd
Research digest 0017 — SSIMULACRA 2 picture_to_linear_rgb SIMD (T3-1 phase 3)¶
Research digest 0018 — SSIMULACRA 2 snapshot-JSON regression gate (T3-3)¶
0018-ssimulacra2-snapshot-gate
Research Digest 0019 — Tiny-AI Training on the Netflix VMAF Corpus¶
Research digest 0020 — cambi GPU port strategies¶
Research Digest 0021 — Corpus + Architecture Survey for the Next Tiny-AI Iteration¶
0021-tiny-ai-corpus-arch-survey
Research Digest 0022 — LOSO results for mlp_small on the Netflix corpus¶
Research Digest 0023 — 3-arch LOSO results on the Netflix corpus¶
Research-0024 — VIF kernelscale: fork-vs-upstream divergence and port strategy¶
Research-0025 — FoxBird outlier resolved via Netflix + KoNViD-1k combined training¶
0025-foxbird-resolved-via-konvid
Research-0026 — Cross-metric feature fusion for tiny-AI¶
0026-cross-metric-feature-fusion
Research-0027 — Phase-2 feature correlation, MI, and importance results¶
0027-phase2-feature-importance
Research-0028 — Phase-3 MLP arch sweep on Top-K feature subsets¶
Research-0029 — Phase-3b: StandardScaler retry of the subset sweep¶
0029-phase3b-standardscaler-results
Research-0030 — Phase-3b multi-seed validation (Gate 1)¶
0030-phase3b-multiseed-validation
Research-0031: Intel AI-PC NPU + EP applicability to tiny-AI / dnn/¶
0031-intel-ai-pc-applicability
Research Digest 0032 — cambi Vulkan integration trade-offs¶
Research-0033 — Bristol VI-Lab Neural Video Compression review (2026-04 preprint)¶
Research-0034: CHANGELOG + ADR-index fragment-file pattern¶
0034-changelog-fragment-pattern
Research-0040: Codec-aware FR conditioning for the tiny FR regressor¶
0040-codec-aware-fr-conditioning
Research-0041: GPU-generation ULP calibration corpus design¶
Research-0042: Vulkan async pending-fence model for the VkImage zero-copy import path¶
0042-vulkan-async-pending-fence
Research-0043 — DISTS extractor design digest¶
Research-0044: Quality-aware encode automation (vmaf-tune) — option-space digest¶
0044-quality-aware-encode-automation
Research-0045: Vulkan picture preallocation — option-space digest¶
0045-vulkan-picture-preallocation
Research-0046 — Bristol VI-Lab dataset feasibility for tiny-AI training and parity soak¶
0046-bristol-vi-lab-feasibility
Research-0046 — vmaf_tiny_v3 mlp_medium evaluation vs v2 mlp_small¶
0046-vmaf-tiny-v3-mlp-medium-evaluation
Research-0046: vmaf-tune saliency-aware ROI tuning (Bucket #2)¶
Research-0047 — CUDA graph capture feasibility for the per-frame submit chain¶
0047-cuda-graph-capture-feasibility
Research-0048: vmaf_tiny_v4 (mlp_large) evaluation — does the arch ladder saturate?¶
0048-vmaf-tiny-v4-mlp-large-evaluation
Research-0050: vmaf_tiny v3 + v4 — multi-seed Netflix LOSO + KoNViD 5-fold¶
0050-vmaf-tiny-v3-v4-multiseed-konvid-eval
Research-0051: SpEED-QA full-reference metric — feasibility on the lusoris fork¶
Research-0051 — ssimulacra2 GPU XYB shader precision re-investigation (T-GPU-OPT-VK-3)¶
0051-ssimulacra2-gpu-xyb-precision
Research-0052: Vulkan submit-side template + fence pool + descriptor pre-alloc¶
Research-0053 — MobileSal real-weights swap blocker¶
0053-mobilesal-real-weights-blocker
Research-0053: OSSF Scorecard investigation and remediation plan¶
0053-ossf-scorecard-investigation
Research Digest 0053 — Post-merge CPU profile, 2026-05-03¶
0053-post-merge-cpu-profile-2026-05-03
Research-0053: Vulkan 1.4 bump — NVIDIA-only float contraction regression¶
0053-vulkan-1-4-nvidia-fp-contraction-regression
Research-0054: iqa_convolve block-of-N tap widen — bit-exactness post-mortem¶
0054-iqa-convolve-block-widen-attempt
Research-0055 — ciede2000 Vulkan NVIDIA places=4 root cause: f32 vs f64 colour-space chain¶
0055-ciede-vulkan-nvidia-f32-f64-root-cause
Research-0055 — U-2-Net u2netp saliency replacement survey¶
0055-u2netp-saliency-replacement-survey
Research-0055: VIF AVX-512 polynomial-log2 attempt — bit-exactness forbids¶
0055-vif-polynomial-log2-attempt
Research-0056: precise decoration audit on vif.comp + ciede.comp — Step A implementation findings¶
0056-vif-ciede-precise-step-a-implementation
Research-0057 — vmaf_tiny_v5 corpus expansion (5-corpus = 4-corpus + UGC)¶
0057-vmaf-tiny-v5-corpus-expansion
Research 0058 — canonical-6 permutation importance for vmaf_tiny_v2¶
0058-canonical-6-permutation-importance
Research-0058: FR regressor v2 (codec-aware) feasibility¶
0058-fr-regressor-v2-feasibility
Research digest 0054 — libFuzzer scaffold for the Y4M parser¶
Research-0060: vmaf-tune fast — proxy-based recommend (Phase A.5)¶
Research-0061: CUDA integer_ms_ssim_cuda drain_batch wire-up — host-blocking syscall profile¶
0061-cuda-ms-ssim-drain-batch-profile
Research-0061 — Docs-only PR fast-track for CI matrix¶
Research-0061: vmaf-tune capability audit — beyond Phases A/B + the fast path¶
0061-vmaf-tune-capability-audit
Research-0062 — Content-aware fr_regressor_v2 feasibility¶
0062-content-aware-fr-regressor-feasibility
Research-0062 — Fork-trained saliency student on DUTS-TR¶
0062-saliency-student-from-scratch-on-duts
Research-0063 — Encoder knob-space stratifies by rate-control mode (CQ vs VBR)¶
0063-encoder-knob-space-cq-vs-vbr-stratification
Research-0063: T7-5 NOLINT sweep — citation audit + closeout¶
Research-0064: GPU-lane NOLINT audit — which suppressions are actually live¶
Research-0064 — vmaf-tune resolution-aware model selection + CRF offsets¶
0064-vmaf-tune-resolution-aware
Research-0065: NVENC codec adapters for vmaf-tune — option-space digest¶
Research Digest 0054 — vmaf-tune Intel QSV codec adapters¶
Research-0067: probabilistic fr_regressor_v2 — deep-ensemble + conformal¶
0067-fr-regressor-v2-probabilistic
Research-0067: fr_regressor_v2 PROD — LOSO ship-gate evaluation¶
0067-fr-regressor-v2-prod-loso
Research-0067 — K150K-A corpus integration feasibility and pipeline design¶
Research-0067: ssimulacra2_cuda cuda-reviewer follow-up (2026-05-09)¶
0067-ssimulacra2-cuda-leaks-perf
Research-0067: vmaf-tune Phase F — adaptive recipe-aware composition feasibility¶
0067-vmaf-tune-phase-f-feasibility-2026-05-08
Research-0068: vmaf-tune Phase E — per-title bitrate-ladder algorithm survey¶
0068-vmaf-tune-phase-e-bitrate-ladder
Research-0069: Region-of-interest VMAF — option-space digest¶
0069-vmaf-roi-saliency-weighted
Research-0070: vmaf-tune codec-agnostic encode dispatcher¶
0070-vmaf-tune-encode-multi-codec
Research-0071: vmaf-tune GPU score backend — option-space digest¶
0071-vmaf-tune-gpu-score-backend
Research-0072 — vmaf-tune HDR-aware encoding + scoring¶
Research-0073: vmaf-tune content-addressed cache¶
Research-0074: vmaf-tune Apple VideoToolbox codec adapters¶
0074-vmaf-tune-videotoolbox-adapters
Research-0075: fr_regressor_v2 ensemble — production flip LOSO protocol¶
0075-fr-regressor-v2-ensemble-prod-flip
Research-0076: vmaf-tune fast — production wiring¶
0076-vmaf-tune-fast-path-prod-wiring
Research-0077 — Encoder knob-space Pareto frontiers (per source × codec × rc_mode)¶
0077-encoder-knob-space-pareto-frontiers
Research-0078: ENCODER_VOCAB v3 — 16-slot schema expansion + retrain plan¶
0078-encoder-vocab-v3-schema-expansion
Research-0079: vmaf-tune Phase E ladder default sampler — gap analysis¶
0079-vmaf-tune-ladder-default-sampler
Research-0080: Encoder knob-sweep — populated Pareto hulls and recipe regressions¶
0080-encoder-knob-sweep-findings
Research-0081: fr_regressor_v2 ensemble — real-corpus retrain methodology¶
0081-fr-regressor-v2-ensemble-real-corpus-methodology
Research-0082 — BVI-DVC corpus feasibility for fr_regressor_v2¶
0082-bvi-dvc-corpus-feasibility
Research-0083: libFuzzer harness expansion — target survey¶
0083-libfuzzer-harness-expansion-target-survey
Research-0084: FFmpeg-side integration patterns for video-quality tools (vmaf-tune)¶
0084-ffmpeg-patch-vmaf-tune-integration-survey
Research-0085 — Vendor-neutral VVC GPU encode landscape¶
0085-vendor-neutral-vvc-encode-landscape
Research-0086: ADR Proposed status sweep — 2026-05-08¶
0086-adr-proposed-status-sweep-2026-05-08
Research-0086 — Contributor data-pack expansion via web-researched data¶
0086-contributor-pack-web-data-expansion-2026-05-08
Research-0086: KonViD-150k corpus feasibility for the lusoris fork¶
0086-konvid-150k-corpus-feasibility
Research-0086 — Local sidecar training feasibility¶
0086-local-sidecar-feasibility
Research-0086 — Stale-marker sweep, 2026-05-08¶
0086-stale-marker-sweep-2026-05-08
Research-0086: docs/state.md audit + backfill — 2026-05-08¶
0086-state-md-audit-2026-05-08
Research-0086 — SYCL / oneAPI / QSV toolchain audit (2026-05-08)¶
0086-sycl-toolchain-audit-2026-05-08
Research-0086 — Tiny-AI SOTA deep dive: is the lusoris approach state of the art?¶
0086-tiny-ai-sota-deep-dive-2026-05-08
Research-0086 — u2netp.pth fork-local mirror: Apache-2.0 §4 compliance + supply-chain surface¶
0086-u2netp-fork-mirror-license-compliance
Research-0086: Usage-doc coverage audit against ADR-shipped surfaces¶
0086-usage-doc-coverage-audit-2026-05-08
Research-0086 — Video-temporal saliency feasibility for ROI-encode tuning¶
0086-video-saliency-feasibility-2026-05-08
ChatGPT vision + Claude-bias audit — VMAFx/vmafx fork¶
0087-chatgpt-vision-and-claude-bias-audit
Research-0087 — External-competitor benchmark survey (Synamedia x264-pVMAF, DOVER-Mobile)¶
0087-external-bench-competitor-survey-2026-05-08
Research-0087: vmaf-tune codec-adapter dispatcher pivot (HP-1 audit)¶
0087-vmaftune-codec-adapter-dispatcher-hp1
Docs cleanup punch list (2026-05-08)¶
Research digest 0088: hardware-capability priors for the FR-regressor corpus¶
0088-hardware-capability-priors-2026-05-08
Research-0089 — AVX-512 follow-up audit sweep (2026-05-09)¶
0089-avx512-audit-sweep-2026-05-09
CI cost-optimization audit (2026-05-09)¶
0089-ci-cost-optimization-audit-2026-05-09
Research-0089 — HDR VMAF model: source / train / placeholder¶
Research-0089: libvmaf WebAssembly compilation feasibility¶
0089-libvmaf-wasm-feasibility-2026-05-09
Research-0089: mkdocs --strict gate — warning census + carve-out justification¶
Research-0089: MoltenVK feasibility on the fork's Vulkan shader inventory¶
0089-moltenvk-feasibility-on-fork-shaders
Research-0089: Multi-corpus MOS aggregation — scale unification feasibility¶
0089-multi-corpus-aggregation-feasibility
Research-0089: saliency_student_v2 — Resize-decoder ablation¶
0089-saliency-student-v2-resize-decoder
Research-0089: Upstream sync coverage — 2026-05-08¶
0089-upstream-sync-coverage-2026-05-08
Research-0089: vmaf-tune Phase F.5 — recipe calibration on K150K (2026-05-09)¶
0089-vmaf-tune-phase-f5-calibration-2026-05-09
Research-0089: vmaf-tune Apple ProRes VideoToolbox codec adapter¶
0089-vmaf-tune-prores-videotoolbox-adapter
Research-0089: Vulkan VIF API-1.4 NVIDIA residual — CPU double vs Vulkan float stage bisect¶
0089-vulkan-vif-fp-residual-bisect-2026-05-08
AGENTS.md per-package coverage audit (2026-05-09)¶
0090-agents-md-coverage-audit-2026-05-09
Research-0090 — arXiv tech-note draft: production-flip gates and conformal prediction for VMAF predictors¶
0090-arxiv-techNote-prodflip-conformal-2026-05-09
Research-0090 — KonViD MOS head v1 design¶
Research-0090: LSVQ corpus feasibility for nr_metric_v1¶
Research-0090: MCP runtime v3 SSE transport implementation choice¶
0090-mcp-sse-transport-implementation
Research-0090: Phase-A-promotion audit — 2026-05-08¶
0090-phase-a-promotion-audit-2026-05-08
Research-0090 — Pre-push PR-body deliverables validator¶
0090-pr-body-validator-prepush-hook
Research-0090: Sanitizer matrix test-set scope decision¶
0090-sanitizer-matrix-test-scope
Research-0090: Semgrep OSS warnings audit — 2026-05-09¶
0090-semgrep-warnings-audit-2026-05-09
Research-0090: docs/state.md comprehensive verify-every-row audit, 2026-05-09¶
0090-state-md-row-audit-2026-05-09
Research 0090 — T3-15 GPU coverage long-tail batch 4: actual gap re-audit + per-kernel ordering¶
0090-t3-15-gpu-coverage-long-tail-2026-05-09
Research-0090: Upstream SKIP-bucket triage — 2026-05-08¶
0090-upstream-skip-bucket-triage-2026-05-08
Research-0090: vmaf-tune Phase B target-VMAF bisect — feasibility¶
0090-vmaf-tune-phase-b-bisect-feasibility
Research-0090: Vulkan VIF API-1.4 NVIDIA residual — Phase 3b stronger-fence experiments¶
0090-vulkan-vif-fp-residual-phase-3b-2026-05-09
Research Digest 0091 — CAMBI CUDA integration trade-offs¶
Research digest 0091 — cambi × {CUDA, SYCL, HIP} GPU port planning¶
0091-cambi-gpu-port-planning-2026-05-09
Research-0091: Partial-integration audit — features the AI can't learn from yet¶
0091-partial-integration-audit-2026-05-08
psnr_hvs AVX-512 bench-first audit (2026-05-09)¶
0091-psnr-hvs-avx512-bench-2026-05-09
Research-0091: openai/symphony SPEC review — adoptable shapes for the fork¶
Research digest 0091 — Vulkan per-WG accumulator readback bottleneck and two-level GPU reduction¶
0091-vulkan-gpu-reduction-perf-analysis
Research-0091 — Vulkan readback buffer VMA flag: performance analysis¶
0091-vulkan-readback-alloc-flag-perf-analysis
Research-0091: Waterloo IVC 4K-VQA corpus feasibility for nr_metric_v1¶
0091-waterloo-ivc-4k-corpus-feasibility
Research-0091: YouTube UGC corpus feasibility for nr_metric_v1¶
0091-youtube-ugc-corpus-feasibility
Research-0092: CUDA motion sub-4K performance root cause (2026-05-10)¶
0092-motion-cuda-sub4k-perf-root-cause-2026-05-10
Research Digest 0092 — Multi-Backend Performance Benchmark¶
0092-perf-bench-multi-backend-2026-05-10
Research-0092: Symbol-Visibility Audit — Round-4 Bug-Hunt Finding¶
0092-round4-symbol-visibility-audit
Research-0093: Vulkan GCC 16 -Wreturn-mismatch build-break root-cause¶
0093-vulkan-gcc16-return-mismatch-2026-05-10
Research-0094: integer_motion_v2 flush() dictionary leak — round-7 stability audit¶
0094-motion-v2-flush-dict-leak-round7
Research-0094 — T-PYPSNR-AST-EVAL: str/ast.literal_eval log serialization incompatible with numpy 2.x¶
0094-pypsnr-ast-eval-json-fix-2026-05-10
Research-0095 — ATOM_FEATURES port (3dee9666 + 7209110e) blocker¶
0095-atom-features-port-blocker-2026-05-10
Research-0096 — Feature-extractor deduplication by provided-feature names¶
0096-fex-dedup-by-provided-feature-2026-05-10
Research-0096: K150K GPU driver investigation — CUDA timing and double-write root cause¶
0096-k150k-gpu-driver-investigation-2026-05-10
Research-0097 — MCP describe_worst_frames temporary directory leak¶
0097-mcp-describe-worst-frames-tmpdir-leak-2026-05-10
Research-0097: vmaf_thread_pool_create — unchecked pthread_create and racy n_threads read in destroy¶
0097-thread-pool-pthread-create-unchecked-2026-05-10
Research-0098: Round-9 endianness audit — 10-bit YUV read path on big-endian hosts¶
0098-round9-endianness-audit-2026-05-10
Research Digest 0099 — Tiny-AI VMAF Training: 2024–2026 Distillation and ONNX Runtime Update¶
0099-tiny-ai-netflix-training-update
Research-0100: vmaf-tune / Tiny-AI Scaffold-State Audit¶
0100-vmaf-tune-ai-scaffold-audit-2026-05-14
Training Discovery Synthesis — 2026-05-14¶
0101-training-discovery-synthesis-2026-05-14
Research 0101: vmaf-tune Predictor Real Hardware Retrain (2026-05-14)¶
0101-vmaf-tune-predictor-real-hardware-retrain-2026-05-14
Research 0102: vmaf-tune compare bisect CLI closure¶
0102-vmaf-tune-compare-bisect-cli-closure-2026-05-14
Research 0103 — vmaf-tune Usage-Doc Stub Label Sweep¶
0103-vmaf-tune-usage-doc-stub-label-sweep-2026-05-14
Research 0104 — vmaf-tune fast --time-budget-s Scaffold Closure¶
0104-vmaf-tune-fast-time-budget-scaffold-2026-05-14
Research 0105 — vmaf-tune tune-per-shot Bisect Closure¶
0105-vmaf-tune-per-shot-bisect-cli-closure-2026-05-14
Research 0106 — Public Doc Stub Sweep¶
0106-doc-stub-sweep-2026-05-14
Research-0106: vmaf-tune corpus benchmark gap check¶
0106-vmaf-tune-corpus-benchmark
Research 0107: vmaf-tune x264 two-pass gap¶
0107-vmaf-tune-x264-two-pass-2026-05-14
Research 0108 — KonViD-150k Split Score Layout¶
0108-konvid-150k-split-score-layout-2026-05-14
Research-0108: vmaf-roi-score saliency materialiser¶
0108-vmaf-roi-score-materialiser-2026-05-14
Research-0108: Vulkan VIF manual int64 subgroup reduction¶
0108-vulkan-vif-int64-subgroup-reduction-2026-05-14
Research-0108: Y-FUNQUE+ wavelet-domain atom features — feasibility + constants¶
Research-0109: tiny-AI real-weight limitation cleanup¶
0109-tiny-ai-real-weight-doc-limitations-2026-05-15
Research-0109: vmaf-tune auto winner selection¶
0109-vmaf-tune-auto-winner-selection-2026-05-14
Research 0109: vmaf-tune Bisect Sample-Clip Wiring¶
0109-vmaf-tune-bisect-sample-clip-2026-05-14
Research-0110: ADM AVX2 direct-LUT UBSan fix¶
0110-adm-avx2-clz-ubsan-2026-05-15
Research-0110: testdata bench_perf portability¶
0110-testdata-bench-perf-portability-2026-05-14
Research-0111 — DISTS-Sq Extractor Implementation Notes¶
0111-dists-sq-extractor-2026-05-14
Research-0112 — Public Doc Gap Batch¶
0112-public-doc-gap-batch-2026-05-14
Research 0113: FR-from-NR CUDA Feature Split¶
0113-fr-from-nr-cuda-feature-split-2026-05-15
Research-0113: Tiny-model directory jail documentation closeout¶
0113-tiny-model-dir-doc-closeout-2026-05-14
Research-0114: Model overview tiny-AI documentation closeout¶
0114-model-overview-tiny-ai-doc-closeout-2026-05-14
Research-0115: fr_regressor_v2 model-card status closeout¶
0115-fr-regressor-v2-card-closeout-2026-05-14
Research-0116: float_adm AVX2/AVX-512 F2+F3 precision audit¶
0116-float-adm-avx2-512-f2-f3-precision
Research-0116: vmaf-train tune CLI closure¶
0116-vmaf-train-tune-cli-2026-05-14
Research-0117: vmaf-roi-score synthetic mask materialisation¶
0117-roi-score-synthetic-mask-materialisation-2026-05-14
Research-0118: external-bench wrapper schema validation¶
0118-external-bench-wrapper-schema-validation-2026-05-14
Research 0119: ROI-Score High-Bit-Depth Mask Materialisation¶
0119-roi-score-high-bit-depth-mask-2026-05-15
Research 0120: Runtime doc-status audit batch¶
0120-mcp-runtime-doc-status-2026-05-15
Research 0121: MCP compute_vmaf high-bit-depth input support¶
0121-mcp-compute-vmaf-high-bitdepth-2026-05-15
Research-0122 — vmaf-tune sidecar CLI wiring¶
0122-vmaf-tune-sidecar-cli-2026-05-15
Research 0123: vmaf-roi High-Bit-Depth Input¶
0123-vmaf-roi-high-bitdepth-2026-05-15
Research 0124: vmaf-perShot 4:2:2 / 4:4:4 Input¶
0124-vmaf-pershot-422-444-2026-05-15
Research 0125: Tiny-AI RGB Extractors High-Bit-Depth Input¶
0125-tiny-ai-rgb-high-bitdepth-2026-05-15
Research 0126: saliency docs status sweep¶
0126-saliency-doc-status-2026-05-15
Research 0126 — SYCL preallocation API doc status¶
0126-sycl-preallocation-api-doc-status-2026-05-15
Research-0126: vmaf-tune HDR dispatch coverage¶
0126-vmaf-tune-hdr-dispatch-coverage-2026-05-15
Research 0127: Tiny-AI quantisation doc status¶
0127-tiny-ai-quantization-doc-status-2026-05-15
Research 0128: DNN API execution-provider doc status¶
0128-dnn-api-ep-doc-status-2026-05-15
Research 0129: ensemble training kit doc status¶
0129-ensemble-training-kit-doc-status-2026-05-15
Research 0130: CUDA psnr_hvs DCT parallelisation¶
0130-cuda-psnr-hvs-dct-parallel-2026-05-15
CHUG HDR Audit + Content Splits¶
0131-chug-hdr-audit-splits-2026-05-15
CHUG Parquet Metadata Enrichment¶
0132-chug-parquet-metadata-enrichment-2026-05-15
Research 0133: vmaf-tune temporal saliency aggregation¶
0133-vmaf-tune-temporal-saliency-aggregation-2026-05-15
Research 0134: saliency per-block evaluation¶
0134-saliency-per-mb-eval-2026-05-15
Research-0135: CAMBI CUDA spatial-mask shared-memory tile design¶
0135-cambi-cuda-smem-tile-2026-05-16
Research-0135: CHUG extraction I/O cost breakdown and Win 1 + Win 2 optimisations¶
0135-chug-extraction-perf-2026-05-16
Research-0135: CHUG vmaf Column Always NaN — Diagnosis and Removal¶
0135-chug-vmaf-column-always-nan-2026-05-16
Research-0135 — CHUG vmaf column: SDR baseline via vmaf_v0.6.1¶
0135-chug-vmaf-sdr-baseline-2026-05-16
Research-0135: dev-MCP container stage-3 EACCES diagnosis¶
0135-dev-mcp-container-stage-3-fix-2026-05-16
Research: Dispatch-Strategy Registry Audit — 2026-05-15¶
0135-dispatch-strategy-registry-audit-2026-05-15
Research Digest 0135: SSIMULACRA2 CUDA Blur Performance Optimisation¶
0135-ssimulacra2-cuda-perf-2026-05-16
Research Digest 0135: SYCL CAMBI/SSIM/CIEDE GPU-Sync and SLM Performance Analysis¶
0135-sycl-cambi-ssim-ciede-perf-2026-05-16
Research Digest 0135 — VIF CPU Performance: Workspace Hoist + Fast Log2 Audit¶
0135-vif-cpu-perf-workspace-hoist-2026-05-16
Research 0135: VIF CUDA shared-memory staging (wins #1 and #4)¶
0135-vif-cuda-smem-staging-2026-05-16
Research Digest 0135 — Vulkan Dispatch Overhead Characterization¶
0135-vulkan-dispatch-overhead-2026-05-15
Research-0136: CAMBI SYCL GPU Queue-Drain Overhead (SY-1)¶
0136-cambi-sycl-event-chain-sy1-2026-05-16
Research-0136: HDR/UGC Dataset License Audit (2026-05-15)¶
0136-hdr-ugc-dataset-license-audit-2026-05-15
Research-0136: PSNR enable_chroma cross-backend silent divergence — 2026-05-16¶
0136-psnr-enable-chroma-cross-backend-2026-05-16
Research-0137: MS-SSIM GPU option parity — enable_db, clip_db, enable_lcs¶
0137-ms-ssim-gpu-option-parity-enable-db-lcs-2026-05-16
Research-0138: dev-MCP container backend exposure (SYCL / Vulkan / HIP)¶
0138-dev-mcp-container-backend-exposure-2026-05-18
Metal kernel batch T8-1c through T8-1k — reduction strategy and numerical stability¶
0421-metal-kernel-batch-t8-1c-k
Research-0431: Drop ssimulacra2 from CHUG/K150K self-vs-self extraction¶
0431-chug-drop-ssimulacra2-self-vs-self-extraction
Research-0432: HIP (AMD ROCm) backend applicability¶
Research-0433 — CI pipeline audit (2026-05-01)¶
0433-ci-pipeline-audit-2026-05
Research-0434: VIF log2 LUT Shrink and Gaussian Filter Cache¶
0434-vif-lut-shrink-and-filter-cache
Research-0435: ADM decouple gather locality (2026-05-18)¶
0435-adm-decouple-gather-locality
Research-0436: Vulkan VIF fp32 vs fp64 Cross-Vendor Empirical Study¶
0436-vulkan-vif-fp32-cross-vendor-empirical
Research-0480: Bootstrap Name Owner Restoration¶
0480-bootstrap-name-owner-restoration
Research-0513: tiny-model loader and the rank-2 / external-data gap¶
0518-tiny-model-loader-feature-rank
Research-0519: HIP vmaf_hip_import_state library-side wiring¶
Research-0519: --tiny-codec / --tiny-preset / --tiny-crf CLI surface¶
0522-tiny-codec-preset-crf-cli-flags
Research-0530: compare rate-quality chart from bisect samples¶
0534-compare-rate-quality-chart-from-bisect-samples
Research digest: HIP integer-VIF kernel fault analysis (ADR-0537)¶
0537-hip-integer-vif-kernel-fix
Research-0537: premium-archival VMAF target defaults + bisect reaches VMAF 95+¶
0537-premium-vmaf-target-defaults-and-bisect
Research digest: dev-MCP container SYCL+HIP runtime UAPI pin (ADR-0541)¶
0541-dev-container-sycl-hip-runtime-fix
Research-0541: dev-mcp container full GPU plumbing — root cause and fix matrix¶
0542-dev-container-full-gpu-plumbing
0546 — Codec-adapter two_pass_args real implementations¶
0546-codec-adapter-two-pass-real
Research digest 0548 — vmaf-tune ergonomics: container sources + CRF sweep mode¶
0548-vmaf-tune-full-file-and-no-bisect
Research-0550 — Cross-Backend Parity Matrix (2026-05-18)¶
0550-cross-backend-parity-matrix-2026-05-18
Research-0551: VCQ-223 LocalExplainer CI timeout — root cause investigation¶
Research digest: ADR-0552 — HIP VIF deterministic wavefront reduction¶
0552-hip-vif-deterministic-reduce
Research digest: ADR-0561 — HIP gfx_targets fallback widening¶
0561-hip-gfx-targets-fallback-widening
Research digest — HIP extractor audit (9 remaining scaffold claims) (ADR-0563)¶
0563-hip-extractor-audit-verification
Research digest: Real integer_ssim GPU kernels (ADR-0564)¶
0564-integer-ssim-gpu-real-kernels
Research-0567: SpEED GPU Kernel Design — CPU/GPU Split and Backend Implementation Choices¶
Research-0576: ffmpeg-patches surface-gap audit (n8.1.1 + feature-exposure)¶
0576-ffmpeg-patches-feature-gap-audit
Research digest: macOS SIGSEGV deep-fix in output.c writers (ADR-0606)¶
0606-macos-vmaf-write-output-segv-deep-fix
Research Digest 0609: Dynamic Optimizer (DO)¶
0609-dynamic-optimizer-research
Research Digest 0610: Per-Shot ABR Rendition Selection¶
0610-per-shot-abr-rendition-research
Research Digest 0611: Fast NR Pre-Scoring¶
0611-fast-nr-prescoring-research
Research Digest 0607 — Tiny-AI Netflix corpus training: 2024–2026 literature refresh¶
0612-tiny-ai-netflix-training-scaffold-2026-05-19
Research Digest 0612: VMAF NEG Integration¶
0612-vmaf-neg-integration-research
Research Digest 0613: Cross-Shot Complexity Weighting¶
0613-cross-shot-complexity-weighting-research
Research Digest 0614: Content-Aware Classifier¶
0614-content-aware-classifier-research
Research Digest 0615 — Tiny-AI Netflix corpus training: 2026-05-20 literature refresh¶
0615-tiny-ai-netflix-training-2026-05-20
Research: Dev-container encoder probe hardening¶
0641-dev-container-encoder-probe-hardening
Research: AI full-feature refresh defaults¶
0642-ai-refresh-full-feature-defaults
Research-0643: vmaf-tune Encoder Profile Reports¶
0643-vmaf-tune-encoder-profile-contract
Research: vmaf-tune codec runtime variants¶
0644-vmaf-tune-codec-runtime-variants
Research-0645: Integer ADM p-norm SIMD callback audit¶
Research Digest: Attached DNN Multi-Output Routing¶
0646-dnn-attached-multi-output
Research-0647: fr_regressor_v1 refresh on the 2026-05-20 Netflix table¶
0647-ai-fr-regressor-v1-refresh-20260520
Research 0648: CHUG HDR MOS Trainer Surface¶
0648-mos-head-feature-jsonl-chug-id
Research 0649: CHUG HDR Wide MOS Feature Schema¶
0649-chug-hdr-wide-mos-feature-schema
Research Digest: Signal-Mix Audit¶
Research-0651: CHUG HDR Row Metadata¶
Research-0652: CHUG Visual-Signal Primitives¶
0652-chug-visual-signal-primitives
Research-0653: CHUG Display Profile Training¶
0653-chug-display-profile-training
Research-0654: Predictor Saliency Signal Wiring¶
0654-predictor-saliency-signals
Research-0655: Saliency Feature Materializer¶
0655-saliency-feature-materializer
Research-0656: external-bench fork wrapper schema mismatch¶
0656-external-bench-wrapper-schema
Research Digest: Second-Opinion Feature Materializer¶
0657-second-opinion-feature-materializer
Research-0658: Project modernization audit¶
0658-project-modernization-audit
Research-0659: Modernization audit false-positive filter¶
0659-modernization-audit-false-positive-filter
Research-0660: Tiny-AI disabled-runtime gate¶
0660-tiny-ai-disabled-runtime-gate
Research-0661: AI run manifest provenance¶
0661-ai-run-manifest-provenance
Research 0662: Vulkan Motion Lavapipe Parity¶
0662-vulkan-motion-lavapipe-parity
Research-0663: MOS Label Materializer¶
Research-0664: AI Model Sidecar Run Provenance¶
0664-ai-fr-regressor-run-provenance
Research-0664: Windows CUDA toolkit installer failure¶
0664-windows-cuda-toolkit-installer
Research-0665: AI Eval / Validate Report Run Provenance¶
0665-ai-eval-report-run-provenance
Research-0666: Legacy AI Eval Report Run Provenance¶
0666-ai-legacy-eval-report-provenance
Research-0667: Predictor v2 Report Run Provenance¶
0667-predictor-v2-report-provenance
Research-0668: vmaf_tiny Train Stats Run Provenance¶
0668-vmaf-tiny-train-stats-provenance
Research-0669: AI Materializer Audit Provenance¶
0669-ai-materializer-audit-provenance
Research-0670: Ensemble Seed Export Provenance¶
0670-ensemble-seed-export-provenance
Research-0671: Ensemble LOSO Report Provenance¶
0671-ensemble-loso-report-provenance
Research-0672: vmaf-train CLI Report Provenance¶
0672-vmaf-train-cli-report-provenance
Research 0673 — Feature-correlation report provenance¶
0673-feature-correlation-report-provenance
Research 0674 — Phase-3 subset-sweep report provenance¶
0674-phase3-subset-report-provenance
Research-0675: Per-EP quantisation report provenance¶
0675-quant-ep-report-provenance
Research 0676 — Phase F calibration provenance¶
0676-phase-f-calibration-provenance
Research 0677 — NR threshold calibration provenance¶
0677-nr-threshold-calibration-provenance
Research 0678 — Ensemble manifest provenance¶
0678-ensemble-manifest-provenance
Research 0679: DNN exporter manifest provenance¶
0679-dnn-exporter-manifest-provenance
Research 0680: Saliency-student metrics provenance¶
0680-saliency-student-metrics-provenance
Research 0681: tiny-VMAF validator report provenance¶
0681-vmaf-tiny-validator-report-provenance
Research 0682: quantisation report provenance¶
0682-quantization-report-provenance
Research 0683: AI validation report provenance¶
0683-ai-validation-report-provenance
Research 0684: CHUG extraction report provenance¶
0684-chug-extraction-report-provenance
Research 0685: Fast-NR calibration quality guard¶
0685-fast-nr-calibration-quality-guard
Research-0685: Modernization Audit Contract-Noise Follow-Up¶
0685-modernization-audit-contract-noise
Research 0686: vmaf-tune report quick takeaways¶
0686-tune-report-quick-takeaways
Research Digest 0687 — vmaf-tune score backend native priority¶
0687-vmaf-tune-score-backend-native-priority
Research Digest 0688 — AI derived table provenance¶
0688-ai-derived-table-provenance
Research-0688: HIP AMD Raphael iGPU divergence — carry-loss in int64 wavefront reduction¶
0688-hip-raphael-igpu-divergence
Research Digest 0689 — AI corpus JSONL provenance¶
0689-ai-corpus-jsonl-provenance
Research Digest 0690 — AI legacy corpus extraction manifests¶
0690-ai-legacy-corpus-extraction-manifests
Research-0691 — U2NetP mirror exporter¶
Research-0692: Saliency Materializer Temporal Controls¶
0692-saliency-materializer-temporal-controls
Research-0693: Saliency Materializer Batch Manifest¶
0693-saliency-materializer-batch-manifest
Research-0694: Second-Opinion Materializer Batch Manifest¶
0694-second-opinion-materializer-batch-manifest
Research-0695: MOS Label Materializer Batch Manifest¶
0695-mos-label-materializer-batch-manifest
Research-0696: Full-Feature Exporter Manifests¶
0696-full-feature-exporter-manifests
Research-0697: MOS Corpus Adapter Manifests¶
0697-mos-corpus-adapter-manifests
Research-0698: AI Dataset Fetch Manifests¶
0698-ai-dataset-fetch-manifests
Research-0699: AI Run Manifest Helper¶
CI Draft Auto-Merge Gate¶
Research-0699: VMAFX Helm chart and Kubernetes GPU scheduling (2026-05-28)¶
0699-vmafx-helm-chart-k8s-2026-05-28
Research 0700: AI CLI Helper Pattern¶
Research 0701: AI Script Bootstrap Helper¶
0701-ai-script-bootstrap-helper
Research 0702: AI Materializer Bootstrap Sweep¶
0702-ai-materializer-bootstrap-sweep
Research 0703: AI Materializer CLI Helper Sweep¶
0703-ai-materializer-cli-helper-sweep
Research 0704: AI Corpus Adapter CLI Helper Sweep¶
0704-ai-corpus-adapter-cli-helper-sweep
Research 0705: VMAF Tiny Script Bootstrap Sweep¶
0705-ai-vmaf-tiny-cli-bootstrap-sweep
Research 0706: VMAF Tiny Eval Script Bootstrap Sweep¶
0706-ai-vmaf-tiny-eval-cli-bootstrap-sweep
Research Digest 0706 — Tiny-AI Netflix Training Prep: 2026-05-22 Literature Update¶
0706-tiny-ai-netflix-training-prep-2026-05-22
Research 0707: AI Quant/Export Script Bootstrap Sweep¶
0707-ai-quant-export-cli-bootstrap-sweep
Research: Rust/cbindgen Integration for libvmaf Feature Extractors (ADR-0707)¶
Research 0708: AI Dataset-Prep CLI Bootstrap Sweep¶
0708-ai-dataset-prep-cli-bootstrap-sweep
Research 0709: AI Model Export/Validator CLI Bootstrap Sweep¶
0709-ai-model-export-validator-cli-bootstrap-sweep
AI full-feature CLI bootstrap sweep¶
0710-ai-full-feature-cli-bootstrap-sweep
AI evaluation and calibration CLI bootstrap sweep¶
0711-ai-eval-calibration-cli-bootstrap-sweep
Research 0712: AI Training CLI Bootstrap Sweep¶
0712-ai-training-cli-bootstrap-sweep
Research 0713: CHUG HDR CLI Bootstrap Sweep¶
0713-ai-chug-cli-bootstrap-sweep
Research-0714 — vmaf-tune report output helper¶
0714-vmaf-tune-report-output-helper
Research 0715: CHUG HDR MOS Head — Held-Out Test Partition Results (2026-05-27)¶
0715-chug-hdr-held-out-test-validation-2026-05-27
Research-0715 — vmaf-tune strict JSON helper¶
Research-0716 — vmaf-tune strict JSON artifact emitters¶
0716-vmaf-tune-jsonio-artifact-emitters
Research-0717 — vmaf-tune CLI strict JSON helper¶
0717-vmaf-tune-cli-jsonio-helper
Research-0718 — vmaf-tune sidecar strict JSON persistence¶
0718-vmaf-tune-sidecar-jsonio-helper
Research-0719 — vmaf-tune cache strict JSON metadata¶
0719-vmaf-tune-cache-jsonio-helper
Research-0720 — vmaf-tune executor strict JSONL results¶
0720-vmaf-tune-executor-jsonio-helper
Research-0721 — AI run manifest strict JSON¶
0721-ai-run-manifest-strict-json
Research-0722 — tool report strict JSON¶
Research-0723 — AI registry strict JSON helper¶
Research-0724 — AI strict JSON stdout helper¶
Research-0725 — AI strict JSONL row helper¶
Research-0726 — AI legacy strict JSON writers¶
0726-ai-legacy-json-writer-helper
Research-0727 — LPIPS exporter helper alignment¶
Research 0728: Train / Aggregate Scripts Bootstrap Sweep¶
0728-ai-train-konvid-aggregate-helpers
Research-0729 — Zed 1.3.6 config refresh + agent_servers wiring¶
Research-0730: Cross-backend numerical parity — Intel Arc A380 (2026-05-27)¶
0730-cross-backend-arc-parity-20260527
Research digest 0730: ffmpeg libvmaf filter end-to-end smoke test (2026-05-27)¶
0730-ffmpeg-libvmaf-smoke-20260527
Research Digest 0730 — Tiny-AI Netflix Training Prep: 2026-05-27 Literature Update¶
0730-tiny-ai-netflix-training-prep-2026-05-27
Research-0731 — VMAFx Org Migration + Monorepo Split Audit¶
Research Digest 0732: Netflix Pipeline Functions Backlog Audit¶
0732-netflix-pipeline-backlog-audit
Research-0732: VMAFX C++23 Internals Migration Plan¶
0732-vmafx-cpp23-internals-migration-plan
Research-0733 — Feature Importance Audit and Drop Recommendations (2026-05-28)¶
0733-feature-importance-audit-2026-05-28
Research 0733 — Hardware GPU backend audit and drop recommendation (2026-05-28)¶
0733-hardware-backend-audit-2026-05-28
Research-0733 — VMAFX eBPF Optimization Target¶
0733-vmafx-ebpf-optimization-target
Research-0733 — VMAFX Sidecar Online Training Architecture¶
0733-vmafx-sidecar-training-architecture
Research-0734: Cross-Backend Throughput Baseline + SYCL on Intel Arc (2026-05-28)¶
0734-cross-backend-baseline-with-sycl-20260528
Research-0734 — CUDA 13.3 Fix-List Deep Audit: Per-Issue Exposure Mapping¶
0734-cuda-13.3-fix-list-deep-audit
Research-0734: CUDA 13.3 Non-Blog Findings — Bug Fixes and APIs the Developer Blog Skipped¶
0734-cuda-13.3-non-blog-findings
Research-0734: CUDA ADM kernel ncu hotpath analysis (576x324)¶
Research-0734: CUDA __mul24 silent-corruption sweep (CUDA 13.3 fix surface)¶
0734-cuda-mul24-corruption-audit
Research-0734: CUDA VIF filter1d ncu Hotpath Profile (2026-05-28)¶
0734-cuda-vif-filter1d-ncu-hotpath-20260528
Research digest 0734 — cuDNN version audit for ONNX Runtime / tiny-AI inference¶
0734-cudnn-version-audit-20260528
Research Digest 0734: NVIDIA R610.43.02 Driver Changelog Audit¶
0734-r610-driver-changelog-audit-2026-05-28
Research-0734 — SYCL Backend Audit (2026-05-29)¶
0734-sycl-backend-audit-20260529
Research-0735 — CI Required-Check Failures Round 3 (2026-05-28)¶
0735-ci-required-failures-round-3-2026-05-28
Research-0735: CUDA motion kernel ncu hotpath analysis (576x324)¶
Research-0736: CUDA SSIM kernel ncu hotpath analysis (576x324)¶
Research-0737: CUDA MS-SSIM kernel ncu hotpath analysis (576x324)¶
Research-0738: CUDA remaining kernels ncu hotpath summary (adm, motion, ssim, ms_ssim)¶
0738-cuda-other-kernels-ncu-summary
Research-0744 — Cross-Backend Parity Baseline (Pre-ncu-opt)¶
0744-cuda-cross-backend-baseline-pre-ncu-perf
Research-0748: CUDA VIF filter1d 1080p re-measurement (PR #76 production-workload validation)¶
0748-cuda-vif-filter1d-1080p-remeasure
Research-0749: Hardware Measurement of PR perf/cuda-ms-ssim-decimate-adm-cm-ncu-driven-20260528¶
0749-cuda-ms-ssim-decimate-adm-cm-1080p-measure
Research-0751: Cross-backend 4K Baseline + PR #79 adm_cm A/B at 4K¶
0751-cross-backend-4k-baseline-and-pr79-adm-cm-4k-measure
Research-0753: CUDA Resolution-Aware Dispatch — Design Rationale¶
0753-cuda-resolution-aware-dispatch-design
Research-0754 — CUDA SSIM vert_combine: __ldg() + __launch_bounds__ + pinned-host leak¶
0754-cuda-ssim-vert-combine-ldg-launch-bounds-2026-05-29
Research-0755: PR #91 576p NO_BOUNDS dispatch validation — A/B at SMALL workload class¶
0755-cuda-pr91-576p-no-bounds-dispatch-validation-20260529
Research-0755 — HIP Backend Audit (2026-05-29)¶
0755-hip-backend-audit-20260529
Research-0760: CUDA motion kernel ncu multi-resolution profile (2026-05-29)¶
0760-cuda-motion-ncu-multi-resolution-20260529
Research-0761 — Metal Backend Audit (2026-05-29)¶
0761-metal-backend-audit-20260529
Research-0762: Public-header ISO-reserved include-guards — Round 27 audit A.1 (2026-05-31)¶
0762-public-header-iso-reserved-guards-2026-05-31
Research-0763 — CUDA adm_decouple F3 __ldg() fix¶
0763-cuda-adm-decouple-ldg-f3-fix
Research-0764: psnr_hvs CUDA kernel — F3 __ldg() + __launch_bounds__(64) (2026-05-29)¶
0764-cuda-psnr-hvs-ldg-launch-bounds-2026-05-29
Research-0773: CUDA ADM decouple-inline — __ldg() F3 fix (active path)¶
0773-cuda-adm-decouple-inline-ldg-f3-fix
IWYU (include-what-you-use) Audit — Fork-Added C/C++ Code¶
Research-0870: Helm values.schema.json + dev-MCP Containerfile rebuild audit¶
0870-helm-values-schema-and-container-rebuild-audit
Research-0876: printf-format portability audit (CERT FIO47-C / MISRA 21.6)¶
0876-printf-format-portability-audit
Research-0877: Error-code consistency audit — fork-added C surfaces¶
0877-error-code-consistency-audit-2026-05-30
Research digest — Trivy container scan baseline (2026-05-30)¶
0878-trivy-container-scan-baseline
Coverage-overrides audit — 2026-05-30¶
0881-coverage-overrides-audit-2026-05-30
Research-0887: vmaf_model_destroy heap-buffer-overflow from per-feature array length mismatch¶
0887-vmaf-model-slopes-feature-mismatch-validation
0889 — Vendored libsvm 3.24 audit (2026-05-30)¶
0889-libsvm-vendored-audit-2026-05-30
Research-0892: Conventional-Commits + changelog-fragment hygiene audit¶
0892-conventional-commits-audit-2026-05-30
Research digest — Pre-commit config audit (2026-05-30)¶
0893-pre-commit-audit-20260530
Research 0905 — .gitignore + .github/workflows/ staleness audit (2026-05-30)¶
0905-gitignore-and-workflow-audit
Research 0907 — Perf regression gate audit (2026-05-30)¶
0907-perf-regression-gate-audit-20260530
Research-0912: Pixel-format edge coverage audit¶
0912-pixel-format-edge-coverage
Research-0913 — CHANGELOG.md renderer splice bug + 23 k+ line drift audit¶
0913-changelog-renderer-and-drift-2026-05-31
Research-0914: Python test orchestrator audit (2026-05-31)¶
0914-python-test-orchestrator-audit-2026-05-31
Research-0922: Coverage ratchet sizing + per-PR delta gate design¶
0922-coverage-ratchet-aggressive
Research-0923: BuildKit cache mounts + ccache rollout (2026-05-31)¶
0923-buildkit-cache-mounts-rollout-2026-05-31
Research-0924 — Native bash pre-commit hook (opt-in)¶
Research digest — ADR-0930: Helm NetworkPolicy + Pod Security Standards baseline¶
Research digest 0935 — Go errors.Join cleanup-path audit + slog key standardisation (2026-05-31)¶
0935-go-errors-join-slog-audit
Research digest — SYCL kernel coverage round 3 (ADR-0946, 2026-05-31)¶
0946-sycl-kernel-coverage-round3-2026-05-31
GitHub Actions custom-action + reusable-workflow audit — 2026-05-31¶
0951-github-actions-custom-audit
Research digest — Vendored libsvm + IQA coverage uplift (ADR-0952)¶
0952-vendored-libsvm-iqa-coverage-2026-05-31
Research digest — SYCL kernel coverage round 4 (ADR-0957, 2026-05-31)¶
0957-sycl-kernel-coverage-round4-2026-05-31
HIP kernel parity coverage — round 4 audit (2026-05-31)¶
0958-hip-kernel-coverage-round4-2026-05-31
Research digest — Metal kernel coverage round 4 (closeout)¶
0959-metal-kernel-coverage-round4-closeout
Research digest: GPU runtime error-path leak fixes (round-25 audit)¶
0960-gpu-runtime-error-path-leaks-2026-05-31
Research digest: PullWork post-update rollback (round-25 audit B.1)¶
0961-queue-pullwork-rollback-2026-05-31
Research digest — Controller infrastructure fixes (ADR-0962)¶
0962-controller-streamjobs-reaper-fixes-2026-05-31
Research: NaN Propagation Guards in ai/src/vmaf_train — Round-25 Audit¶
0963-ai-nan-propagation-guards-2026-05-31
Research-0964 — speed_internal.c missing TU, cross-backend wiring gap¶
0964-speed-internal-cross-backend-2026-05-31
Research digest: CUDA SpEED TU repair (ADR-0965, 2026-05-31)¶
0965-cuda-speed-tu-repair-2026-05-31
Research-0966 — dev/Containerfile libvmaf → core rename (Round 26 audit C.1)¶
0966-dev-containerfile-libvmaf-rename-2026-05-31
Research digest: MCP HTTP transport security hardening (ADR-0967, 2026-05-31)¶
0967-mcp-http-transport-security-2026-05-31
Research digest — CI scripts Round 26 audit (D.1 + D.2)¶
0968-ci-scripts-rebrand-tempfile-2026-05-31
Research Digest — Helm Chart PSA B.1 + B.3 Fixes (ADR-0969)¶
0969-helm-seccomp-node-image-2026-05-31
Research digest: test_gpu_picture_pool.c Round 27 cleanup (D.3 + D.4)¶
0970-test-gpu-picture-pool-cleanup-2026-05-31
Research digest: Test suite unchecked malloc sweep (Round 27 audit D.1)¶
0971-test-unchecked-malloc-sweep-2026-05-31
Research-0973: Master CI regressions — verified reproduction and root-cause analysis¶
0973-master-ci-regressions-verified-2026-05-31
Python-surfaces bug-audit research digest¶
0983-python-surfaces-bug-audit-2026-05-31
Research-0985: SYCL parity divergence — float_ssim + ssimulacra2 + float_ansnr (2026-06-03)¶
0985-sycl-parity-divergence-2026-06-03
Research-0993: KoNViD / UGC / BVI-DVC Saliency Batch Launch Investigation¶
0993-konvid-ugc-bvi-saliency-batch-launch
Research-1052: ARM motion_v2 re-registration bisect¶
1052-arm-motion-v2-re-register
Research digest: Helm chart security hardening (ADR-1058)¶
1058-helm-chart-security-hardening
Research digest — ADR-1061: vendored pdjson/cJSON depth-limit and overflow fixes¶
1061-vendored-cjson-pdjson-depth-overflow
Research: SYCL -fsycl link-propagation and feature-name aliasing¶
1099-sycl-fsycl-link-propagation
Research-1100: ΔE-ITP (Delta E ITP) — HDR/WCG colour-difference metric feasibility¶
Research digest 1101 — BRISQUE no-reference CPU extractor¶
Research-1101: Vendoring the Pelorus interop ABI into vmafx¶
1101-vendor-pelorus-interop-abi
Research-1102: Pelorus side-data perceptual weighting of VMAF pooling¶
1102-perceptual-sidedata-weighting
Research-1116: vmaf-tune autotune prefilter control plane (Pelorus deband)¶
1116-autotune-prefilter-control-plane
Research digest: MCP tiny-AI / feature / CTC coverage (2026-06-14)¶
1117-mcp-tiny-ai-feature-coverage
Research digest 1119 — Adopting the golusoris fx framework across vmafx Go¶
1119-golusoris-go-framework-adoption
GPU SpEED — covariance geometry + eigenvalue-basis correctness (2026-06-20)¶
1120-gpu-speed-covariance-eigenbasis-correctness-2026-06-20
Research-1127: Single SemVer release stream¶
1127-single-semver-release-stream
Research-1128: Fragment-owned release cuts¶
1128-fragment-owned-release-cuts
Research-1129: Release container runtime alignment¶
1129-release-container-runtime-alignment
Research: modernize-use-nullptr on C23 translation units, and the c-rework-core lint pass¶
1138-c23-nullptr-msvc-upstream-parity
Research digest: integer ADM upstream-mirror rework (ADR-1141)¶
1141-integer-adm-upstream-mirror-rework
Research-1147: HIP Backend Gap Resolution and AMD iGPU Parity¶
1147-hip-backend-gap-resolution
VMAFx/vmafx — release-please / release-pipeline audit¶
1151-release-please-audit-2026-09-02
Research digest 1152 — which paths do Renovate PRs actually touch?¶
1152-dependency-classifier-surface-audit
Research 1166 — upstream Netflix/vmaf issue harvest, 2026-09-03¶
1166-upstream-issue-harvest-2026-09-03
Research Digest: Dev Container Release Artifact Compilation and Runner Architecture¶
1178-dev-container-image-publish
Research-1182: Windows UTF-8 Path Contract Investigation & Design¶
1182-windows-utf8-path-contract
Research digest: MCP sidecar-binary and gRPC control-plane bridge (2026-09-05)¶
1184-mcp-sidecar-and-grpc-bridge
Research 1185 — refreshing the per-backend performance baselines (2026-09-06)¶
1185-backend-perf-baselines-2026-09-06
Research-1188 — percentile temporal pooling: what the two surfaces actually compute¶
1188-percentile-pooling-methods
Research 1193 — the PSNR psnr_max ceiling truncates real scores¶
1193-psnr-cap-truncation-2026-09-06
Research: how ROCm 10 is distributed, and what it changed under /opt/rocm¶
1225-rocm-10-distribution-and-layout
Research digest — base-image single source, and what the oneAPI 2026 move actually requires¶
Research Digest: Version Single-Source Tree & Python Dependency Unification¶
1236-version-single-source-tree
Research-1238: Predictor card G304 and missing Go merge gate¶
1238-go-security-required-gate
Agent cleanup preservation¶
1239-agent-cleanup-preservation
Research-1241: Local hook wiring and worktree lifetime¶
Research-1242: Generated ADR metadata audit¶
Research-1245: Cppcheck branch budgets and exhaustive analysis¶
1245-cppcheck-exhaustive-configured-analysis
Cppcheck public entrypoint model¶
1246-cppcheck-public-entrypoints
Research 1277 — local workspace contract cleanup¶
1277-workingdir-contract-cleanup
Research-1280: Worktree-safe private-state synchronization¶
1291 — What the silent-revert gate was actually reporting on the zero-warning train¶
1291-silent-revert-gate-false-positives
Research-1292: Bounded FIFO producer startup wait¶
1292-fifo-bounded-startup-wait-2026-09-23
Backlog tracker schema migration evidence — 2026-09-23¶
1303-backlog-tracker-schema-migration-2026-09-23
Research-1305¶
1305-python-dependency-hash-lock-audit
Research-1306: dropping nvidia/cuda base images in favor of version-locked apt installs¶
Research-1311: Source ADR citation provenance audit¶
1311-source-adr-citation-provenance-audit
Research-1314: Semgrep registry SARIF routing¶
1314-semgrep-registry-sarif-routing
Research-1317: Netflix Golden Gate Build Isolation and Compiler FP Contraction Drift¶
1317-golden-gate-build-isolation
Research-1318: Bounded vmaf-perShot scan loop with operator frame ceiling¶
1318-pershot-endless-input-ceiling-2026-09-25
Research-1333: Meson test environment secret credential sanitization¶
1333-meson-test-secret-env-sanitization
Research-1338: Go fix clean-tree gate¶
Research-1341: RC correctness, benchmark, and retraining sequence¶
1341-rc-correctness-benchmark-retrain-sequence
Research-1342: RC1 external tester readiness and report evidence¶
1342-rc1-external-tester-report-bundle
Research-1346: Building native release artifacts on a hosted runner inside the build-deps stage¶
1346-hosted-slim-container-release-build
Research-1354: Building the native Linux bundle on the Debian 13 release track¶
1354-native-bundle-release-track
Research-1356: Release provenance under the organisation's SHA-pinning policy¶
1356-release-provenance-attest
Research-1358: Device-resident SpEED on SYCL — what the device must do to match the CPU bit for bit¶
1358-sycl-speed-device-resident
Research-1362: The AIM pass on the SYCL integer ADM twin — what it takes to match the CPU bit for bit¶
Research-1363: Device-resident ssimulacra2 and a single-wait MS-SSIM on SYCL¶
1363-sycl-ssimulacra2-msssim-device-resident
Research-1366: Frame read-ahead in the vmaf CLI¶
Research-1367: One strict FP line for every SYCL feature TU — what it guarantees, what it changes, what it costs¶
1367-sycl-strict-fp-every-feature-tu
Research-1369: SYCL psnr_hvs, motion_v2 and psnr at 4K — where the frame time went and how the twins now share the uploaded planes¶
1369-sycl-shared-planes-light-twins
Research-1371: motion_sycl tiny-frame parity — blur order, not edges¶
1371-sycl-motion-diff-first-pipeline
Research-1372: CUDA RC3 parity port — motion order, CPU options, tiny-frame guards¶
Research-1377: HIP RC3 CPU parity — motion order, tiny-frame reads, CPU options¶
Research-1378: HIP CAMBI and SpEED on the device — rounding, replays, what holds without a device¶
1378-hip-cambi-speed-device-resident
Research-1379: Device-resident CAMBI and SpEED on CUDA — the CUDA rounding contract, the CPU's libm, and verification by emulation and on an RTX 4090¶
1379-cuda-cambi-speed-device-resident
Research-1383: Replaying real docs/state.md rebase conflicts through the three-way resolver¶
1383-state-md-three-way-conflict-replay
Research-1391: Device-resident ssimulacra2 on CUDA¶
1391-cuda-ssimulacra2-device-resident
Research-1395: Scratch memory in the SYCL kernels on an Arc A380 under the xe driver¶
Research-1397: Making a GPU psnr_hvs twin return the CPU's bits — what differs, what it costs, and how to tune it¶
1397-psnr-hvs-twins-cpu-float-sum
Research-1399: float_ssim decimation and CPU-typed convolution on CUDA¶
1399-cuda-float-ssim-device-decimation
Research-1400: integer SSIM on identical small frames — where the CPU's sum is not exactly 1¶
1400-hip-integer-ssim-raster-sum
Research-1401: Bit-exact psnr_hvs on SYCL and HIP — the CPU's masking threshold without fp64, measurements on an Arc A380 and a gfx1036, and the cost¶
1401-psnr-hvs-sycl-hip-exact-twins
Research-1402: Removing the int16 wrap of the integer ADM masking centre tap¶
Research-1403: FMA contraction off for every CUDA kernel — which kernels change, what each twin does against the CPU, what it costs¶
1403-cuda-strict-fp-every-kernel
Research-1405: float_ssim decimation on the HIP device¶
1405-hip-float-ssim-device-decimation
Research-1407: contraction off in every HIP kernel — parity and cost per twin on a gfx1036¶
1407-hip-strict-fp-every-kernel
Research-1408: one upload of each frame plane for all HIP twins — output identity, throughput and the rejected upload variants on a gfx1036¶
Research-1412: Why float_vif_cuda was not the CPU's float_vif — four causes, their sizes, and the cost of removing them¶
1412-cuda-float-vif-cpu-arithmetic
Research-1413: The integer ADM enhancement gain limit under a non-integer limit¶
1413-adm-gain-limit-truncated-double-product
Research-1416: Why adm_cuda was not the CPU's adm — a copied CSF weight routine, a per-warp fold and an fp32 shift¶
1416-cuda-adm-cpu-row-rounding
Research-1417: Integer AIM above 1 on a flat reference¶
1417-integer-aim-unclipped-upstream-parity
Research-1420: Why float_adm_cuda was not the CPU's float_adm — nine causes, their sizes, and a division that belongs to the host processor¶
1420-cuda-float-adm-cpu-arithmetic
Research-1422: float_vif on SYCL without fp64 — the four differences on an Arc A380, a pair that is almost enough, and the kernel shapes measured¶
1422-sycl-float-vif-fp64-free-statistic
Research-1424: integer_ssim_cuda and the CPU's frame sum — what an exact sum costs, and how it could be made parallel¶
Research-1426: Why ciede_cuda was 1e-5 from the CPU — fp32 where the CPU is fp64, and what is left once that is fixed¶
1426-cuda-ciede-cpu-arithmetic
Research-1430: Which speed_chroma_cuda outputs differ from a glibc CPU, and that log2f is the only reason¶
1430-cuda-speed-chroma-log2f-bound
Research-1433: Forming the CPU's sequential double sums on a GPU, bit for bit¶
1433-cuda-ssimulacra2-cpu-sum-order
Research-1434: float_adm on SYCL without fp64 — the differences on an Arc A380 one at a time, and three fp64 expressions as fp32 pairs¶
1434-sycl-float-adm-fp64-free-arithmetic
Research-1436: ciede2000 on SYCL without fp64 — each fp32 piece of the old twin on its own, elementary functions on fp32 pairs, and where the last pixels come from¶
Research-1437: Which HIP twins return the CPU extractor's bits — a sweep of every gate feature on a gfx1036¶
Research-1442: What changes when float ADM divides instead of using the processor's reciprocal estimate¶
1442-float-adm-reference-divides
Research-1443: fixed-point ssim on SYCL without fp64 — where the fp32 twin differed, the fp64 term in 64-bit integers, and what each part costs on an Arc A380¶
Research-1446: SSIMULACRA 2 on SYCL without fp64 — the CPU's fp64 terms in integers, the CPU's sequential sums from parallel pieces, and what a slow single lane does to the design¶
1446-sycl-ssimulacra2-cpu-bits
Research-1457: Which CUDA twins return the CPU extractor's bits — a sweep of every gate feature on an RTX 4090, stress content and small frames included¶
1457-cuda-twin-exactness-sweep
Research-1472: Which term of the integer ADM contrast-masking reduction wraps in Barten mode, and what weight keeps it in range¶
1472-integer-adm-cm-weight-budget
Research-1475: the fork's integer ADM against Netflix master, before and after the quantisation step went back to float¶
1475-integer-adm-quant-step-upstream-float
Research-1477: SpEED against Netflix master, and what a GPU twin has to do to return an fp64 log2¶
1477-speed-upstream-double-math
Research-1487: Where the fork's CPU extractors differ from Netflix/vmaf, and why¶
Research-1489: the fork's float ADM against Netflix master, before and after its CSF weights went back to float¶
1489-float-adm-barten-upstream-float
Research-1498: Metal Shading Language floating-point semantics for exact twins¶
1498-metal-shading-language-fp-semantics
Research-1590: GPU device code compression¶
Research-1900: Verification and contract analysis for VPL decode retry ceiling¶
1900-vpl-decode-ceiling-verification-2026-10-05
SYCL QSV Zero-Copy — Two-Bug Investigation (Phase 3 fix digest)¶
2026-06-30-sycl-qsv-zerocopy-two-bug-investigation
VMAF v1 Models — Research Digest¶
Research digest — c-rework-vif-motion: bit-exact rework of the scalar VIF and float-motion TUs (2026-09-02)¶
2026-09-02-c-rework-vif-motion-bit-exact
Research digest — the shared CUDA/SYCL CAMBI parity drift (2026-09-05)¶
2026-09-05-gpu-cambi-parity-drift
Research: Intel Arc SYCL Default Model Crashes and Parity Fixes¶
2026-09-05-sycl-v1-model-crash-digest
Research digest — Netflix benchmark re-run on cd52f2670 (2026-09-06)¶
2026-09-06-netflix-benchmark-rerun
Research: oneAPI 2025.3.2 production builder compatibility¶
20260831-oneapi-2025-3-2-builder-runtime-compatibility
Lint / CI carve-out inventory (2026-09-02)¶
2027-lint-carveout-inventory-2026-09-02
Code Scanning Alert Sweep & Security Dismissal Re-Audit (2026-09-03)¶
2028-code-scanning-audit-2026-09-03
Research-2029: Tiny-AI Int8 Static PTQ and QAT Readiness Smoke for Epic #1246¶
2029-int8-static-qat-readiness
Research-2030: an fp64-free, bit-identical integer-ADM angle_flag¶
Research digest: AI teacher single source, table provenance, and combiner refusal¶
Research digest 2030 — CPU hot path under the default model: SpEED matrix_mul and CAMBI¶
2030-speed-matmul-and-cambi-cpu-hot-path
Research-2031 — cpp/integer-multiplication-cast-to-long on convolve, moment, and PSNR¶
2031-codeql-float-widening-multiplication
Research-2032: Finding two GPU parity bugs by varying one constant (2026-09-06)¶
2032-gpu-parity-resolution-blind-spot
2033 — Options whose default is an arithmetic identity are untested by default-options parity tests¶
2033-identity-default-option-blind-spot
Research-2033: Testing the property instead of the proxy (2026-09-06)¶
2034 — Kernel-local constants that shadow an option name defeat both review and the type system¶
2034-kernel-local-constants-shadowing-options
2035 — Designing a fixture that can actually observe a numerical-degeneracy bug¶
2035-singular-covariance-fixture-design
Research digest 2036 — Benchmark tuning pass: CAMBI AVX2 anti-dithering, SpEED SIMD QR, and HIP threaded pipeline¶
2036-benchmark-tuning-pass-cambi-speed-hip
2036 — Three CAMBI divergences behind one vacuous fixture¶
2036-cambi-twin-drift-and-vacuous-fixtures
2037 — Advertised-but-unimplemented options on the float-ADM GPU twins¶
2037-advertised-but-unimplemented-gpu-options
2038 — When a twin implements the right-sounding thing: clip_db¶
2039 — Why the code-scanning alerts never clear¶
2039-why-code-scanning-alerts-never-clear
2040 — Code Scanning Alert Audit and Cleanup for Issue #1243¶
2040-code-scanning-cleanup-1243
Research: CUDA Tile C++ has no additive reduction, and VIF is made of them¶
2040-cuda-tile-additive-reduction-gap
Research-2041: Thread-pool backpressure and shutdown lifetime¶
Integer ADM SIMD native lint cleanup — 2026-09-08¶
2042-adm-simd-native-lint-2026-09-08
Research-2043: CAMBI production lint without numerical changes¶
2043-cambi-production-lint-2026-09-08
Research-2044: Feature-extractor test read-only inputs¶
2044-feature-extractor-test-const-2026-09-08
Research-2045: integer VIF AVX2 stages — 2026-09-08¶
2045-integer-vif-avx2-stages-2026-09-08
Research-2046: Integer VIF AVX-512 stage cleanup¶
2046-integer-vif-avx512-stage-lint
Research-2047: Option-aware feature context registration¶
2047-option-aware-context-registration-2026-09-08
Research-2048: Feature registration option-copy ownership¶
2048-model-registration-ownership-2026-09-08
Research-2049: Preserve observation-only SVM tests while cleaning native lint¶
2049-svm-observation-test-lint-2026-09-08
Research-2050: SpEED test native lint cleanup¶
Research-2051: CAMBI AVX2 parameter shadowing¶
2051-cambi-avx2-parameter-names-2026-09-08
Research-2052: DNN test const inputs and fixture I/O errors¶
Research-2053: Read-only IQA and motion coverage fixtures¶
2053-observation-fixture-const-2026-09-08
Research-2054: Preserve motion, SSIM and PSNR coverage during lint cleanup¶
2054-metric-coverage-const-2026-09-08
Research-2055: OpenSSF passing evidence and truthful support policy¶
2055-best-practices-passing-evidence
Research-2056 — Fedora Dockerfile heredoc repair¶
Research-2057: Repository security enforcement¶
2057-repository-security-enforcement
Research-2058 — One verified envtest tool version¶
2059 — Why no dependency bump can merge¶
2059-why-no-dependency-bump-can-merge
2060 — Three gates that only fail in combination¶
2060-the-matrix-found-what-one-lane-hid
2061 — The address that was not an address¶
2061-win64-cannot-realign-the-stack
Research-2062: CAMBI spatial-mask row SIMD — what each ISA actually gains¶
Research-2063: September 2026 upstream sync — ADM / VIF SIMD fixes and checkasm¶
2063-upstream-sync-2026-09-adm-vif-simd
2064 — Praetor adoption: measured behaviour and the two exclusions¶
2064-praetor-adoption-measurements
Research-2065: CAMBI SIMD gaps — AVX-512 and NEON for every stage AVX2 accelerates¶
2066 — Windows on ARM64: what the runner offers, which toolchain, what broke¶
Research-2070: Intel NEO release fetch fail-closed audit¶
2070-intel-neo-fetch-fail-closed
Research-2071: Repository subprocess boundary audit¶
2071-bounded-process-execution
Research-2072: Pelorus v0.2.2 interop parser safety sync — 2026-09-20¶
2072-pelorus-interop-v022-sync-2026-09-20
Research-2073: FFmpeg n9.0.2 stable refresh¶
2073-ffmpeg-n9-0-2-stable-refresh
Research-2074: HISS-21 replay-evidence gap¶
Research-2075: High-signal non-CUDA Cppcheck cleanup¶
2075-cppcheck-high-signal-noncuda
Research-2076: CUDA ADM sign-bit rounding without conversion warnings¶
Research-2077: Go duplicate implementation cleanup¶
2077-go-duplicate-implementation-cleanup
Research-2078: MSVC-compatible strict-FP flags — 2026-09-23¶
2078-msvc-compatible-strict-fp-flags-2026-09-23
Research-2079: SYCL Clang-Tidy Required Gate Verification and Hardening¶
2079-sycl-clang-tidy-required-gate-verification
Research-2080 — required-workflow path-filter closure¶
2080-rust-ci-path-filter-coverage
Research-2081: Float moment Metal exact reduction restoration¶
2081-float-moment-metal-exact-reduction-restoration
Research-2082: SYCL upload host-buffer lifetime¶
2082-sycl-upload-host-lifetime
Research-2083: dev-MCP smoke-probe contract restoration¶
2083-dev-mcp-smoke-probe-contract-restoration
Research-2084: dev-MCP resilience restoration¶
2084-dev-mcp-resilience-restoration
Research-2085: Strict tool JSON restoration¶
2085-strict-tool-json-restoration
Research-2086: BUG-048 strict JSON emitter restoration¶
2086-bug048-strict-json-emitter-restoration
Research-2087: GPU public-header Doxygen contracts survived in code, not prose¶
2087-gpu-public-header-doxygen-contracts
Research-2088: BUG-048 AI CLI Helper Restoration¶
2088-bug048-ai-cli-helper-restoration
Research-2089: Archived issue-reference provenance after the repository migration¶
2089-archived-issue-reference-provenance
Research-2090: SYCL silent-revert residuals — 2026-09-24¶
2090-sycl-silent-revert-residuals-2026-09-24
Research-2091: restoring silent-reverted test seams¶
2091-bug048-test-contract-restoration
Research-2092: Pre-push type-check baseline evaluates branch checker configuration¶
2092-mypy-pre-push-config-baseline
Research-2093: Resolution of CodeQL cpp/include-non-header alerts¶
2093-codeql-include-non-header-alerts
Research-2094: CodeQL SVM solver lifecycle and parser loop alerts¶
2094-codeql-svm-lifecycle-and-loop-alerts
Research-2095: Semgrep Warning Alerts 946–949 Audit and Resolution¶
2095-semgrep-warning-alerts-946-949-audit
Research-2096: CodeQL unused-static-function identity audit¶
2096-codeql-unused-static-alerts-2026-09-24
Research-2097: Resolving live CodeQL cpp/equality-on-floats alerts — 2026-09-24¶
2097-codeql-equality-on-floats-2026-09-24
Research-2098: Resolution of CodeQL AVX-512 large-parameter alerts¶
2098-vif-avx512-large-parameter-codeql
Research-2099: Non-finite score laundering in the metric engine¶
2099-nonfinite-score-laundering-2026-09-23
Research-2100: feature-collector source authority — 2026-09-24¶
2100-feature-collector-source-authority-2026-09-24
Research-2101 — BUG-048 section E SYCL init-unwind restoration (2026-09-24)¶
2101-bug048-sycl-init-unwind-restoration-2026-09-24
Research-2102: SYCL Clang-Tidy Wrapper Executable Path Resolution for Safe Subprocess¶
2102-sycl-tidy-wrapper-safe-subprocess-path
Research-2103: required mypy gate restoration — 2026-09-25¶
2103-mypy-ci-fail-closed-2026-09-25
Research-2104: GPU option-alias parity — 2026-09-25¶
2104-gpu-option-alias-parity-2026-09-25
Research-2105: GPU option-value capability fallback — 2026-09-25¶
2105-gpu-option-value-capability-fallback-2026-09-25
Research-2106: CUDA and HIP device target header dependency tracking — 2026-09-25¶
2106-cuda-hip-kernel-header-dependency-tracking
Research-2107: Pre-RC1 scheduled-CI flake triage — 2026-09-25¶
2107-pre-rc1-scheduled-ci-flake-triage-2026-09-25
Research-2108: GPU float-SSIM auto-scale fallback — 2026-09-25¶
2108-gpu-float-ssim-auto-scale-fallback-2026-09-25
Research-2109: Integer ADM Barten fixed-point normalization — 2026-09-25¶
2109-integer-adm-barten-fixed-point-normalization
Research-2110: Metal float_ms_ssim option and score parity — 2026-09-25¶
2110-metal-ms-ssim-option-parity-2026-09-25
Research-2111: Integer-ADM row-rounding observability — 2026-09-25¶
2111-adm-cm-row-rounding-observability
Research-2112: SYCL motion-add-UV fixed-point oracle — 2026-09-25¶
2112-sycl-motion-uv-fixed-oracle-2026-09-25
Research-2113: Metal float-motion lifecycle, debug gating, and flush regression — 2026-09-25¶
2113-metal-float-motion-lifecycle-flush
Research-2114: trusted authority for the research-digest identity ratchet¶
2114-research-digest-ratchet-authority
Research-2115: HIP float-motion lifecycle and flush regression — 2026-09-24¶
2115-hip-float-motion-lifecycle-flush
Research-2116: CodeQL Python alerts triage and exception semantics¶
2116-codeql-python-alerts-triage-2026-09-24
Research-2117: CUDA context-owned resource teardown¶
2117-cuda-context-owned-resource-teardown
Research-2118 — BUG-048 script environment drift¶
2118-bug048-script-environment-drift-2026-09-25
Research-2119: GCC 16 C++ Placement New/Delete Symbol Visibility — 2026-09-25¶
2119-gcc16-placement-new-symbol-visibility
Research-2120: SYCL ciede2000 throughput — 2026-09-29¶
Research-2121: --feature with an explicit GPU --backend — twin pairing and receipt¶
Research-2122: Device-resident CAMBI on SYCL — c-values and exact top-K pooling on the GPU¶
2122-sycl-cambi-device-resident
Research-2123: Arc B580 SYCL crashes — psnr_hvs compile, tile-halo faults¶
2123-sycl-b580-psnr-hvs-and-tile-halo-faults
Research-2124: SYCL AOT images dropped at the link — 2026-09-29¶
2124-sycl-aot-images-dropped-at-link
Research-2125: First native Windows run of the SYCL build on Arc B580 and UHD 770 — 2026-09-29¶
Research-2127: CPU options on the SYCL PSNR, SSIM and float-motion twins — 2026-09-29¶
2127-sycl-twin-cpu-option-parity
Research-2128: oneAPI release image — Arc B580 crash and the Debian 13 rebuild — 2026-09-29¶
2128-oneapi-release-image-runtime
Research-2130: float_ssim decimation on the SYCL device¶
2130-sycl-float-ssim-device-decimation
Research-2131: who frees the vmaf CLI's option dictionaries¶
2131-cli-option-dictionary-ownership
Research-2132: CAMBI on frames shorter or narrower than its window¶
2132-cambi-short-and-narrow-frames
Research-2133: Licence audit of the published tester image and macOS bundle¶
2133-tester-artifact-licence-audit
Research-2136: what an Intel GPU tester image needs at run time, on Linux and under WSL2¶
Research-2137: Documentation site generator, charts, diagrams and readability¶
2137-docs-site-toolchain-charts-diagrams
Research-2138: what an NVIDIA GPU tester image needs at run time, and what it may carry¶
Research-2139: what an AMD GPU tester image needs at run time, and the source its LGPL parts owe¶
Research-2140: Licence audit of the published production artifacts¶
2140-production-artifact-licence-audit
Research-2141: Windows tester zip¶
Research-2142: Package compression and what each consumer opens¶
2142-package-compression-consumers
Research-2143: Root licence files, package licence fields, and where LICENSE-MIT was published¶
2143-root-licence-files-and-package-manifests
Research-2144: integer_adm_metal on the first Apple report, six defects found on the host¶
2144-metal-integer-adm-twin-defects
Research-2151: What the RC4 Rust extractor framework has to match¶
2151-rc4-rust-extractor-framework
Research-2158: VMAFx API redesign, generated surfaces and the VMAFx FFmpeg filters¶
Research-2159: What the HIP runtime of a gfx1036 does with imported frames, fences and GL textures¶
Research-2161: What an EGL dma-buf export of a GL texture is on a gfx1036, and how HIP can read it¶
Research-2170: Windows icx-cl and icpx strict FP spelling — 2026-10-07¶
2170-windows-strict-fp-spelling-2026-10-07
Research-2351: components of the cloud-native platform, their versions, licences and limits¶
2351-cloud-native-platform-components
Research-2399: Packaging the generated rules and dashboards in the Helm chart¶
2399-observability-helm-packaging
Research-2479: what replacing the C and C++ host code with Rust takes¶
ADR-0108 Six-Deliverables Compliance Audit — 2026-05-29¶
adr-0108-compliance-audit-2026-05-29
Research: ADR status-field drift audit (2026-05-30)¶
adr-status-drift-audit-2026-05-30
ARM64 NEON bit-exactness audit — 2026-05-30¶
arm64-neon-bit-exactness-audit-2026-05-30
Research digest: bash strict-mode + trap-cleanup audit (2026-05-30)¶
bash-strict-mode-audit-2026-05-30
BUG-003: REUSE coverage and provenance audit¶
bug-003-reuse-provenance-audit-2026-09-24
BUG-048 backend invariant restoration¶
bug048-backend-invariant-restoration-2026-09-25
BUG-048 — Zed 1.18.1 project-workflow restoration¶
bug048-zed-1-18-restoration-2026-09-24
CAMBI native test cleanup¶
Research digest — cargo-deny baseline scan (2026-05-31)¶
cargo-deny-baseline-2026-05-31
Research — cargo-machete unused-dependency audit (2026-05-30)¶
cargo-machete-audit-2026-05-30
CI fail-closed audit (2026-09-20)¶
ci-fail-closed-audit-2026-09-20
Research: CI Warning Omnibus (2026-05-19)¶
Research digest: clang-tidy modernize-* sweep (2026-05-31)¶
clang-tidy-modernize-sweep-2026-05-31
Codex hook-path portability audit (2026-09-25)¶
codex-hook-path-portability-2026-09-25
Coverage audit — compat/python-vmaf/ leaf utilities¶
compat-python-vmaf-coverage-audit-2026-05-31
Research digest: the vmaf shim deadlock in multiprocessing spawn children¶
compat-shim-spawn-recursion-2026-09-22
Configured clang-tidy warning exit audit¶
configured-lint-warning-fail-closed-2026-09-21
Configured native lint selection — 2026-09-08¶
configured-native-lint-2026-09-08
Const-correctness audit — fork-added C / C++ — 2026-05-30¶
const-correctness-audit-2026-05-30
Continuous Feature-Mix Evaluation Pipeline — Design Specification¶
continuous-feature-mix-evaluation-design-2026-05-18
Horizontal SIMD convolution boundary repair¶
convolution-horizontal-boundary-2026-09-08
Core HISS control-flow cleanup (2026-09-21)¶
core-hiss-control-flow-cleanup-2026-09-21
Core MCP transport coverage push — 2026-05-31¶
core-mcp-coverage-push-2026-05-31
core/test HISS-21 burn-down (2026-09-21)¶
core-test-hiss21-burndown-2026-09-21
Research: Adversarial code review — C++23 conversion wave (PRs #41–#58)¶
cpp23-wave-adversarial-review-20260528
Cppcheck pthread model correction (2026-09-08)¶
cppcheck-pthread-model-2026-09-08
CUDA-kernel parity-test coverage — round 2 audit (2026-05-30)¶
cuda-kernel-coverage-round2-2026-05-30
Research Digest — CUDA kernel parity coverage round 3 (2026-05-31)¶
cuda-kernel-coverage-round3-2026-05-31
Research Digest — CUDA kernel parity coverage round 4 (2026-05-31)¶
cuda-kernel-coverage-round4-2026-05-31
Research: User-input dataclass → pydantic v2 migration triage — 2026-05-31¶
dataclass-to-pydantic-configs-2026-05-31
Dependency audit 2026-05-30¶
Research digest: doxygen public-API clean (2026-05-31)¶
doxygen-public-api-clean-2026-05-31
Research digest: doxygen public-API warnings to zero and fail closed (2026-09-25)¶
doxygen-public-api-fail-closed-2026-09-25
Kubernetes E2E runtime-contract audit — 2026-08-31¶
e2e-k8s-runtime-contract-2026-08-31
Research: POSIX I/O EINTR + return-value audit (2026-05-30)¶
eintr-and-io-error-audit-2026-05-30
Research: Feature Coverage Audit — 2026-05-18¶
feature-coverage-audit-2026-05-18
Feature-extractor coverage round 2 — research digest¶
feature-extractor-coverage-round2-2026-05-31
Feature-extractor coverage round 3 — research digest¶
feature-extractor-coverage-round3-2026-05-31
Stable feature-context pool entries¶
FFmpeg release patch lifecycle evidence¶
ffmpeg-release-patch-lifecycle
Git fixture initialization inside linked-worktree hooks¶
git-fixture-hook-environment-2026-09-08
Research digest: GitHub Actions hardening audit (2026-05-30)¶
github-actions-audit-2026-05-30
Research digest: Go static-analysis audit (2026-05-30)¶
go-nilness-staticcheck-audit-2026-05-30
Go orchestrator HISS cleanup — 2026-09-21¶
go-orchestrator-hiss-cleanup-2026-09-21
Research digest: gosec sweep — per-finding fix rationale¶
gosec-findings-fix-sweep-2026-06-01
Research digest: governance file audit — 2026-05-30¶
GPU backend kernel coverage audit — 2026-05-30¶
gpu-backend-kernel-coverage-audit-2026-05-30
Research: fail-closed self-hosted GPU runner admission¶
gpu-runner-admission-2026-09-25
Research digest — a HIP init unwind ladder that reported success (T-HIP-INIT-UNWIND-REPORTS-SUCCESS-2026-09-22)¶
hip-init-unwind-reports-success-2026-09-22
Research digest — HIP integer_moment HSACO blob registration (ADR-0539)¶
hip-integer-moment-registration
HIP kernel parity coverage — round 2 audit (2026-05-30)¶
hip-kernel-coverage-round2-2026-05-30
HIP kernel parity coverage — round 3 audit (2026-05-31)¶
hip-kernel-coverage-round3-2026-05-31
Research digest — __init__.py export-completeness audit (2026-05-31)¶
init-py-export-audit-2026-05-31
K150K extractor crash-restart row-loss RCA (2026-05-30)¶
k150k-crash-restart-row-loss-rca-2026-05-30
Research digest — LLVM IR diff harness for bit-exact SIMD paths¶
llvm-ir-diff-harness-2026-05-31
Research: Logging consistency audit in fork-added C/C++ code¶
logging-consistency-audit-2026-05-30
Magic-number audit (fork-added C surfaces) — 2026-05-30¶
Research digest: MCP server audit (2026-05-29)¶
Merge-train control investigation, 2026-09-08¶
merge-train-control-2026-09-08
Research: mkdocs nav restructure for 631 ADRs (2026-05-31)¶
mkdocs-nav-restructure-2026-05-31
Research digest: AVX-512 motion parity test coverage gap (2026-05-29)¶
motion-avx512-parity-tests-2026-05-29
Research Digest: Netflix Upstream Feature Additions — CUDA Twin Gap Audit (2026-05-18)¶
netflix-upstream-feature-additions-since-sync-2026-05-18
NOLINT Cluster Audit — 2026-05-29¶
nolint-cluster-audit-2026-05-29
NOLINT inventory audit — ADR-0278 cite-form drift sweep (2026-05-30)¶
nolint-inventory-audit-2026-05-30
Option sentinel cleanup — 2026-09-08¶
option-sentinel-cleanup-2026-09-08
Research digest: OpenTelemetry traces + metrics Phase 1 (2026-05-31)¶
otel-traces-metrics-phase1-2026-05-31
Research digest — parquet schema v2 (zstd-3 + canonical column order + schema metadata)¶
pdjson nesting boundary and parser lint cleanup¶
pdjson-nesting-and-lint-2026-09-08
Perf Snapshot Audit — 2026-05-29¶
perf-snapshot-audit-2026-05-29
Phase 4b Status Digest — 2026-05-29¶
phase4b-status-digest-20260529
Picture pool / framesync lifecycle audit — 2026-05-29¶
picture-pool-framesync-lifecycle-audit-2026-05-29
Research Digest: integer_motion pipelined rename (upstream a4a1492d3)¶
port-a4a1492d3-integer-motion-pipelined-rename
Research Digest: 2160p at 1.5H CSF support (upstream c2155d6cd)¶
Research — pre-existing test failures across ai/, tools/vmaf-tune/, mcp-server/¶
pre-existing-test-failures-2026-05-30
Research: Pre-existing macOS / Python / Tiny-AI CI Failures (2026-06-01)¶
preexisting-macos-python-tinyai-failures-2026-06-01
PSNR format-table analyzer cleanup¶
psnr-format-table-lint-2026-09-08
Pyright strict audit — fork-local Python packages (2026-05-30)¶
pyright-strict-audit-2026-05-30
Python dependency freshness audit — 2026-05-30¶
python-dep-freshness-2026-05-30
Python harness coverage push round 2 (2026-05-31)¶
python-harness-coverage-push-2026-05-31
Research digest: Python / MCP / AI stub-and-silent-fallback audit¶
python-mcp-ai-audit-2026-05-18
Research: Python type-annotations audit (2026-05-30)¶
python-type-annotations-audit-20260530
Research: Python warning root causes (2026-09-21)¶
python-warning-root-causes-2026-09-21
Release Preview — v3.1.0-lusoris.0¶
release-preview-3.1.0-lusoris.0
Research: what Renovate can and cannot rewrite for the CUDA pin¶
Research: Renovate custom manager file-pattern delimiters¶
renovate-file-pattern-delimiters
Research-0734: CUDA 13.3 impact assessment for the VMAFX fork¶
research-0734-cuda-13.3-impact-assessment
Research-0743: CUDA VIF filter1d ncu-driven performance implementation¶
research-0743-cuda-vif-filter1d-perf-impl
Research-0744: CUDA ms_ssim_decimate smem tiling + adm_cm_line_kernel_8 register reduction¶
research-0744-cuda-ms-ssim-adm-cm-perf-impl
Research-0747 — CUDA extern "C" name-mangling sweep¶
research-0747-cuda-extern-c-sweep
Research-0749: Sunset VmafLegacyQualityRunner¶
research-0749-sunset-legacy-vmaf-feature-extractor
Research-0752 — Multi-Resolution Performance Benchmark Baseline¶
research-0752-perf-bench-multi-resolution-baseline
Research-0752: Phase 4b.8 — libvmaf C ABI Break Scoping¶
research-0752-phase-4b8-c-abi-break-scoping
Research-0756: CUDA F3 Struct-by-Value Kernel Audit¶
research-0756-cuda-f3-struct-by-value-audit
Research-0757 — C++23 Feasibility: feature_collector + feature_extractor¶
research-0757-feature-collector-extractor-cpp23-feasibility
Research-0759: HIP ADM AdmBufferHip by-pointer refactor¶
research-0759-hip-adm-buffer-by-pointer
Research-0760: Rust Crate Audit — TAD Feature Extractor and vmafx-sys Bindings¶
research-0760-rust-crate-audit
Research-0775: DNN ORT Backend Audit¶
research-0775-dnn-ort-backend-audit
Research-0787 — libvmaf Public API Error-Path Consistency Audit¶
research-0787-libvmaf-api-error-path-audit
Research-0848: Per-Surface Documentation Compliance Audit — Last 30 PRs (2026-05-29)¶
research-0848-per-surface-doc-compliance-audit-20260529
Research-0932: iter.Seq[T] adapter audit for Go packages¶
research-0932-iter-seq-adapter
HIP Backend Completeness Audit — 2026-06-03¶
research-0933-hip-completeness-audit-2026-06-03
Research-0991: Second-Opinion Batch Materializer — Smoke-Run Scaffold¶
research-0991-second-opinion-batch-2026-06-03
Research-1284: detecting merges that silently revert the target¶
research-1284-silent-revert-detection
ROCm 26.04 pin restoration — 2026-09-08¶
rocm-2604-restoration-2026-09-08
Research: ROCm Version Audit (2026-05-19)¶
ROI reader and placeholder boundary audit (2026-09-08)¶
Research: Local Sanitizer Audit — 2026-05-30¶
Signing and attestation audit — 2026-05-30¶
signing-and-attestation-audit-2026-05-30
Research digest: SIMD twin coverage inventory (2026-05-29)¶
simd-twin-inventory-2026-05-29
Research digest — Skills library expansion (2026-05-31)¶
skills-library-expansion-2026-05-31
Research digest — slow-test audit (2026-05-30)¶
SYCL integer VIF warning cleanup — 2026-09-21¶
sycl-integer-vif-warning-cleanup-2026-09-21
SYCL kernel coverage round 2 — gap audit (2026-05-30)¶
sycl-kernel-coverage-round2-2026-05-30
SYCL strict-diagnostics cleanup — 2026-09-21¶
sycl-strict-diagnostics-2026-09-21
SYCL Zero-Copy NaN Diagnosis — Phase 1 Research Digest¶
Tensor I/O test lint cleanup (2026-09-08)¶
tensor-io-test-cleanup-2026-09-08
Thread-Safety Audit: CUDA / SYCL / HIP Backends¶
thread-safety-audit-backends-2026-05-29
Research: reproducing the ADR-1142 clang-tidy ratchet off the CI runner¶
tidy-ratchet-host-reproduction-2026-09-22
Research digest: migrating tiny-AI QAT to torchao pt2e¶
tinyai-qat-pt2e-migration-2026-09-22
TSan race audit — libvmaf threadpool + dispatcher paths (2026-05-30)¶
Floating-point VIF native lint repair — 2026-09-08¶
Research: vmaf_mcp_stop() double-join SIGSEGV¶
vmaf-mcp-stop-double-join-segv-2026-05-31
Copyright 2026 Lusoris¶
vmaftune-compare-real-smoke-20260527
Research — vmaf-tune profile-report renderer audit (T-VMAFTUNE-PROFILE-REPORT-AUDIT)¶
vmaftune-profile-report-audit-2026-05-27
Research: vmaf-dev-mcp Container Smoke Test — 2026-05-27¶
vmafx-container-smoke-20260527
Research digest — VMAFx copyright policy: drop "and Claude (Anthropic)"¶
vmafx-copyright-policy-drop-anthropic
Research digest: vmafx-mcp Go port (2026-05-28)¶
pytest suite audit — ai/, tools/vmaf-tune/, mcp-server/vmaf-mcp/ (2026-05-27)¶
vmafx-pytest-suite-audit-20260527