Research window: 2026-10-08 06:00 ~ 2026-10-09 06:00 (24 hours) Sources: GitHub (org: flagos-ai, full audit of pushed_at across 54 repos, full verification of 75 commit searches, merged and open PR search verification, atomic cross-check of default branches across 9 repos and all FlagQuantum branches, atomic audit of Releases/tags across 24 repos), Google News RSS (24 sets of Chinese and English query terms, via proxy), HN Algolia, official CSDN column and BAAI Community verification (see appendix source list for details)

Index for This Issue

  • Today’s Highlights: Post-holiday volume surge day — 16 consecutive competition Track1 kernel merges and 63 new commits merged (10-08 ~ 10-09)
  • I. Open Source Project Progress (GitHub Activity)
    • 1.1 FlagGems-sglang: 16 competition Track1 kernels merged in a concentrated batch — bulk landing in 32 minutes (10-08)
    • 1.2 FlagGems: 19 merges — KernelGen operators, Kunlunxin fix batch, and infrastructure wrap-up (10-08)
    • 1.3 KernelGen pipeline: 8 Nvidia batch merges, Ascend batch under review (10-08)
    • 1.4 FlagAttention: 7 attention operator merges in a batch (10-08)
    • 1.5 FlagTree: 7 merges covering AMD / HCU / PPU / mthreads and new backend thrive (10-08)
    • 1.6 FlagQuantum: CI fix achieved, QBoson provider PR opened (10-08)
    • 1.7 Other activity: FlagCX / vllm-plugin-FL / FlagGems-vllm / docs / FlagFFT (10-08 ~ 10-09)
  • II. News Coverage and Ecosystem
    • 2.1 Two ecosystem-side items: Tsingmicro expands to Shanghai, Enflame post-IPO analysis (10-08)
    • 2.2 Search and official channel verification: component lines remain quiet, no new CSDN posts (10-08 ~ 10-09)
  • III. Member Company Deep Dive
    • 3.1 Kunlunxin: fix batch merged, six new PRs in succession, and P800 vLLM 0.24 landing (10-08)
    • 3.2 Enflame: GCU fp8 fix merged and post-IPO attention (10-08)
    • 3.3 Hygon: vLLM 0.28 adaptation begins, Hygon Device API CI and HCU fixes (10-08)
    • 3.4 Moore Threads / MetaX: benchmark and operator lines in parallel (10-08)
    • 3.5 DAMO XuanTie / Ascend: compiler and CI line activity (10-08)
    • 3.6 Other members: Tsingmicro / Iluvatar CoreX / Cambricon / Horizon (within window)
  • IV. Summary and Trend Observations
  • Appendix: Source Verification Table
  • Complete Source List

Today’s Highlights: Post-Holiday Volume Surge — 16 Competition Track1 Kernels Merged in a Row and 63 New Merges (10-08 ~ 10-09)

Date: 2026-10-08 ~ 2026-10-09　Sources: FlagGems-sglang #101, #133, FlagGems #6887, FlagQuantum #598, etc. (see appendix for full list)

On the second working day after the holiday, the organization saw a full-scale volume surge: 11 of 54 repositories had pushes within the window, commit searches returned 75 commits, PR merge searches returned 69 merges (of which 63 were later than the previous issue’s cutoff and are new to this issue), and another 151 open PRs were updated within the window; there were no new Releases and no new tags during the window. Three main threads emerged throughout the day:

  • Batch entry of competition results: 16 “FlagOS Competition-Track1” kernels in FlagGems-sglang were merged in a concentrated burst within 32 minutes from 16:23 to 16:55 on 10-08 — covering quantized matrix multiplication, fused norms, linear attention components, speculative sampling, and convolution updates; in addition, 7 attention variants in FlagAttention and 11 in the KernelGen series (3 in FlagGems + 8 in Experimental) were merged in batches (see 1.1 / 1.3 / 1.4 for details).
  • Parallel multi-vendor fixes and adaptations: Kunlunxin had 7 fix merges in a single day, followed by six new PRs in quick succession in the evening (see 3.1 for details); Enflame GCU fp8 crash fixes, Hygon HCU fixes, Ascend KMCompiler linear optimization, and DAMO XuanTie PPU CI updates each had merges; the vLLM 0.28.0 multi-vendor adaptation wave (DAMO XuanTie PPU / Moore Threads / Hygon / Ascend) kicked off in a concentrated fashion on the same afternoon (see 1.7 and 3.x for details).
  • Quantum line expands to multiple providers: FlagQuantum merged two CI fixes and opened PRs for the QBoson (Boson Quantum “Kaiwu”) provider and QDiffusion integration — the quantum stack is extending from “release certification” to “multi-provider onboarding” (see 1.6 for details).

Assessment: This issue marks the first full volume window after the National Day holiday, with both merge and review metrics returning to high levels; moreover, the merge pattern shows a clear “batch pipeline” characteristic (sglang 16 merges / 32 minutes, Kunlunxin five in a row / 3 minutes, FlagAttention 7 merges / 12 minutes), signaling that all lines have entered the execution phase of “review clearance + concentrated merging,” in contrast to the pre-holiday rhythm of “quiet window + single-point breakthroughs.”

I. Open-Source Project Progress (GitHub Activity)

1.1 FlagGems-sglang: 16 Competition Track1 Kernels Merged in a Batch — 32-Minute Bulk Landing (10-08)

Date: 2026-10-08　Source: #101 · #133 and 16 total (see appendix)

  • Merge cadence: 16 “FlagOS Competition-Track1” Triton kernel PRs (numbered #88–#133) were merged into the master branch in a concentrated 32-minute window from 16:23 to 16:55 on 10-08 — a one-shot bulk consolidation of competition submissions into the operator library.
  • Contributors: c2flowDS (8), xuanzhengdu-eng (3), liangkw16 (2), and yzw1128, hangglider5, William7743 (1 each) — the competition channel is funneling new contributors into the library.
  • Kernel coverage (16 items): w8a8_block_int8_matmul (blockwise INT8 quantized matmul), log_scaling_tau, hc_head, fused_norm_rope_stacked, fused_gdn_gating, fused_dual_residual_rmsnorm, fla_layernorm_gated, extend_attention, ernie45_rope_fused, chunked_sgmv_shrink, chunked_sgmv_expand,
    chunked_embedding_lora_a, chunk_scaled_dot_kkt, chain_speculative_sampling, causal_conv1d_update, l2norm — covering quantization, fused normalization, linear attention (GDN / DeltaNet components), speculative sampling, and convolution updates across the SGLang inference pipeline.
  • Notes: Most PRs entered the review queue in late September (created September 19–30) and were merged in this concentrated batch; Track1 competition results have entered the “bulk landing” phase.

1.2 FlagGems: 19 Merges — KernelGen Operators, Kunlunxin Fix Batch, and Infrastructure Consolidation (10-08)

Date: 2026-10-08　Source: #5304 · #6369 · #6887 · #6904 and 19 total (see appendix)

  • KernelGen (Nvidia) operators, 3: linalg_lu_solve (#5304), batch_norm_update_stats (#5754), foreach unary operators (#5730, shared multi-tensor executor implementation).
  • Kunlunxin fix batch, 7: native_group_norm switched to a tle.gpu (cluster DMA) single fused kernel (#6369); TE-series fixes (dswiglu #6753, te_rmsnorm_fwd #6768, geglu #6781, hc skip vllm #6791); greater_scalar performance (#6721); reflection fix and rollback (#6767).
  • Ascend, 1: linear operator optimization in the KMCompiler generation path (#6887).
  • Infrastructure and quality, 8: PPU CI updated to flagtree 0.7.0 and switched to the torch image (#6883), torch_gcu (Enflame GCU) fp8 conversion crash fix (#6904), Enflame CI environment variable fix (#5453), pytest-marker-based test scanning and location (#6891), C++ runtime .so relative path fallback (#4958), pure-Triton HIP libdevice fallback for isfinite (#5841), OP test log prefix (#6854), second batch of explicit operator invocation tests (#6877).
  • Under review: six new Kunlunxin PRs (#6896–#6901: log_softmax_out / softmax_out fusion, mkldnn_rnn_layer LSTM, linalg_matrix_norm, nonzero_numpy, thnn_fused_lstm_cell_backward, etc.); the XPU3 batch (#6867 updated, addmm / cholesky_solve newly opened as #6910 / #6911), and others — dense review updates throughout the day.

1.3 KernelGen Pipeline: Nvidia Batch of 8 Merged, Ascend Batch Under Review (10-08)

Date: 2026-10-08　Source: FlagGems-Experimental #647 and 8 total (see appendix)

  • Nvidia batch of 8 merged (FlagGems-Experimental): nested_tensor_size (#647), _choose_qparams_per_tensor (#681), slow_conv_dilated3d (#695), chain_matmul (#688), copy_sparse_to_sparse (#687), diagflat (#683), slow_conv_dilated2d (#694), _new_zeros_with_same_feature_meta (#661) — merged in two batches between 10:55 and 17:50.
  • Backlog batch advancing: roughly 25 KernelGen PRs were updated in a concentrated burst from evening into the early hours of the next day (including dstack, coalesce, the sparse* sparse series, the convolution series, etc.), with the review pipeline running continuously.
  • Ascend batch: 7 PRs including beam_search_score, round_, batch_norm_no_update, histc, selu_ (#6638–#6644) were updated in a concentrated burst late at night.
  • Tooling: #6902 “Define categorized Triton kernel requirements for all” newly opened (KernelGen-side engineering).

1.4 FlagAttention: 7 Attention Operators Merged in a Batch (10-08)

Date: 2026-10-08　Source: #65 and 7 total (see appendix)

  • Bulk merge (19:26–19:38, 7 PRs in 12 minutes): TLE Log-Linear Attention forward operator (#65), InfLLM-V2 attention (#87), Forgetting Attention / ACP (#89), DiffKV attention (#78), TLE fused AttnRes forward kernel (#47), wall attention (#88), Inkling FA4 attention (#86, Triton/TLE kernel + tests + benchmarks).
  • Coverage areas: long-context sparsification (InfLLM-V2, wall), KV compression (DiffKV), memory decay (Forgetting), linearized attention (Log-Linear), and residual attention fusion (AttnRes) — frontier attention variants continue to be “operator-library-ized.”
  • Continuity: Along with the 09-30 moba / Hy3 / parallax batch, this is part of the repo’s periodic expansion of its “attention variant library.”

1.5 FlagTree: 7 Merges Covering AMD / HCU / PPU / mthreads and New Backend thrive (10-08)

Date: 2026-10-08　Source: #1085 · #1039 · #1288 · #1329 and others (see appendix)

  • New backend (thrive): #1085 adds a ThvTile dialect and chiplet-to-chiplet communication support (tle.chiplets) for the thrive backend — including ChipletToThvTile conversion, plus test examples for chiplet barriers / producer-consumer / allgather-GEMM / reduce-scatter / MoE all2all; this backend is a new third_party directory (a third-party technology review attributes it to the DAMO line; no official statement has been seen).
  • AMD: gfx1201 TopK provider selection and tuning (#1039), TLE matmul load rematerialization after WMMA selection (#1033).
  • Hygon: #1288 fixes volatile load handling in HCU backend LLVM lowering (merged 22:47).
  • DAMO XuanTie (PPU): #1329 updates PPU CI to vllm-plugin-fl 0.3.0rc2.post1.
  • Others: mthreads vllm benchmark TP adjusted to 2 (#1325), FlagCX commit pinned to 2026/09/30 (#1326).
  • Under review: XPU backend and internal triton sync (#1330), tle.gpu.alloc alias extended to mthreads / hcu (#1316), Ascend XCS batch (#1238 / #1275), etc.

1.6 FlagQuantum: CI Fix Landed, QBoson (Kaiwu) Provider PR Opened (10-08)

Date: 2026-10-08　Source: #597 · #599 · #598

  • CI fix landed (#597, merged 13:46): CI now reads main by revision and enforces validation, fixing the concurrency defect identified in the previous window where “merging a green PR caused main validation to be cancelled”; #599 (14:50) subsequently updated the workflow counting metric and recorded the cancellation cases.
  • QBoson provider opened (#598, newly opened 10:40): adds provider integration for QBoson’s “Kaiwu” platform — including Ising / QUBO conversion, precision evidence, bounded sampling, resumable task lifecycle, and provider billing reconciliation; also adds a QDiffusion channel pinned to Kaiwu 1.3.1 (offline readiness, source and artifact provenance, two-node A800 system probe, three-seed protein training, ESM2 local evaluation, etc.). The integration keeps the provider API outside the stable facade and does not distribute the proprietary SDK; the branch side pushed more than a dozen commits in succession from the evening of 10-08 (17:42–20:18).
  • Vocabulary migration batch: WQ-4 (runtime segment) and WQ-5 (algorithm segment) slices continue to advance — two WQ-4 commits (10-08 10:00 / 10:30), two WQ-5 commits (with 10-09 timestamp scheduling annotations), executed per the Proposal 065 naming table with gate results recorded.
  • Others: the parity/w6-recovery branch advanced throughout the day (COBYLA constrained optimizer, IR internal hierarchy contracts, QEC hypergraph decoding, and 10+ other commits); a new bench/compiler-optimization-levels-vs-qiskit benchmark branch was created (measuring optimization depth and gate counts against Qiskit).

1.7 Other Activity: FlagCX / vllm-plugin-FL / FlagGems-vllm / docs / FlagFFT (10-08 ~ 10-09)

Date: 2026-10-08 ~ 2026-10-09　Source: FlagCX #642 · vllm-plugin-FL #561 · #576 and others (see appendix)

  • FlagCX: #640 completed GDR visibility handling and validation (merged 15:22); #642 P2P Engine WRITE correctness coverage (merged 10-09 00:26). Under review, #643 (Hygon Device API and IR CI tests) and #645 (backporting the 0.14.0-rc2 packaging changes to main — a 0.14 version-line signal) were newly opened.
  • vllm-plugin-FL: #561 merged (11:44) — Kunlunxin P800 upgraded to vLLM 0.24.0 (entered review in late September); in the afternoon, a concentrated wave of vLLM 0.28.0 adaptation PRs was opened: #574 (DAMO XuanTie PPU), #575 (Moore Threads MUSA), #576 (Hygon), and #487 (Ascend 910C).
  • FlagGems-vllm: #895 moved ascend-related imports into try protection (merged 14:51); under review, #896 (Qwen4 HC / PLE / sparse QSA operators), #902 (extension-free M3 inference operators with optional W8A8 fusion), #840 (MetaX fused Marlin MoE), etc.
  • docs / Megatron-LM-FL: docs #544 bot updated ModelScope documentation (18:35); Megatron-LM-FL #104 (MegaLens probe and parallel analysis integration) updated in the evening.
  • FlagFFT: dev branch development continues (pushed 10-08 17:28; no PR updates within the window).

II. News Coverage and Ecosystem

2.1 Two Ecosystem Items Included: Tsingmicro Sets Up in Shanghai, Enflame Post-IPO Analysis (10-08)

Date: 2026-10-08　Source: Sohu · Zhangtongshe Tech · Huxiu

  • Tsingmicro: Tsingmicro wholly established “Shanghai Zhisuan Qingxin Information Technology Co., Ltd.” in Xuhui District, Shanghai; Tsingmicro is a FlagOS member (reconfigurable computing route, with the RAISA software stack natively supporting the domestic open-source framework ecosystem), with cumulative compute card orders exceeding 30,000 units in 2025 and products deployed at scale in more than ten intelligent computing centers.
  • Enflame: Huxiu published a post-IPO analysis piece (originating from the WeChat public account “Zhihe pro”), focusing on its customer structure with 83% of revenue dependent on Tencent and the independence of its DSA route (listed on the STAR Market on September 11 at an issue price of 142.18 yuan/share) — this is a capital-markets-side development with no direct connection to the FlagOS technical line, included as fundamental background on a member unit.

2.2 Search and Official Channel Check: Component Line Remains Quiet (10-08 ~ 10-09)

Date: 2026-10-08 ~ 2026-10-09　Source: Google News RSS (24 sets of Chinese and English query terms, via proxy), HN Algolia, flagos.csdn.net, BAAI Hub

  • Component-term searches (14 sets including FlagOS / FlagGems / FlagScale / FlagTree / FlagPerf / FlagCX / KernelGen, etc.) returned zero hits within the window, continuing the quiet streak.
  • Ecosystem-term searches returned 6 hits: 2 included (see 2.1); the remaining 4 were excluded per established criteria (2 stock market pieces, 1 industry summit recap, 1 incubation financing piece).
  • Six HN Algolia queries returned no direct hits; the official CSDN column had no new articles within the window (latest content date remains 09-24); the BAAI Hub homepage showed no FlagOS-themed entries.

III. Member Deep Dives

3.1 Kunlunxin: Fix Batch Merged, Six New PRs in Quick Succession, and P800 vLLM 0.24 Landed (10-08)

Date: 2026-10-08　Source: FlagGems #6369 · #6896 · vllm-plugin-FL #561 et al. (see appendix)

  • Merged: 7 FlagGems fix PRs (see 1.2); vllm-plugin-FL #561 landed P800 support up to vLLM 0.24.0.
  • Six new PRs opened (#6896 ~ #6901, 18:45 ~ 18:53): log_softmax_out long-row reduction fusion, softmax_out tle.gpu fusion path, mkldnn_rnn_layer LSTM (MMA payload), linalg_matrix_norm reduction optimization, nonzero_numpy (tle.raw cluster payload), thnn_fused_lstm_cell_backward single-launch fusion.
  • XPU3 line: SDNN 46-item fix batch updated (#6867); new addmm device baseline and cholesky_solve Triton FP32 fast path opened (#6910 / #6911).
  • Periphery: FlagTree synced XPU backend with internal triton (#1330, under review); Torch-FL #273 added P800 CI line for the 2.9 stable branch.
  • Assessment: Kunlunxin is the most active vendor in this window—merges, new PRs, inference upgrades, and compiler sync all running in parallel, exhibiting a “small batches, fast cadence” rhythm.

3.2 Enflame: GCU fp8 Fix Merged and Post-IPO Attention (10-08)

Date: 2026-10-08　Source: FlagGems #6904 · #5453

  • #6904 (merged 20:34): Fixed the issue on the torch_gcu (Enflame GCU) backend where “unsupported FP8 conversions directly abort the entire process” (such errors on this backend surface as process aborts rather than Python exceptions); the test side now actually executes conversion probes within try/except.
  • #5453 (merged 15:23): Fixed environment variable settings for Enflame CI.
  • Ecosystem: Post-IPO analysis pieces drew attention (see 2.1); both the technical and capital tracks are in an active phase.

3.3 Hygon: vLLM 0.28 Adaptation Started, Hygon Device API CI, and HCU Fixes (10-08)

Date: 2026-10-08 ~ 2026-10-09　Source: vllm-plugin-FL #576 · FlagCX #643 · FlagTree #1288

  • Inference side: vllm-plugin-FL #576 started Hygon adaptation for vLLM 0.28.0 (opened 16:05).
  • Communication side: FlagCX #643 enabled Hygon Device API and IR CI tests (opened 10-09 00:32).
  • Compiler side: FlagTree #1288 fixed volatile load handling in HCU backend LLVM lowering (merged 22:47).
  • Operator side: FlagGems #6559 cudnn_convolution (Nvidia / Iluvatar / Hygon) review updated.

3.4 Moore Threads / MetaX: Benchmarks and Operator Lines in Parallel (10-08)

Date: 2026-10-08　Source: FlagTree #1325 · FlagGems #6903 · FlagGems-vllm #840

  • Moore Threads: FlagTree #1325 (mthreads vllm benchmark TP adjusted to 2) merged, #1316 (tle.gpu.alloc alias) updated; FlagGems mthreads fixes (#6903 empty index treated as no-op, #5981 sort radix, #6878 SDPA); vllm-plugin-FL #575 started MUSA adaptation for vLLM 0.28.0.
  • MetaX: FlagGems-vllm #840 (fused Marlin MoE W4A16 / W8A16 dual backend), #878 (w8a8 flash_attn_varlen) continue under review.

3.5 DAMO XuanTie / Ascend: Compiler and CI Line Updates (10-08)

Date: 2026-10-08　Source: FlagTree #1329 · FlagGems #6887 · FlagScale #1300

  • DAMO XuanTie (PPU): FlagTree #1329 (PPU CI updated to vllm-plugin-fl 0.3.0rc2.post1) merged; FlagGems #6883 (PPU CI updated to flagtree 0.7.0) merged; vllm-plugin-FL #574 started THead PPU adaptation for vLLM 0.28.0; FlagScale #1300 added PPU 810E training CI (under review).
  • Ascend: FlagGems #6887 (KMCompiler-generated path linear optimization) merged; KernelGen Ascend batch (#6638 ~ #6644) under review update; FlagTree Ascend XCS batch (#1238 / #1275) under review; vllm-plugin-FL #487 advances Ascend 910C support for vLLM 0.28.

3.6 Other Members: Tsingmicro / Iluvatar / Cambricon / Horizon (Within Window)

Date: 2026-10-08 ~ 2026-10-09

  • Tsingmicro: Ecosystem move into Shanghai (see 2.1); no new public FlagOS technical updates within the window.
  • Iluvatar, Cambricon, Horizon: no new public updates within the window.

IV. Summary and Trend Observations

  1. First volume surge window after the holiday: 11 repos with pushes, 75 commits, 69 merges (63 new) and 151 review updates—organizational cadence fully restored; merges took a “batch pipeline” form (sglang 16 PRs / 32 minutes, Kunlunxin five in a row / 3 minutes, FlagAttention 7 PRs / 12 minutes), signaling that all lines have entered a “review clearance + concentrated merge” phase.
  2. Competition results begin landing in bulk: FlagGems-sglang’s Track1 kernels closed out 16 at once; the competition channel’s delivery to the operator library has moved from “review queue” to “results landing,” with contributor distribution showing the breadth of community participation.
  3. Multi-vendor vLLM upgrade wave begins: After Kunlunxin P800 merged into vLLM 0.24.0, DAMO XuanTie PPU, Moore Threads, Hygon, and Ascend all concentrated on starting vLLM 0.28.0 adaptation on the same afternoon—a version-line shift in the inference plugin layer.
  4. KernelGen pipeline and operator fixes both high-frequency: Nvidia 8 merges, Ascend batch under review, FlagGems 3 operators + vendor fixes (Kunlunxin 7, Enflame fp8, Hygon HCU, Ascend KMCompiler, etc.), showing post-holiday “clearing the books” characteristics.
  5. Quantum line expands to multiple providers: QBoson provider and QDiffusion integration PR opened, w6-recovery branch advancing in large strides (COBYLA, IR internal hierarchy, QEC), and a new benchmark branch against Qiskit created—FlagQuantum moving from “release certification” into a “capability extension” phase.
  6. Next observation points: review results of the vLLM 0.28 adaptation batch; merge cadence of remaining sglang competition PRs (add3, etc.); landing of FlagTree thrive / XPU / tle.gpu.alloc three lines; progress on FlagQuantum #598 merge; and when the first new Release since 09-28 will appear.

Appendix: Source Verification Table

Category Source Verification Method Result
GitHub org: flagos-ai repos API Full verification of pushed_at across 54 repos 11 repos had pushes within the window (FlagGems / FlagGems-sglang / FlagGems-Experimental / FlagTree / FlagAttention / FlagQuantum / FlagCX / vllm-plugin-FL / FlagGems-vllm / docs / FlagFFT); no new repos
GitHub Commit search (committer-date basis) Full item-by-item verification of all 75 commits FlagGems 19 / FlagGems-sglang 16 / FlagQuantum 13 / FlagGems-Experimental 8 / FlagTree 7 / FlagAttention 7 / vllm-plugin-FL 2 / FlagCX 2 / FlagGems-vllm 1
GitHub Merge search (is:merged) Item-by-item verification of all 69 merges 63 merges later than the previous issue’s cutoff; the other 6 (FlagQuantum 5 + vllm-plugin-FL 1) were already covered in the previous issue
GitHub Open search (is:open) 151 updates; 100 sampled for verification FlagGems 48 / FlagGems-Experimental 25 / vllm-plugin-FL 7 / FlagTree 6 / FlagGems-vllm 5 / FlagCX 4 / 5 others
GitHub Branches and commit atoms Top-ordered commits on default branches of 9 repos + full-branch check of FlagQuantum Top-ordered commits on main branches consistent with commit search; no merge record into main found for the vocabulary migration slice
Releases Tag / Release atom verification releases across 24 repos + tags across 6 repos No new tags / new Releases within the window
News Google News RSS (24 sets of Chinese and English query terms, via proxy) 24-hour window filtering + item-by-item review 6 hits: 2 included, 4 excluded
Community HN Algolia Review of 6 sets of queries No direct hits
Official channels CSDN column / BAAI Hub Page-scrape review No new articles within the window on CSDN; no FlagOS entries on the BAAI Hub homepage
Historical comparison 10-08 daily report (previous issue) Deduplication verification Previous issue covered up to 10-08 10:00; the 4-hour overlap between this window and the previous one has been deduplicated by cutoff

Complete Source List