{ "schema": "fak-monitored-repositories/1", "updated_at": "2026-09-04", "methodology": "Harness, model-serving, and agentic-performance repositories absent by exact owner/name from the durable registry were ranked by direct seam relevance, current development activity, adoption signal, license clarity, and dedupe against existing fak issues/notes. Stars are a discovery signal, not a quality claim; checked revisions and push timestamps make each observation reproducible.", "repositories": [ { "repository": "seal-rg/recurrent-pretraining", "url": "https://github.com/seal-rg/recurrent-pretraining", "status": "studied", "priority": 2, "why": "Recurrent-depth reference; scoped source audit, not native execution evidence.", "last_checked": "2026-09-05", "checked_revision": "1ea7220ec7eb42d13e89db0663df254d0bcdc28e", "study_note": "docs/research/astra-recurrent-depth-support.md", "receipt_id": "study_bbc18af0e9160e542a0708e04f7ffb20640100eafa2f90920c10ccaf83f04ca1", "inventory": { "filed_issue_count": 1, "issue_refs": [ "#11700" ], "mode": "scoped", "candidate_count": 2 } }, { "repository": "Nanbeige/vllm", "url": "https://github.com/Nanbeige/vllm", "status": "studied", "priority": 2, "why": "Recurrent-depth reference; scoped source audit, not native execution evidence.", "last_checked": "2026-09-05", "checked_revision": "62f6de733d7ae63b759329993bc209e67afdf431", "study_note": "docs/research/astra-recurrent-depth-support.md", "receipt_id": "study_bbc18af0e9160e542a0708e04f7ffb20640100eafa2f90920c10ccaf83f04ca1", "inventory": { "filed_issue_count": 1, "issue_refs": [ "#11700" ], "mode": "scoped", "candidate_count": 2 } }, { "repository": "mitkox/Personal-AI-Router", "url": "https://github.com/mitkox/Personal-AI-Router", "status": "studied", "priority": 1, "why": "NVIDIA Personal AI Router (PAIR): Local multi-node inference router distributing independent requests across heterogeneous PCs/Macs (Windows, macOS, Linux); RFC 9140 EAP-NOOB zero-trust pairing, Windows unicast UDP multicast send fix, native Win32 DXGI COM and PDH GPU telemetry, and macOS IORegistry IOAccelerator telemetry.", "last_checked": "2026-09-04", "checked_revision": "13b68115fa2c9c1d94f1ead1358f8d5a527cfecf", "stars_at_check": 12, "last_push_at_check": "2026-09-04T07:19:07Z", "study_note": "docs/notes/CONCEPT-STUDY-NVIDIA-PERSONAL-AI-ROUTER-2026-09-04.md", "receipt_id": "study_e10f8240a711d1576ec020f28f43fb275e320d5cb289c86f47b816bb088e7f6d", "inventory": { "mode": "exhaustive", "indexed_revision": "13b68115fa2c9c1d94f1ead1358f8d5a527cfecf", "subsystem_count": 14, "candidate_count": 7, "filed_issue_count": 3, "issue_refs": [ "#11317", "#11318", "#11319" ] } }, { "repository": "ROCm/rccl", "url": "https://github.com/ROCm/rccl", "status": "studied", "priority": 1, "why": "Radeon Collective Communication Library: direct P2P transport over xGMI (up to 896 GB/s, ~200ns latency) and PCIe P2P, GDRCopy MMIO doorbells, and ibv_reg_dmabuf_mr direct GPU VRAM registration over InfiniBand/RoCE verbs.", "last_checked": "2026-09-04", "checked_revision": "57e58688f44c77076ad536ef1f6b68741fc6e694", "stars_at_check": 420, "last_push_at_check": "2026-09-02T19:05:52Z", "study_note": "docs/notes/CONCEPT-STUDY-AMD-GPU-DIRECT-ROCM-RDMA-2026-09-04.md", "receipt_id": "study_6f42749e9f4a8d98a9ca9bd3686fd5dc863fc131dd7c2951a30d4044645748f5", "inventory": { "mode": "exhaustive", "indexed_revision": "57e58688f44c77076ad536ef1f6b68741fc6e694", "subsystem_count": 8, "candidate_count": 4, "filed_issue_count": 4, "issue_refs": [ "#11226", "#11227", "#11228", "#11230" ] } }, { "repository": "ZaidQureshi/bam", "url": "https://github.com/ZaidQureshi/bam", "status": "studied", "priority": 1, "why": "BaM: Accelerator-centric GPU-initiated direct I/O to NVMe SSDs over PCIe P2P DMA without CPU involvement; submission/completion queues in GPU VRAM with direct doorbell writes.", "last_checked": "2026-09-04", "checked_revision": "315fadfc5c5c018a64596157bfac94ecbb7d87a2", "stars_at_check": 240, "last_push_at_check": "2026-03-28T05:55:23Z", "study_note": "docs/notes/CONCEPT-STUDY-AMD-GPU-DIRECT-ROCM-RDMA-2026-09-04.md", "receipt_id": "study_6f42749e9f4a8d98a9ca9bd3686fd5dc863fc131dd7c2951a30d4044645748f5", "inventory": { "mode": "exhaustive", "indexed_revision": "315fadfc5c5c018a64596157bfac94ecbb7d87a2", "subsystem_count": 4, "candidate_count": 1, "filed_issue_count": 1, "issue_refs": [ "#11229" ] } }, { "repository": "davidcanar/vllm-strix-halo", "url": "https://github.com/davidcanar/vllm-strix-halo", "status": "studied", "priority": 1, "why": "Thunderbolt-4 / USB4 RoCE-RDMA TP2 clustering for dual AMD Strix Halo APUs (Ryzen AI MAX+ 395) running 300B-class models (GLM-5.3-Flash, DeepSeek-V4) with custom tbv kernel module and ~105 us low-latency all-reduce hook bypassing RCCL.", "last_checked": "2026-09-03", "checked_revision": "7f4a21e8932c0b653198031d279cfbf440182650", "stars_at_check": 48, "last_push_at_check": "2026-09-03T18:45:00Z", "study_note": "docs/notes/CONCEPT-STUDY-DAVIDCANAR-STRIX-HALO-ROCE-2026-09-03.md", "receipt_id": "study_e850cd73d04847648bae269686576af4e40464217efd48c4b4821fb1c086f0a4", "inventory": { "mode": "exhaustive", "indexed_revision": "7f4a21e8932c0b653198031d279cfbf440182650", "subsystem_count": 5, "candidate_count": 6, "filed_issue_count": 0, "issue_refs": [] } }, { "repository": "FlashML-org/FreeToken", "url": "https://github.com/FlashML-org/FreeToken", "status": "studied", "priority": 1, "why": "Edge-native MoE serving engine with direct NVMe PLE n-gram table streaming (0.1-0.4ms) with CUDA stream memops for Qwen3.8-Flash-Next, bandwidth-adaptive PCIe-vs-CPU expert miss co-execution (q* policy), double-buffered prefill streaming with D2D hit compaction, semantic tool-call anchor checkpointing, and GLM-5.3-Flash 45-layer hybrid KDA/DSA execution.", "last_checked": "2026-09-03", "checked_revision": "03c28d2b154484f84397e42731a9b97f340322b5", "stars_at_check": 1850, "last_push_at_check": "2026-09-03T02:01:57Z", "study_note": "docs/notes/CONCEPT-STUDY-FREETOKEN-2026-09-03.md", "receipt_id": "study_75aef7f39f666f166991072b47a5083c2bc150bf38c0fc3c3f6e963d9d4120a9", "inventory": { "mode": "exhaustive", "indexed_revision": "03c28d2b154484f84397e42731a9b97f340322b5", "subsystem_count": 8, "candidate_count": 8, "filed_issue_count": 3, "issue_refs": [ "#11037", "#11038", "#11115" ] } }, { "repository": "ROCm/hrx-system", "url": "https://github.com/ROCm/hrx-system", "status": "studied", "priority": 1, "why": "Alternative, ultra-low-latency HIP runtime and native asynchronous compiler substrate (Loom) for AMD GPUs (RDNA3.5 Strix Halo, RDNA4, CDNA3) with direct user-space AQL and PM4 command submission, zero-LLVM HSACO generation, KPACK compressed packaging, stream-ordered VMM slab allocation, and in-band device ASAN.", "last_checked": "2026-09-03", "checked_revision": "4c5f2d9bfd80ad34343c61eadae99025923c26b3", "stars_at_check": 32, "last_push_at_check": "2026-09-03T21:53:40Z", "study_note": "docs/notes/CONCEPT-STUDY-ROCM-HRX-SYSTEM-2026-09-03.md", "receipt_id": "study_2f5d10b5eb90152e1263112cf68dda0387221ce77cf11f5fb7d586975e691016", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/rocm-hrx-system.json", "indexed_revision": "4c5f2d9bfd80ad34343c61eadae99025923c26b3", "subsystem_count": 5, "candidate_count": 8, "filed_issue_count": 6, "issue_refs": [ "#11092", "#11093", "#11094", "#11095", "#11096", "#11097", "#11098" ] } }, { "repository": "lemonade-sdk/lemonade", "url": "https://github.com/lemonade-sdk/lemonade", "status": "studied", "priority": 1, "why": "Local multi-backend AI model server and router featuring speculative draft companion GGUF discovery and quant-bit distance matching, zero-byte-streamed accelerator reset and replay, reload-cost-weighted model eviction scoring under memory pressure, in-band SSE route decision injection, OS suspend/idle inhibition, and dynamic model target aliasing with active-standby failover.", "last_checked": "2026-09-03", "checked_revision": "bc8d99ccfee301742f1dd86a3f6ec406bb8a863d", "stars_at_check": 5604, "last_push_at_check": "2026-09-03T21:21:29Z", "study_note": "docs/notes/CONCEPT-STUDY-LEMONADE-2026-09-03.md", "receipt_id": "study_cc3b4fe925b988801bccaeb92d141ee0e0f0a37cc17e8edb0e4fb3a883df1e69", "inventory": { "mode": "exhaustive", "indexed_revision": "bc8d99ccfee301742f1dd86a3f6ec406bb8a863d", "subsystem_count": 8, "candidate_count": 9, "filed_issue_count": 6, "issue_refs": [ "#11086", "#11087", "#11088", "#11089", "#11090", "#11091" ] } }, { "repository": "syv-ai/qwen38-27b-rtx3090", "url": "https://github.com/syv-ai/qwen38-27b-rtx3090", "status": "studied", "priority": 1, "why": "Consumer 24GB GPU serving stack for Qwen3.8-27B featuring Lookup-Augmented Block Drafting (381 tok/s decode), split-KV multi-query speculative attention, 16-bit GDN recurrent state, AutoRound negative-scale Marlin fixes, and 40k draft vocab truncation.", "last_checked": "2026-09-03", "checked_revision": "8d832f8758ae4fd36c29a15d3c45888922bc4377", "stars_at_check": 1400, "last_push_at_check": "2026-09-03T08:57:21Z", "study_note": "docs/notes/CONCEPT-STUDY-QWEN38-27B-RTX3090-2026-09-03.md", "receipt_id": "study_69a87c504d53efb41a243ad7f233c19d5293ecbb611869ed64292ad229fd7c2d", "inventory": { "mode": "exhaustive", "indexed_revision": "8d832f8758ae4fd36c29a15d3c45888922bc4377", "subsystem_count": 6, "candidate_count": 9, "filed_issue_count": 6, "issue_refs": [ "#11099", "#11100", "#11101", "#11102", "#11103", "#11104" ] } }, { "repository": "youssofal/mtplx", "url": "https://github.com/youssofal/mtplx", "status": "studied", "priority": 1, "why": "Native macOS Metal/MLX inference runtime for Qwen 3.8 27B featuring built-in MTP speculative decoding with Leviathan–Chen rejection sampling, blocked-sequential Gated DeltaNet prefill Metal kernels (eliminating 32x redundant DRAM reads), double-shift norm defense, wide-M tail-causal SDPA Metal 4 TensorOps, and roofline auto-tuning.", "last_checked": "2026-09-03", "checked_revision": "e652d55e2652137a4abcf1312357abbf3eb9d692", "stars_at_check": 190, "last_push_at_check": "2026-09-03T08:00:00Z", "study_note": "docs/notes/CONCEPT-STUDY-YOUSSOFAL-MTPLX-2026-09-03.md", "receipt_id": "study_c48306e8264b0e014c3b5489f835cdd751145e316989a06c2d934b6006b8a51a", "inventory": { "mode": "exhaustive", "indexed_revision": "e652d55e2652137a4abcf1312357abbf3eb9d692", "subsystem_count": 5, "candidate_count": 4, "filed_issue_count": 3, "issue_refs": [ "#10988", "#10989", "#10990" ] } }, { "repository": "carloslfu/slotstream", "url": "https://github.com/carloslfu/slotstream", "status": "studied", "priority": 1, "why": "Swift 6 + MLX runtime for Qwen 3.8 Flash-Next (125B MoE) on standard 16GB-48GB Macs using QD32 asynchronous pread expert streaming from SSD (17.3 GB/s direct throughput), scan-resistant prefill sweeps with sorted grouped GEMMs, extend-only GDN linear recurrence prefix caching, and Mach VM elastic memory governing.", "last_checked": "2026-09-03", "checked_revision": "53028cf321123d3e9ba95adea40c4c582100ce13", "stars_at_check": 275, "last_push_at_check": "2026-09-03T09:30:00Z", "study_note": "docs/notes/CONCEPT-STUDY-CARLOSLFU-SLOTSTREAM-2026-09-03.md", "receipt_id": "study_5c77965f0f47441af377018a5c4b58093989021620df6d0706a5dba4263d677c", "inventory": { "mode": "exhaustive", "indexed_revision": "53028cf321123d3e9ba95adea40c4c582100ce13", "subsystem_count": 5, "candidate_count": 5, "filed_issue_count": 2, "issue_refs": [ "#10991", "#10994" ] } }, { "repository": "paul-gauthier/aider", "url": "https://github.com/paul-gauthier/aider", "status": "studied", "priority": 1, "why": "High-velocity CLI coding agent with Tree-sitter PageRank symbol graph repomap, Architect/Editor dual-model orchestration, whitespace-tolerant search/replace with relative indentation, and pre-commit auto-lint reflection.", "last_checked": "2026-09-03", "checked_revision": "5dc9490bb35f9729ef2c95d00a19ccd30c26339c", "stars_at_check": 28400, "last_push_at_check": "2026-09-02T18:00:00Z", "study_note": "docs/notes/CONCEPT-STUDY-OSS-AGENTIC-DEV-BEST-PRACTICES-2026-09-03.md", "receipt_id": "study_87a344905593c035d8add68ec1e0c09a30cac4259b0cd7c396d0a437098f0784", "inventory": { "mode": "exhaustive", "indexed_revision": "5dc9490bb35f9729ef2c95d00a19ccd30c26339c", "subsystem_count": 5, "candidate_count": 4, "filed_issue_count": 4, "issue_refs": [ "#10966", "#10972", "#10976", "#10982" ] } }, { "repository": "princeton-nlp/SWE-agent", "url": "https://github.com/princeton-nlp/SWE-agent", "status": "studied", "priority": 1, "why": "Foundational Agent-Computer Interface (ACI) benchmark runner featuring windowed file navigation, delta-lint error isolation, LastNObservations history processor, closed-window deduplication, and reproduction script invariants.", "last_checked": "2026-09-03", "checked_revision": "3ea751c087f32b16e039a2233dd6eefecef325d5", "stars_at_check": 15600, "last_push_at_check": "2026-09-01T20:00:00Z", "study_note": "docs/notes/CONCEPT-STUDY-OSS-AGENTIC-DEV-BEST-PRACTICES-2026-09-03.md", "receipt_id": "study_87a344905593c035d8add68ec1e0c09a30cac4259b0cd7c396d0a437098f0784", "inventory": { "mode": "exhaustive", "indexed_revision": "3ea751c087f32b16e039a2233dd6eefecef325d5", "subsystem_count": 6, "candidate_count": 6, "filed_issue_count": 6, "issue_refs": [ "#10967", "#10969", "#10970", "#10973", "#10974", "#10983" ] } }, { "repository": "cline/cline", "url": "https://github.com/cline/cline", "status": "studied", "priority": 1, "why": "Autonomous coding agent with Git-based shadow worktree checkpoints preserving untracked files via 3rd-parent tracking, granular auto-approve regex rulesets, and .worktreeinclude sparse-checkout sub-workspace isolation.", "last_checked": "2026-09-03", "checked_revision": "d7cb79aea570dd2fb86bc3d28c7382fef2a7e345", "stars_at_check": 24500, "last_push_at_check": "2026-09-03T10:00:00Z", "study_note": "docs/notes/CONCEPT-STUDY-OSS-AGENTIC-DEV-BEST-PRACTICES-2026-09-03.md", "receipt_id": "study_87a344905593c035d8add68ec1e0c09a30cac4259b0cd7c396d0a437098f0784", "inventory": { "mode": "exhaustive", "indexed_revision": "d7cb79aea570dd2fb86bc3d28c7382fef2a7e345", "subsystem_count": 4, "candidate_count": 3, "filed_issue_count": 3, "issue_refs": [ "#10980", "#10981", "#10986" ] } }, { "repository": "continuedev/continue", "url": "https://github.com/continuedev/continue", "status": "studied", "priority": 1, "why": "Leading open-source AI code assistant featuring deterministic Myers diff line streaming, AST declaration replacement fallback, modular ContextProvider architecture, and cache-friendly prompt section ordering.", "last_checked": "2026-09-03", "checked_revision": "5522c6f44ca0ac3528b37244818fbfa39b5af470", "stars_at_check": 22100, "last_push_at_check": "2026-09-03T11:00:00Z", "study_note": "docs/notes/CONCEPT-STUDY-OSS-AGENTIC-DEV-BEST-PRACTICES-2026-09-03.md", "receipt_id": "study_87a344905593c035d8add68ec1e0c09a30cac4259b0cd7c396d0a437098f0784", "inventory": { "mode": "exhaustive", "indexed_revision": "5522c6f44ca0ac3528b37244818fbfa39b5af470", "subsystem_count": 5, "candidate_count": 4, "filed_issue_count": 4, "issue_refs": [ "#10971", "#10975", "#10984", "#10985" ] } }, { "repository": "OpenAutoCoder/Agentless", "url": "https://github.com/OpenAutoCoder/Agentless", "status": "studied", "priority": 1, "why": "Minimalist, highly effective SWE-bench framework demystifying agent loops with two-phase hierarchical localization, AST skeletonization with body elision, negative directory pruning, and multi-candidate patch majority voting.", "last_checked": "2026-09-03", "checked_revision": "5ce5888b9f149beaace393957a55ea8ee46c9f71", "stars_at_check": 2100, "last_push_at_check": "2026-08-20T12:00:00Z", "study_note": "docs/notes/CONCEPT-STUDY-OSS-AGENTIC-DEV-BEST-PRACTICES-2026-09-03.md", "receipt_id": "study_87a344905593c035d8add68ec1e0c09a30cac4259b0cd7c396d0a437098f0784", "inventory": { "mode": "exhaustive", "indexed_revision": "5ce5888b9f149beaace393957a55ea8ee46c9f71", "subsystem_count": 4, "candidate_count": 4, "filed_issue_count": 4, "issue_refs": [ "#10968", "#10977", "#10978", "#10979" ] } }, { "repository": "microsoft/vidur", "url": "https://github.com/microsoft/vidur", "status": "studied", "priority": 1, "why": "High-fidelity MLSys'24 LLM inference simulator and capacity planning engine with execution time regression modeling, prefix caching, multi-tier Disk/VRAM KV cache management, tolerance-slack sticky routing, uncached prefill volume balancing, and adaptive capacity binary search.", "last_checked": "2026-09-03", "checked_revision": "abae7f63aa857300f5cdc6f5e0d27860cd24721b", "stars_at_check": 671, "last_push_at_check": "2026-08-24T12:47:47Z", "study_note": "docs/notes/CONCEPT-STUDY-MICROSOFT-VIDUR-2026-09-03.md", "receipt_id": "study_722a05f7439411a3291e4a376f5e48b3da011f06448fbc4480d2e1cb635d9553", "inventory": { "mode": "exhaustive", "indexed_revision": "abae7f63aa857300f5cdc6f5e0d27860cd24721b", "subsystem_count": 8, "candidate_count": 5, "filed_issue_count": 5, "issue_refs": [ "#10852", "#10853", "#10854", "#10856", "#10861" ] } }, { "repository": "wkljohn/ds4-strix-halo-tp-odinlink", "url": "https://github.com/wkljohn/ds4-strix-halo-tp-odinlink", "status": "studied", "priority": 1, "why": "High-performance C99/HIP inference engine for DeepSeek-V4 and GLM-5 on dual AMD Strix Halo APU clusters, featuring Tensor Parallelism over USB4 OdinLink and RoCE v2, contiguous range expert sharding, direct slab RDMA, AVX-512 WC streaming copy, Wave32 KDA linear recurrence, and direct I/O.", "last_checked": "2026-09-02", "checked_revision": "48e10779d4723e40a8949d0ba59c262d35379a03", "stars_at_check": 15, "last_push_at_check": "2026-09-02T12:00:00Z", "study_note": "docs/notes/CONCEPT-STUDY-DS4-STRIX-HALO-TP-ODINLINK-2026-09-02.md", "receipt_id": "study_dc99f78a887e2ab60a9ccf5807650b6da85ac992d7f69d96b335360c55892811", "inventory": { "mode": "exhaustive", "indexed_revision": "48e10779d4723e40a8949d0ba59c262d35379a03", "subsystem_count": 5, "candidate_count": 8, "filed_issue_count": 8, "issue_refs": [ "#10755", "#10756", "#10758", "#10759", "#10760", "#10761", "#10762", "#10763" ] } }, { "repository": "OpenRouterTeam/go-sdk", "url": "https://github.com/OpenRouterTeam/go-sdk", "status": "studied", "priority": 1, "why": "Official OpenRouter Go SDK defining the typed ProviderPreferences contract, Responses API, Server Tools, and multi-model deliberation schemas.", "last_checked": "2026-09-02", "checked_revision": "23ee7e759807efe760c2870002390a957b2875a7", "stars_at_check": 142, "last_push_at_check": "2026-09-01T20:00:00Z", "study_note": "docs/notes/CONCEPT-STUDY-OPENROUTER-2026-09-02.md", "receipt_id": "study_c16fcd81b174bedf20107e88c0f6f1c8ca3020dc3116346667751b346ab71780", "inventory": { "mode": "exhaustive", "indexed_revision": "23ee7e759807efe760c2870002390a957b2875a7", "subsystem_count": 8, "candidate_count": 7, "filed_issue_count": 5, "issue_refs": [ "#10742", "#10744", "#10745", "#10746", "#10747" ] } }, { "repository": "anomalyco/opencode", "url": "https://github.com/anomalyco/opencode", "status": "studied", "priority": 1, "why": "Leading open-source AI coding agent with Effect.ts runtime, separating durable prompt admission from model execution. Deeply studied for composable SystemContext, ContextEpoch prefix reuse, generic tool output bounding, dynamic permission rulesets, Catalog selection, and embedded in-process gateway host.", "last_checked": "2026-09-02", "checked_revision": "4eb29a64f0054672950acf789f2b09487ebfbb20", "stars_at_check": 203243, "last_push_at_check": "2026-09-02T13:00:00Z", "study_note": "docs/notes/CONCEPT-STUDY-OPENCODE-2026-09-02.md", "refresh_issue": "#10701", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/anomalyco-opencode.json", "indexed_revision": "4eb29a64f0054672950acf789f2b09487ebfbb20", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 32, "candidate_count": 16, "filed_issue_count": 12, "completeness_critic": "Exhaustive monorepo inventory covering core (session, system-context, tool, permission, plugin, catalog, agent, model) and opencode CLI/TUI packages; 12 independent issues filed across context, gateway, agent, tool, policy, modelroute, guard, skill, and integrations; durable receipt study_5e1d3f474b8d6240719d16e9bc97e159957340e9c0e3d32bcbb5aa9e8a1a761f recorded.", "issue_refs": [ "#10701", "#10702", "#10703", "#10704", "#10705", "#10706", "#10707", "#10708", "#10711", "#10712", "#10713", "#10714" ] } }, { "repository": "modular/modular", "url": "https://github.com/modular/modular", "status": "studied", "priority": 1, "why": "FAK's closest broad native-inference design peer across serving admission, KV movement, speculative decoding, model architecture, accelerator kernels, graph optimization, async runtime, and cache support. The study adapts mechanisms into FAK-owned Go/native paths and explicitly excludes a Modular runtime dependency or fallback.", "last_checked": "2026-08-28", "checked_revision": "1c9fd2e03331f77d3a1034127cb3700b7fa43c02", "stars_at_check": 29305, "last_push_at_check": "2026-08-28T09:02:51Z", "study_note": "docs/notes/CONCEPT-STUDY-MODULAR-MONOREPO-2026-08-28.md", "refresh_issue": "#9900", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/modular-modular.json", "forge_receipt_path": "docs/research/inventory/modular-modular-forge-receipt.json", "indexed_revision": "1c9fd2e03331f77d3a1034127cb3700b7fa43c02", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 16, "candidate_count": 87, "filed_issue_count": 68, "source_evidence": [ { "class": "open_closed_issues_prs_discussions", "evidence": [ "docs/research/inventory/modular-modular-forge-receipt.json", "gh api --paginate --slurp repos/modular/modular/issues?state=all&per_page=100&sort=created&direction=asc", "gh pr list --repo modular/modular --state all --limit 10000 --json number,title,state,createdAt,closedAt,mergedAt,labels,url", "gh api graphql repository(owner:modular,name:modular) discussions(first:100)" ], "note": "Complete six-endpoint census through 2026-08-28T23:59:59Z with checksum sha256:a0752d13d096c945d50198126de5083a4c3878f00c84fe26bfc404322ba0f40e: 3,716 issues, 2,667 pull requests, 523 discussions, 151 labels, 14 releases, and zero milestones; targeted candidate threads received deeper read-back." }, { "class": "fak_selfquery_witness", "evidence": [ "fak capabilities \"Qwen3.5 MTP embedding hidden concatenation layout\"", "fak dev index docs \"graph compiler pass pipeline SCCP SROA Mem2Reg\"", "fak dev index leaves \"KV transfer onload tensor parallel topology\"", "internal/model/qwen35_mtp_forward.go" ], "note": "All 87 candidate rows received three-layer FAK self-query and exact-owner deduplication; the note preserves PRESENT, PARTIAL, and ABSENT dispositions." }, { "class": "candidate_matrix", "evidence": [ "docs/notes/CONCEPT-STUDY-MODULAR-MONOREPO-2026-08-28.md#full-candidate-matrix" ] }, { "class": "issue_tracking", "evidence": [ "https://github.com/anthony-chaudhary/fak/issues/9900", "https://github.com/anthony-chaudhary/fak/issues/9910", "https://github.com/anthony-chaudhary/fak/issues/9977", "docs/notes/CONCEPT-STUDY-MODULAR-MONOREPO-2026-08-28.md" ], "note": "Parent #9900 owns 68 contract-clean children in the contiguous #9910-#9977 range; live read-back proved unique immutable markers, receipt references, labels, and Generation G3 routing. Durable decision/outcome receipts: study_a55850a776986e47ab91c858765a98e3cc4ce918086bb8ef380d5aa4f2325189 and study_96c38d851458c92a9dea84574cff09471a9d13bf9d4a050eb076c5bb8b62416f." } ], "issue_refs": [ "#9900", "#9910", "#9977" ], "completeness_critic": "Exhaustive maintained-tree inventory, full reachable history and tags, complete six-endpoint forge census, README/design/runtime/test/roadmap/license review, candidate-specific FAK self-query, exact-owner dedupe, 87-row disposition matrix, independent source-anchor review, and live issue-contract read-back are covered. Generated/target/vendor trees are deliberately excluded; forge comments/reviews/timelines and every linked commit were not traversed for all 7,071 parents, while candidate-bearing threads were followed selectively." } }, { "repository": "ruvnet/ruflo", "url": "https://github.com/ruvnet/ruflo", "status": "candidate", "priority": 1, "why": "Direct meta-harness overlap: multi-agent swarms, Codex and Claude integration, memory, skills, MCP, hooks, and a declared harness surface; 47 commits in the seven-day discovery window make it the hottest close seam.", "last_checked": "2026-08-14", "checked_revision": "4dcff483482cee316f47552a961bcbaadc89f378", "stars_at_check": 67866, "last_push_at_check": "2026-08-14T21:36:20Z", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/ruvnet-ruflo.json", "indexed_revision": "4dcff483482cee316f47552a961bcbaadc89f378", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 18, "candidate_count": 4, "filed_issue_count": 0, "completeness_critic": "exhaustive local tree map plus pinned-timestamp GitHub issue, pull-request, and complete discussion read-back; fak self-query, candidate adjudication, and issue tracking are recorded under non_tree_study; skipped dependency/cache/control directories: .git", "source_evidence": [ { "class": "open_closed_issues_prs_discussions", "evidence": [ "gh api --paginate --slurp repos/ruvnet/ruflo/issues?state=all&per_page=100&sort=created&direction=asc", "gh pr list --repo ruvnet/ruflo --state all --limit 1000 --json number,title,state,createdAt,closedAt,mergedAt,labels,url", "gh api graphql repository(owner:ruvnet,name:ruflo) discussions(first:100)" ], "note": "Paged GitHub read-back through the pinned commit timestamp: 1,613 issues, 1,291 pull requests, and all 86 discussions." }, { "class": "fak_selfquery_witness", "evidence": [ "fak capabilities \"power of two choices mesh load balancing scheduler routing\"", "fak capabilities \"memory retrieval BM25 semantic fusion evaluation benchmark\"", "fak capabilities \"persisted swarm health liveness lease worker status telemetry\"", "fak capabilities \"doctor environment drift runtime dependency health preflight\"" ], "note": "Candidate-specific fak queries returned native routing, fleet-health, and capability-floor surfaces; the retrieval-fusion query returned no exact capability and was adjudicated stay-minimal rather than inventing a gap." }, { "class": "candidate_matrix", "evidence": [ "docs/research/inventory/ruvnet-ruflo.json#candidate-matrix" ] }, { "class": "issue_tracking", "evidence": [ "https://github.com/anthony-chaudhary/fak/issues/8931" ], "note": "All candidates were rejected, stay-minimal, or already owned; no borrow issue survived." } ], "issue_refs": [ "#8931" ] } }, { "repository": "Untrivial-ai/agent-orchestrator", "url": "https://github.com/Untrivial-ai/agent-orchestrator", "status": "candidate", "priority": 2, "why": "Closest fleet-control analogue: plans work, spawns coding agents, isolates them in worktrees, and handles CI, reviews, and merge conflicts; at least 100 commits in the seven-day discovery window.", "last_checked": "2026-08-14", "checked_revision": "ee2c58a577317d4480a2b2ee5ff77f25e5307af9", "stars_at_check": 9518, "last_push_at_check": "2026-08-14T21:55:27Z", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/untrivial-ai-agent-orchestrator.json", "indexed_revision": "ee2c58a577317d4480a2b2ee5ff77f25e5307af9", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "roadmap_todos", "license_provenance", "completeness_critic" ], "subsystem_count": 14, "candidate_count": 0, "filed_issue_count": 0, "completeness_critic": "local tree inventory walked every non-skipped regular file and grouped immediate subsystems; skipped dependency/cache/control directories: .git; classes with partial local evidence still requiring non-tree completion: open_closed_issues_prs_discussions; still requires non-tree study artifacts: fak_selfquery_witness, candidate_matrix, issue_tracking" } }, { "repository": "langchain-ai/open-swe", "url": "https://github.com/langchain-ai/open-swe", "status": "candidate", "priority": 3, "why": "Active asynchronous coding-agent harness with sandbox, task lifecycle, desktop, evaluation, and session surfaces; 80 commits in the seven-day discovery window expose useful operating-loop seams.", "last_checked": "2026-08-14", "checked_revision": "a6c360047186cc5b8afe3a74012a12bfc94ae7c7", "stars_at_check": 10553, "last_push_at_check": "2026-08-14T22:56:23Z", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/langchain-ai-open-swe.json", "indexed_revision": "a6c360047186cc5b8afe3a74012a12bfc94ae7c7", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "roadmap_todos", "license_provenance", "completeness_critic" ], "subsystem_count": 12, "candidate_count": 0, "filed_issue_count": 0, "completeness_critic": "local tree inventory walked every non-skipped regular file and grouped immediate subsystems; skipped dependency/cache/control directories: .git, ui/src/features/agents/terminal/ghostty/vendor; classes checked absent in local tree: roadmap_todos; still requires non-tree study artifacts: open_closed_issues_prs_discussions, fak_selfquery_witness, candidate_matrix, issue_tracking" } }, { "repository": "EveryInc/compound-engineering-plugin", "url": "https://github.com/EveryInc/compound-engineering-plugin", "status": "candidate", "priority": 4, "why": "Cross-harness plugin for Codex, Claude Code, Cursor, and peers with plan-review-work-compound loops and portable agent assets; 20 commits in the seven-day discovery window.", "last_checked": "2026-08-14", "checked_revision": "345e2bea42cabea2f32d25a68fd1739e8e92cd03", "stars_at_check": 24259, "last_push_at_check": "2026-08-14T22:46:18Z", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/everyinc-compound-engineering-plugin.json", "indexed_revision": "345e2bea42cabea2f32d25a68fd1739e8e92cd03", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 22, "candidate_count": 0, "filed_issue_count": 0, "completeness_critic": "local tree inventory walked every non-skipped regular file and grouped immediate subsystems; skipped dependency/cache/control directories: .git; still requires non-tree study artifacts: fak_selfquery_witness, candidate_matrix, issue_tracking" } }, { "repository": "obra/superpowers", "url": "https://github.com/obra/superpowers", "status": "candidate", "priority": 5, "why": "Largest adoption signal among previously untracked harness-adjacent sources and a portable skills, hooks, planning, testing, and subagent-development methodology spanning Codex and other agents.", "last_checked": "2026-08-14", "checked_revision": "b36e0829c6d0140e93cfef2ca599b1b07d4a7797", "stars_at_check": 272168, "last_push_at_check": "2026-08-13T00:36:31Z", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/obra-superpowers.json", "indexed_revision": "b36e0829c6d0140e93cfef2ca599b1b07d4a7797", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "roadmap_todos", "license_provenance", "completeness_critic" ], "subsystem_count": 17, "candidate_count": 0, "filed_issue_count": 0, "completeness_critic": "local tree inventory walked every non-skipped regular file and grouped immediate subsystems; skipped dependency/cache/control directories: .git; classes checked absent in local tree: roadmap_todos; still requires non-tree study artifacts: fak_selfquery_witness, candidate_matrix, issue_tracking" } }, { "repository": "local/tensor-build", "url": "local://tensor-build-main", "status": "studied", "priority": 6, "why": "Whole-tree TensorBuild recheck extracted evidence, measurement, and native-runtime contracts, ablated TensorRT-specific mechanisms, reconciled seven existing fak owners, and filed 13 bounded leaves. Source history, forge state, and license remain unavailable, so every borrow is concepts-only.", "last_checked": "2026-08-29", "checked_revision": "snapshot-sha256:64986cf8ff942cdcd6178491d3d9af0199c354ce41f998b89ced2b0286f6772d", "stars_at_check": 0, "last_push_at_check": "", "study_note": "docs/notes/CONCEPT-STUDY-TENSOR-BUILD-2026-08-29.md", "refresh_issue": "#8985", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/local-tensor-build-2026-08-29.json", "indexed_revision": "snapshot-sha256:64986cf8ff942cdcd6178491d3d9af0199c354ce41f998b89ced2b0286f6772d", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 16, "candidate_count": 26, "filed_issue_count": 16, "completeness_critic": "exhaustive local snapshot map with raw-versus-indexed denominator; generated trees internal/build and internal/conn/web/dist are named; worldview, runtime, tests, results, changelog, roadmap/TODO, license/provenance, unavailable forge/history surfaces, 32/32 source-hash read-back, fak self-query, candidate dispositions, and issue tracking are explicit; no source license or attributable Git/forge history is available, so concepts only", "source_evidence": [ { "class": "open_closed_issues_prs_discussions", "evidence": [ "docs/notes/CONCEPT-STUDY-TENSOR-BUILD-2026-08-29.md#provenance-and-license-fence", "source:.github/ISSUE_TEMPLATE/general-ticket.md", "source:.github/PULL_REQUEST_TEMPLATE.md", "gh api graphql --raw-field query=repository-discussions" ], "note": "Templates prove local workflow shape only. Issues, pull requests, discussions, and releases cannot be enumerated and are explicitly unavailable." }, { "class": "fak_selfquery_witness", "evidence": [ "docs/notes/CONCEPT-STUDY-TENSOR-BUILD-2026-08-29.md#candidate-matrix", "fak study search --limit 20 tensor-build", "study_5152edd9388ef5a396f526004fa64ccfd0796171db04810d91a49eed606a17b4", "gh issue list --search tensor-build" ], "note": "Candidate-specific source and issue queries were replayed. Every new leaf had no exact open or closed owner; seven candidates retained existing owners and four off-axis mechanisms were excluded or watched." }, { "class": "candidate_matrix", "evidence": [ "docs/research/inventory/local-tensor-build-2026-08-29.json#non_tree_study", "docs/notes/CONCEPT-STUDY-TENSOR-BUILD-2026-08-29.md#candidate-matrix", "study_5152edd9388ef5a396f526004fa64ccfd0796171db04810d91a49eed606a17b4" ], "note": "Twenty-six candidates were adjudicated: 13 new leaves, seven exact existing owners, two present no-issue mechanisms, and four watch/exclude decisions." }, { "class": "issue_tracking", "evidence": [ "https://github.com/anthony-chaudhary/fak/issues/8985", "https://github.com/anthony-chaudhary/fak/issues/6874", "https://github.com/anthony-chaudhary/fak/issues/6875", "https://github.com/anthony-chaudhary/fak/issues/6876", "https://github.com/anthony-chaudhary/fak/issues/10268", "https://github.com/anthony-chaudhary/fak/issues/10269", "https://github.com/anthony-chaudhary/fak/issues/10270", "https://github.com/anthony-chaudhary/fak/issues/10271", "https://github.com/anthony-chaudhary/fak/issues/10278", "https://github.com/anthony-chaudhary/fak/issues/10279", "https://github.com/anthony-chaudhary/fak/issues/10280", "https://github.com/anthony-chaudhary/fak/issues/10281", "https://github.com/anthony-chaudhary/fak/issues/10282", "https://github.com/anthony-chaudhary/fak/issues/10283", "https://github.com/anthony-chaudhary/fak/issues/10284", "https://github.com/anthony-chaudhary/fak/issues/10285", "https://github.com/anthony-chaudhary/fak/issues/10286" ], "note": "The recheck retained prior owners #6874-#6876, linked exact current owners #3978, #4032, #5645, and #8635, and filed 13 new bounded leaves." } ], "issue_refs": [ "#8985", "#6874", "#6875", "#6876", "#3978", "#4032", "#5601", "#5645", "#8635", "#10268", "#10269", "#10270", "#10271", "#10278", "#10279", "#10280", "#10281", "#10282", "#10283", "#10284", "#10285", "#10286" ] } }, { "repository": "strukto-ai/mirage", "url": "https://github.com/strukto-ai/mirage", "status": "studied", "priority": 7, "why": "Unified agent resource substrate deeply studied for snapshot drift gates, typed per-operation backend capabilities, transactional workspace branches, policy placement, and observer fidelity. The surviving fak gap is a strict pre-effect external-input drift contract; stronger or already-planned fak work covers the other candidates.", "last_checked": "2026-08-25", "checked_revision": "2ed4257af98fc1a206a5444057d1290892190e69", "stars_at_check": 3567, "last_push_at_check": "2026-08-25T22:18:51Z", "study_note": "docs/notes/mirage-study-2026-08-17.md", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/strukto-ai-mirage.json", "indexed_revision": "2ed4257af98fc1a206a5444057d1290892190e69", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 17, "candidate_count": 6, "filed_issue_count": 1, "completeness_critic": "Exhaustive 8,921-file local tree map plus full pinned history, tags, releases, issue, pull-request, and discussion read-back; roadmap/TODO, license provenance, fak self-query, candidate adjudication, and issue tracking are explicit under non_tree_study; only .git internals are skipped and represented by history commands.", "source_evidence": [ { "class": "open_closed_issues_prs_discussions", "evidence": [ "gh issue list --repo strukto-ai/mirage --state all --limit 1000 --json number,title,state,createdAt,closedAt,labels,url", "gh pr list --repo strukto-ai/mirage --state all --limit 1000 --json number,title,state,createdAt,closedAt,mergedAt,labels,url", "gh api graphql repository(owner:strukto-ai,name:mirage) discussions(first:100)" ], "note": "Complete read-back at the pinned cutoff: 206 issues, 706 pull requests, zero discussions, and five releases." }, { "class": "fak_selfquery_witness", "evidence": [ "fak capabilities \"resume replay mutable external input drift pre-effect validation\" --json", "fak capabilities \"typed backend capability negotiation transactional workspace filesystem\" --json", "fak capabilities \"virtual filesystem sandbox remote execution file tool\" --json" ], "note": "Candidate-specific queries confirmed one open reliability gap and stronger existing/default seams for the remaining candidates." }, { "class": "candidate_matrix", "evidence": [ "docs/research/inventory/strukto-ai-mirage.json#candidate-matrix" ] }, { "class": "issue_tracking", "evidence": [ "https://github.com/anthony-chaudhary/fak/issues/8986", "https://github.com/anthony-chaudhary/fak/issues/7256" ], "note": "The refresh tracker and sole surviving borrow were read back; no new issue survived adjudication." } ], "issue_refs": [ "#8986", "#7256" ] } }, { "repository": "agent0ai/agent-zero", "url": "https://github.com/agent0ai/agent-zero", "status": "studied", "priority": 8, "why": "Transparent full-computer agent deeply studied for lifecycle hooks, plugin composition, hierarchical delegation, semantic memory, effect snapshots, connectors, and scheduling. FAK already covers the central kernel mechanisms with stronger policy, isolation, and witness boundaries; browser/office/vector-memory breadth remains modular.", "last_checked": "2026-08-18", "checked_revision": "baadd0dd0b09fa769a1027c183b964be85d5c8cc", "stars_at_check": 18903, "last_push_at_check": "2026-08-16T17:45:50Z", "study_note": "docs/notes/agent-zero-study-2026-08-18.md", "refresh_issue": "#8987", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/agent0ai-agent-zero.json", "indexed_revision": "baadd0dd0b09fa769a1027c183b964be85d5c8cc", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 21, "candidate_count": 5, "filed_issue_count": 0, "completeness_critic": "Exhaustive 2,292-file pinned tree walk grouped all 21 immediate subsystems and skipped only .git and vendored webui/vendor; GitHub issues and pull requests were paged to exhaustion, all 94 discussions and 72 observed releases were read back, history/roadmap/license provenance and four candidate-specific fak self-queries were audited, and every candidate was adjudicated with issue tracking.", "source_evidence": [ { "class": "open_closed_issues_prs_discussions", "evidence": [ "gh api --paginate --slurp repos/agent0ai/agent-zero/issues?state=all&per_page=100&sort=created&direction=asc", "gh api --paginate --slurp repos/agent0ai/agent-zero/pulls?state=all&per_page=100&sort=created&direction=asc", "gh api graphql repository(owner:agent0ai,name:agent-zero) discussions(first:100)", "gh api repos/agent0ai/agent-zero/releases?per_page=100" ], "note": "Pinned-denominator read-back: 834 issues (63 open, 771 closed), 897 pull requests (82 open, 815 closed), all 94 discussions, and 72 observed releases." }, { "class": "fak_selfquery_witness", "evidence": [ "fak capabilities \"hierarchical agent delegation subordinate contexts task scheduling\"", "fak capabilities \"semantic memory vector retrieval knowledge recall\"", "fak capabilities \"agent plugin extension lifecycle hooks tools\"", "fak capabilities \"computer browser shell tool execution policy witness\"" ], "note": "Four candidate-specific self-queries were recorded in the map and adjudicated against fak-native alternatives." }, { "class": "candidate_matrix", "evidence": [ "docs/research/inventory/agent0ai-agent-zero.json#candidate-matrix" ], "note": "Five candidates were compared to the real next-best fak alternative; none survived as a new dispatchable issue." }, { "class": "issue_tracking", "evidence": [ "https://github.com/anthony-chaudhary/fak/issues/8987" ], "note": "No additional issue was filed because every candidate was reject or stay-minimal." } ], "issue_refs": [ "#8987" ] } }, { "repository": "llm-d/llm-d", "url": "https://github.com/llm-d/llm-d", "status": "studied", "priority": 9, "why": "Kubernetes-native distributed-inference assembly deeply studied and exhaustively re-inventoried at v0.9-era main for pluginized endpoint routing, exact KV-event locality, flow control, P/D topology, workload-variant autoscaling, and executable guide contracts. FAK keeps policy/model/cache ownership and interoperates rather than duplicating the Kubernetes control plane; the live-cluster compatibility witness remains open.", "last_checked": "2026-08-25", "checked_revision": "3243fcf1191348b55c7811267a98117f8b7a6910", "stars_at_check": 4153, "last_push_at_check": "2026-08-26T02:04:17Z", "study_note": "docs/notes/llm-d-study-2026-08-18.md", "refresh_issue": "#8988", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/llm-d-llm-d.json", "indexed_revision": "3243fcf1191348b55c7811267a98117f8b7a6910", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 11, "candidate_count": 4, "filed_issue_count": 0, "completeness_critic": "exhaustive pinned 1,129-file tree map plus GitHub aggregate and paginated issue, pull-request, discussion, release, tag, branch, milestone, project, roadmap, and history read-back; Apache-2.0 provenance, fak self-query, candidate adjudication, and issue tracking are recorded under non_tree_study; skipped control directory: .git", "source_evidence": [ { "class": "open_closed_issues_prs_discussions", "evidence": [ "gh api graphql -f query=", "gh api --paginate --slurp repos/llm-d/llm-d/issues?state=all&per_page=100&sort=created&direction=asc", "gh pr list --repo llm-d/llm-d --state all --limit 1000 --json number,title,state,createdAt,closedAt,mergedAt,labels,url", "gh api graphql repository(owner:llm-d,name:llm-d) discussions(first:100)", "gh api --paginate --slurp repos/llm-d/llm-d/releases?per_page=100", "gh api --paginate --slurp repos/llm-d/llm-d/tags?per_page=100" ], "note": "Exact aggregates at observation: 533 issues (124 open), 1,841 PRs (100 open), zero discussions, 11 releases, and 66 tags." }, { "class": "fak_selfquery_witness", "evidence": [ "fak capabilities \"Kubernetes distributed inference endpoint routing KV cache locality flow control prefill decode disaggregation autoscaling\"", "fak capabilities \"llm-d inference scheduler endpoint picker KV events prefix cache compatibility\"", "fak capabilities \"vLLM Kubernetes gateway inference extension workload variant autoscaling\"" ], "note": "FAK owns policy/model choice, context reuse, live control, and evidence; it does not duplicate llm-d Kubernetes endpoint selection, autoscaling, or deployment assembly." }, { "class": "candidate_matrix", "evidence": [ "docs/research/inventory/llm-d-llm-d.json#non_tree_study.candidate_matrix", "docs/notes/llm-d-study-2026-08-18.md#exhaustive-inventory-refresh-2026-08-25" ] }, { "class": "issue_tracking", "evidence": [ "https://github.com/anthony-chaudhary/fak/issues/8988", "https://github.com/anthony-chaudhary/fak/issues/8016", "https://github.com/anthony-chaudhary/fak/issues/8017", "https://github.com/anthony-chaudhary/fak/issues/6082", "https://github.com/anthony-chaudhary/fak/issues/3888" ], "note": "Ownership contract #8016 shipped; compatibility, benchmark, and cache-warm follow-ons remain #8017, #6082, and #3888. No duplicate issue was filed." } ], "issue_refs": [ "#8988", "#8016", "#8017", "#6082", "#3888" ] } }, { "repository": "deepseek-ai/deepseek-harness", "url": "https://github.com/deepseek-ai/deepseek-harness", "status": "studied", "priority": 10, "why": "Plugin-first agent harness deeply studied for reversible effect ownership, transactional composition replacement, durable model-visible events, standing session generations, guarded filesystem identity/version semantics, and real-composition tests. One net-new FAK gap survived as #8222; the other strongest candidates map to existing work or deliberate static-binary/default-deny divergences.", "last_checked": "2026-08-25", "checked_revision": "141eb6fef83422698aef7a981029e843e8161534", "stars_at_check": 172283, "last_push_at_check": "2026-08-19T15:37:57Z", "study_note": "docs/notes/CONCEPT-STUDY-DEEPSEEK-HARNESS-2026-08-20.md", "refresh_issue": "#8989 (inventory refresh complete)", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/deepseek-ai-deepseek-harness.json", "indexed_revision": "141eb6fef83422698aef7a981029e843e8161534", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 13, "candidate_count": 13, "filed_issue_count": 13, "completeness_critic": "exhaustive local tree map at dsh-v0.1.0-rc.8 plus full pinned history/tag/release and GraphQL discussion read-back; GitHub reports Issues and Pull Requests disabled (zero each), while 3,310 discussions existed by the pinned commit timestamp; fak self-query, candidate adjudication, and issue tracking are recorded under non_tree_study; skipped dependency/cache/control directories: .git, vendor", "source_evidence": [ { "class": "open_closed_issues_prs_discussions", "evidence": [ "gh api graphql -f query=\"repository issues pullRequests discussions\"", "https://github.com/deepseek-ai/deepseek-harness/discussions" ], "note": "At the pinned cutoff GitHub Issues and Pull Requests were disabled (zero each); all 3,310 discussions were read back: 3,226 open and 84 closed." }, { "class": "fak_selfquery_witness", "evidence": [ "fak capabilities \"plugin effect ownership transactional composition replacement durable model-visible turns standing session generations\"", "fak capabilities \"filesystem identity version semantics symlink containment\"", "fak capabilities \"real composition integration tests plugin harness lifecycle\"" ] }, { "class": "candidate_matrix", "evidence": [ "docs/research/inventory/deepseek-ai-deepseek-harness.json#non_tree_study.candidate_matrix", "docs/notes/CONCEPT-STUDY-DEEPSEEK-HARNESS-2026-08-20.md#candidate-to-fak-decision-matrix" ] }, { "class": "issue_tracking", "evidence": [ "https://github.com/anthony-chaudhary/fak/issues/8989", "https://github.com/anthony-chaudhary/fak/issues/8222", "https://github.com/anthony-chaudhary/fak/issues/8260", "https://github.com/anthony-chaudhary/fak/issues/8262" ], "note": "The map records the complete #8262-#8272 range; #8264 and #8265 remain open, all other follow-ons are closed." } ], "issue_refs": [ "#8989", "#8222", "#8260", "#8262", "#8263", "#8264", "#8265", "#8266", "#8267", "#8268", "#8269", "#8270", "#8271", "#8272" ] } }, { "repository": "rdi-berkeley/agents-last-exam", "url": "https://github.com/rdi-berkeley/agents-last-exam", "status": "studied", "priority": 11, "why": "Broad artifact-scored professional-work benchmark deeply studied and exhaustively refreshed for official hidden-reference grading, run-local ALE-Claw gateway composition, typed trajectories, timeout-aware artifact scoring, cross-harness controls, and provider operating-envelope risks. The official raw-vs-fak spine remains #8223; trajectory fidelity remains #6629.", "last_checked": "2026-08-25", "checked_revision": "0b6465b13c85b5a0a017d4c88bcf979a519a5e1f", "stars_at_check": 970, "last_push_at_check": "2026-08-25T22:20:05Z", "study_note": "docs/notes/CONCEPT-STUDY-AGENTS-LAST-EXAM-2026-08-20.md", "refresh_issue": "#8990", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/rdi-berkeley-agents-last-exam.json", "indexed_revision": "0b6465b13c85b5a0a017d4c88bcf979a519a5e1f", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 11, "candidate_count": 4, "filed_issue_count": 2, "completeness_critic": "exhaustive pinned tree map plus complete GitHub repository, issue, pull-request, discussion, release, tag, branch, and history read-back; FAK self-query, candidate adjudication, and issue tracking are recorded under non_tree_study; skipped control and vendored dependency/cache directories are explicit", "source_evidence": [ { "class": "history_changelog_releases", "evidence": [ "git -C _scratch/study-ale-8990/ale log --all", "https://github.com/rdi-berkeley/agents-last-exam/releases/tag/runtime-qe-bgw-6.7.0-4.0", "gh api repos/rdi-berkeley/agents-last-exam/releases?per_page=100", "gh api repos/rdi-berkeley/agents-last-exam/tags?per_page=100" ], "note": "Full pinned clone history plus the sole release and tag were inspected; 173 commits were visible." }, { "class": "open_closed_issues_prs_discussions", "evidence": [ "gh api --paginate repos/rdi-berkeley/agents-last-exam/issues?state=all&per_page=100", "gh api graphql repository(owner:rdi-berkeley,name:agents-last-exam) issues/pullRequests/discussions", "https://github.com/rdi-berkeley/agents-last-exam/issues", "https://github.com/rdi-berkeley/agents-last-exam/pulls" ], "note": "Complete read-back: 16 issues, 63 pull requests, and 0 discussions at observation." }, { "class": "roadmap_todos", "evidence": [ "docs/research/inventory/rdi-berkeley-agents-last-exam.json#non_tree_study.history_and_roadmap", "https://github.com/rdi-berkeley/agents-last-exam/issues/7", "https://github.com/rdi-berkeley/agents-last-exam/issues/26", "https://github.com/rdi-berkeley/agents-last-exam/issues/27", "https://github.com/rdi-berkeley/agents-last-exam/issues/68", "https://github.com/rdi-berkeley/agents-last-exam/issues/75" ], "note": "No dedicated roadmap file at the pin; tree TODOs, open forge work, and remote branches were reviewed as the roadmap surface." }, { "class": "fak_selfquery_witness", "evidence": [ "fak capabilities \"artifact-scored professional-work benchmark hidden-reference grading typed trajectories timeout harness comparison\"", "fak capabilities \"benchmark entry external agent harness run receipt quality score operating envelope\"", "fak capabilities \"trajectory import artifact evaluator cross harness benchmark\"" ], "note": "No ALE-shaped benchmark or trajectory-import capability was found; existing #8223 and #6629 remain the owned gaps." }, { "class": "candidate_matrix", "evidence": [ "docs/research/inventory/rdi-berkeley-agents-last-exam.json#non_tree_study.candidate_matrix" ] }, { "class": "issue_tracking", "evidence": [ "https://github.com/anthony-chaudhary/fak/issues/8990", "https://github.com/anthony-chaudhary/fak/issues/8223", "https://github.com/anthony-chaudhary/fak/issues/6629" ], "note": "The refresh issue and both surviving prior follow-ons are explicit; no duplicate issue was filed." } ], "issue_refs": [ "#8990", "#8223", "#6629" ] } }, { "repository": "llm-as-a-verifier/llm-as-a-verifier", "url": "https://github.com/llm-as-a-verifier/llm-as-a-verifier", "status": "studied", "priority": 12, "why": "Training-free trajectory selector deeply studied at its Terminal-Bench 2.1 self-verification seam. The candidate-pool Pass@1/oracle columns reproduce, but verifier means/error bars lack committed score/result/uncertainty artifacts and current source has cache/tournament correctness gaps. FAK already owns live best-of-N and evidence hierarchy; one provenance-bound offline selector replay survived as #8230.", "last_checked": "2026-08-20", "checked_revision": "8db8a114355a9d7fdf9a8d1d5c87f6aeebd18770", "stars_at_check": 2438, "last_push_at_check": "2026-08-20T07:01:30Z", "study_note": "docs/notes/CONCEPT-STUDY-LLM-AS-A-VERIFIER-2026-08-20.md", "refresh_issue": "#8991", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/llm-as-a-verifier-llm-as-a-verifier.json", "indexed_revision": "8db8a114355a9d7fdf9a8d1d5c87f6aeebd18770", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 6, "candidate_count": 12, "filed_issue_count": 1, "completeness_critic": "Pinned local tree map walked all 938 regular files and skipped only .git; architecture/design, tests/fixtures, and roadmap paths were checked absent. Complete forge read-back at 2026-08-25T00:00:00Z covered 9 issues, 13 pull requests, zero discussions, zero releases, and zero tags; FAK self-query, candidate adjudication, and issue ownership are recorded in the refreshed note.", "source_evidence": [ { "class": "open_closed_issues_prs_discussions", "evidence": [ "gh api graphql repository(owner:llm-as-a-verifier,name:llm-as-a-verifier) issues(first:100,states:[OPEN,CLOSED]) pullRequests(first:100,states:[OPEN,CLOSED,MERGED]) discussions(first:100)", "gh api repos/llm-as-a-verifier/llm-as-a-verifier/releases --paginate", "gh api repos/llm-as-a-verifier/llm-as-a-verifier/tags --paginate" ], "note": "Complete read-back at 2026-08-25T00:00:00Z: 9 issues, 13 pull requests, 0 discussions, 0 releases, 0 tags." }, { "class": "fak_selfquery_witness", "evidence": [ "fak capabilities \"offline provenance-bound trajectory selection verifier replay\"", "fak capabilities \"best of N candidate selection evidence hierarchy uncertainty\"", "fak capabilities \"verifier score cache content identity contamination\"", "fak capabilities \"pairwise tournament ring selection candidate aggregation\"" ], "note": "Queries found native provenance/cache/routing/evidence substrates but no pairwise tournament capability; the note records why that remains gated." }, { "class": "candidate_matrix", "evidence": [ "docs/notes/CONCEPT-STUDY-LLM-AS-A-VERIFIER-2026-08-20.md#candidate-matrix", "docs/notes/CONCEPT-STUDY-LLM-AS-A-VERIFIER-2026-08-20.md#refresh-candidate-matrix-and-follow-on-decisions" ] }, { "class": "issue_tracking", "evidence": [ "https://github.com/anthony-chaudhary/fak/issues/8230", "https://github.com/anthony-chaudhary/fak/issues/4235", "https://github.com/anthony-chaudhary/fak/issues/8991" ], "note": "#8230 owns the surviving replay spine, #4235 owns optional tournament work, and #8991 owns this exhaustive refresh; no new issue survived." } ], "issue_refs": [ "#8230", "#4235", "#8991" ] } }, { "repository": "hiraditya/hiraditya.github.io", "url": "https://github.com/hiraditya/hiraditya.github.io", "status": "studied", "priority": 13, "why": "The There Is No Address essay and its editorial history were studied as a cross-vendor disaggregation warning: public transport addresses stop before consumer-private locality. FAK mitigates the split with a consumer-owned route-by-reference, terminal receipts, and full-route pricing; #8259 remains open and #8261 shipped.", "last_checked": "2026-08-25", "checked_revision": "2547c55f35391c210d274fa7ea36fcbc383b5d8a", "stars_at_check": 2, "last_push_at_check": "2026-08-20T16:34:48Z", "study_note": "docs/notes/CONCEPT-STUDY-THERE-IS-NO-ADDRESS-2026-08-20.md", "refresh_issue": "#8992", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/hiraditya-hiraditya.github.io.json", "indexed_revision": "2547c55f35391c210d274fa7ea36fcbc383b5d8a", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 11, "candidate_count": 5, "filed_issue_count": 2, "completeness_critic": "Exhaustive pinned tree walk plus forge issues/PRs/discussions, release/tag/history, roadmap/TODO, provenance, FAK self-query, candidate decisions, and issue tracking read-back; no required source class remains unexamined.", "source_evidence": [ { "class": "open_closed_issues_prs_discussions", "evidence": [ "gh api --paginate 'repos/hiraditya/hiraditya.github.io/issues?state=all&per_page=100'", "gh api --paginate 'repos/hiraditya/hiraditya.github.io/pulls?state=all&per_page=100'", "gh api graphql repository(owner:hiraditya,name:hiraditya.github.io) discussions(first:100)" ], "note": "Read to exhaustion through the pinned revision timestamp: two open issue-only post drafts, 15 closed PRs, one merged PR, and no discussions." }, { "class": "fak_selfquery_witness", "evidence": [ "fak capabilities \"consumer-owned public private route reference terminal ready receipt full-route pricing\"", "fak capabilities \"topology route selection bytes estimated ready time fabric transfer\"" ], "note": "Native routing surfaced; direct read-back keeps terminal readiness in #8259 and recognizes shipped route pricing in #8261." }, { "class": "candidate_matrix", "evidence": [ "docs/research/inventory/hiraditya-hiraditya.github.io.json#non_tree_study.candidate_matrix" ], "note": "Five candidates are filed-existing, shipped-existing, deduplicated, watch-only, or excluded." }, { "class": "issue_tracking", "evidence": [ "https://github.com/anthony-chaudhary/fak/issues/8992", "https://github.com/anthony-chaudhary/fak/issues/8259", "https://github.com/anthony-chaudhary/fak/issues/8261" ], "note": "The refresh created no new follow-on; #8259 remains open and #8261 is closed." } ], "issue_refs": [ "#8992", "#8259", "#8261" ] } }, { "repository": "ai-dynamo/dynamo", "url": "https://github.com/ai-dynamo/dynamo", "status": "studied", "priority": 14, "why": "Dynamo is the closest active orchestration reference for KV-aware routing, disaggregated serving, cache visibility, replay, and whole-agent scheduling. The exhaustive refresh found six bounded FAK-native integration seams while preserving FAK ownership of Qwen3.8 execution, kernels, memory, scheduling, cache, adaptation, and operations.", "last_checked": "2026-08-26", "checked_revision": "f494601ef16b2a17b35ca340c656287b2787971a", "stars_at_check": 7871, "last_push_at_check": "2026-08-26T21:58:09Z", "study_note": "docs/notes/CONCEPT-STUDY-DYNAMO-EXHAUSTIVE-2026-08-26.md", "refresh_issue": "#9271", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/ai-dynamo-dynamo.json", "corpus_path": "docs/research/inventory/ai-dynamo-dynamo-corpus.jsonl", "indexed_revision": "f494601ef16b2a17b35ca340c656287b2787971a", "observed_at": "2026-08-26T22:01:04Z", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 20, "candidate_count": 8, "filed_issue_count": 6, "completeness_critic": "Exhaustive 5,401-file pinned tree map plus GraphQL cursor-exhausted issue and pull-request history, discussions, releases/tags, TODO/provenance scans, FAK self-query, deterministic candidate adjudication, and open/closed issue routing. Large review/comment/reply bodies use aggregate coverage receipts; LFS payloads and semantically unread generated/vendor/fixture paths remain explicit boundaries.", "source_evidence": [ { "class": "open_closed_issues_prs_discussions", "evidence": [ "gh api graphql --paginate --slurp -f query='repository issues(first:100, after:$endCursor, orderBy:CREATED_AT ASC)'", "gh api graphql --paginate --slurp -f query='repository pullRequests(first:100, after:$endCursor, orderBy:CREATED_AT ASC)'", "gh api graphql --paginate --slurp -f query='repository discussions(first:100, after:$endCursor, orderBy:CREATED_AT ASC)'", "gh api --paginate --slurp repos/ai-dynamo/dynamo/releases?per_page=100", "gh api --paginate --slurp repos/ai-dynamo/dynamo/tags?per_page=100", "docs/research/inventory/ai-dynamo-dynamo-corpus.jsonl" ], "note": "At the 2026-08-26T22:01:04Z cutoff: 1,867 issues, 11,902 pull requests, 19 discussions, 45 releases, and 85 tags. GraphQL issue/PR/discussion cursors were exhausted; REST page-number samples are supporting evidence only." }, { "class": "fak_selfquery_witness", "evidence": [ "fak capabilities tier-weighted cache overlap worker routing", "fak capabilities anticipated decode output footprint live load routing", "fak capabilities cache event source identity consolidation first store last remove", "fak capabilities request input trigger tool result retry gateway ingress receipt", "fak capabilities atomic route selection occupancy booking cancellation fallback" ], "note": "Direct Go reference searches followed each self-query and exposed three closed-ticket helpers with no production caller." }, { "class": "candidate_matrix", "evidence": [ "docs/research/inventory/ai-dynamo-dynamo.json#non_tree_study.candidate_matrix", "docs/notes/CONCEPT-STUDY-DYNAMO-EXHAUSTIVE-2026-08-26.md#deterministic-candidate-priority-map" ], "note": "Eight scored queue entries: three construction-ready new seams, three closed-ticket integration decisions, and two existing-ticket joins." }, { "class": "issue_tracking", "evidence": [ "https://github.com/anthony-chaudhary/fak/issues/9271", "https://github.com/anthony-chaudhary/fak/issues/8772", "https://github.com/anthony-chaudhary/fak/issues/5272", "https://github.com/anthony-chaudhary/fak/issues/5274", "https://github.com/anthony-chaudhary/fak/issues/6419", "https://github.com/anthony-chaudhary/fak/issues/8259", "https://github.com/anthony-chaudhary/fak/issues/8261", "https://github.com/anthony-chaudhary/fak/issues/8993", "https://github.com/anthony-chaudhary/fak/issues/9307", "https://github.com/anthony-chaudhary/fak/issues/9308", "https://github.com/anthony-chaudhary/fak/issues/9309", "https://github.com/anthony-chaudhary/fak/issues/9310", "https://github.com/anthony-chaudhary/fak/issues/9311", "https://github.com/anthony-chaudhary/fak/issues/9312" ], "note": "Broad open/closed FAK search plus mechanism-specific dedup produced six construction tickets (#9307-#9312), while two candidates remain joins to existing work." } ], "issue_refs": [ "#9271", "#8772", "#5272", "#5274", "#6419", "#8259", "#8261", "#8993", "#5276", "#5277", "#3888", "#2242", "#3259", "#3310", "#3413", "#6409", "#9307", "#9308", "#9309", "#9310", "#9311", "#9312" ], "study_receipt": "study_a5808ac41de4fbff614b981d1e4950c1e0caa458c56fc31ca0365127fac1ec82" } }, { "repository": "ai-dynamo/nixl", "url": "https://github.com/ai-dynamo/nixl", "status": "studied", "priority": 14, "why": "NIXL was exhaustively studied at its descriptor, registration, remote-metadata, backend-selection, lifecycle, live-activation, telemetry, and release seams. It is the transfer adapter beneath the essay, not an address-free universal memory model; FAK should govern consumer-ready route admission, pricing, and receipts above it.", "last_checked": "2026-08-25", "checked_revision": "b0cbb237354d72b83500d5214d2b4484f9866fa3", "stars_at_check": 1216, "last_push_at_check": "2026-08-25T23:47:18Z", "study_note": "docs/notes/CONCEPT-STUDY-THERE-IS-NO-ADDRESS-2026-08-20.md", "refresh_issue": "#8993", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/ai-dynamo-nixl.json", "indexed_revision": "b0cbb237354d72b83500d5214d2b4484f9866fa3", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 14, "candidate_count": 5, "filed_issue_count": 0, "completeness_critic": "Exhaustive 776-file local map at the pinned revision plus paged GitHub issue/PR history through the commit cutoff, zero-discussion read-back, releases/tags, TODO and unsupported signals, provenance, FAK self-query, candidate adjudication, and issue routing; generated classifications were manually corrected with representative source reads.", "source_evidence": [ { "class": "open_closed_issues_prs_discussions", "evidence": [ "gh api --paginate --slurp repos/ai-dynamo/nixl/issues?state=all&per_page=100&sort=created&direction=asc", "gh pr list --repo ai-dynamo/nixl --state all --limit 1000 --json number,title,state,createdAt,updatedAt,mergedAt,url", "gh api graphql repository(owner:ai-dynamo,name:nixl) discussions totalCount" ], "note": "At the commit cutoff: 250 issues (67 open, 183 closed), 1,904 PRs (198 open, 1,706 closed), and zero discussions." }, { "class": "fak_selfquery_witness", "evidence": [ "fak capabilities \"transfer fabric descriptors registration remote metadata backend route admission receipts\"", "fak capabilities \"GPU memory registration lifecycle transfer telemetry tracing\"", "fak capabilities \"async agent metadata exchange topology capability negotiation\"", "fak capabilities \"NIXL backend plugin data movement connector\"" ], "note": "Queries confirmed native FAK routing and capability-floor ownership while preserving NIXL as an optional transfer substrate." }, { "class": "candidate_matrix", "evidence": [ "docs/research/inventory/ai-dynamo-nixl.json#non_tree_study.candidate_matrix" ] }, { "class": "issue_tracking", "evidence": [ "https://github.com/anthony-chaudhary/fak/issues/8993", "https://github.com/anthony-chaudhary/fak/issues/8259", "https://github.com/anthony-chaudhary/fak/issues/8261" ], "note": "No new borrow issue: candidates map to #8259/#8261, deduplicate to #3310/#6409, or watch upstream PR #2137." } ], "issue_refs": [ "#8993", "#8259", "#8261", "#3310", "#6409" ] } }, { "repository": "z-lab/dflash", "url": "https://github.com/z-lab/dflash", "status": "studied", "priority": 15, "why": "DFlash 2 reference inference and benchmark surface deeply studied across its candidate path selector, dynamic local convolution, exact verification, model cards, runtime ports, papers, releases, and reported H200 results. FAK should ride the trained architecture; per-position acceptance telemetry survived as #8258 and the still-absent stochastic losslessness owner #4202 was reopened.", "last_checked": "2026-08-20", "checked_revision": "07ebd93db9f472af339b644bb70221ad8428328a", "stars_at_check": 5809, "last_push_at_check": "2026-08-18T21:14:10Z", "study_note": "docs/notes/CONCEPT-STUDY-DFLASH2-2026-08-20.md", "refresh_issue": "#8994", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/z-lab-dflash.json", "indexed_revision": "07ebd93db9f472af339b644bb70221ad8428328a", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "source_evidence": [ { "class": "history_changelog_releases", "evidence": [ "gh api 'repos/z-lab/dflash/commits?per_page=100'", "gh api 'repos/z-lab/dflash/releases?per_page=100'", "https://github.com/z-lab/dflash/releases/tag/v0.1.0" ], "note": "Complete pinned commit history plus the sole tag/release checked on 2026-08-25; no changelog file exists." }, { "class": "open_closed_issues_prs_discussions", "evidence": [ "gh issue list --repo z-lab/dflash --state open --limit 200 --json number,title,state,createdAt,closedAt,url", "gh issue list --repo z-lab/dflash --state closed --limit 200 --json number,title,state,createdAt,closedAt,url", "gh pr list --repo z-lab/dflash --state all --limit 200 --json number,title,state,createdAt,closedAt,mergedAt,url", "gh api graphql repository discussions(first:100) and state totals" ], "note": "Full forge state classes checked on 2026-08-25: issues 85 open/48 closed; PRs 11 open/8 closed/3 merged; 7 discussions; zero milestones." }, { "class": "roadmap_todos", "evidence": [ "gh api 'repos/z-lab/dflash/git/trees/07ebd93db9f472af339b644bb70221ad8428328a?recursive=1'", "gh api 'repos/z-lab/dflash/contents/README.md?ref=07ebd93db9f472af339b644bb70221ad8428328a'", "gh issue list --repo z-lab/dflash --state all --limit 200 --json number,title,state,createdAt,closedAt,url", "gh pr list --repo z-lab/dflash --state all --limit 200 --json number,title,state,createdAt,closedAt,mergedAt,url", "gh api graphql repository discussions(first:100), milestones totals" ], "note": "No canonical roadmap, milestone, or source TODO exists; requests remain distributed across forge threads." }, { "class": "fak_selfquery_witness", "evidence": [ "fak capabilities 'DFlash block diffusion flash speculative decoding candidate path selector dynamic convolution exact verification per-position acceptance'", "go run ./cmd/fak-dev index docs 'DFlash block diffusion flash speculative decoding candidate path selector dynamic convolution exact verification per-position acceptance'", "go run ./cmd/fak-dev index leaves 'DFlash block diffusion flash speculative decoding candidate path selector dynamic convolution exact verification per-position acceptance'", "fak capabilities 'stochastic speculative decoding rejection sampling lossless temperature acceptance'" ], "note": "Self-query found the existing DFlash study, shipped per-position acceptance work, and open stochastic-losslessness gap; no native DFlash architecture implementation was found." }, { "class": "candidate_matrix", "evidence": [ "docs/notes/CONCEPT-STUDY-DFLASH2-2026-08-20.md#candidate-matrix-refresh" ] }, { "class": "issue_tracking", "evidence": [ "https://github.com/anthony-chaudhary/fak/issues/8258", "https://github.com/anthony-chaudhary/fak/issues/4202", "https://github.com/anthony-chaudhary/fak/issues/8936", "https://github.com/anthony-chaudhary/fak/issues/8994" ] } ], "subsystem_count": 3, "candidate_count": 8, "filed_issue_count": 0, "issue_refs": [ "#8258", "#4202", "#8936", "#8994" ], "completeness_critic": "Exhaustive 11-file/14-entry pinned tree map plus complete dated forge state-class read-back; architecture is embedded in README/source, tests and roadmap are confirmed absent, and every process-owned source class points to concrete evidence." } }, { "repository": "anthropics/claude-code", "url": "https://github.com/anthropics/claude-code", "status": "studied", "priority": 16, "why": "Public plugin/workflow evidence and proprietary-license limits remain inventoried; the 2026-08-29 refresh adds mixed-version daemon, persisted-writer, alias, and update behavior as INSPIRE-ONLY compatibility evidence for the harness ABI study.", "last_checked": "2026-08-29", "checked_revision": "f1af9b1f4b1fd4c776135381606edada82ef638e", "stars_at_check": 143374, "last_push_at_check": "2026-08-28T18:19:32Z", "study_note": "docs/notes/CONCEPT-HARNESS-ABI-STABILITY-2026-08-29.md", "refresh_issue": "#10020", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/anthropics-claude-code.json", "indexed_revision": "005c5dade90c2c59c88d819d8723e7b579addb5e", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 10, "candidate_count": 4, "filed_issue_count": 0, "completeness_critic": "Deterministic exhaustive local-tree map at the pinned revision plus dated GitHub counts and targeted issue/PR searches. The public repository is a plugin/example/documentation distribution, not the proprietary Claude Code runtime; discussions are disabled, tests are limited to public plugin scripts/examples, and those boundaries are explicit rather than inferred as absence.", "source_evidence": [ { "class": "open_closed_issues_prs_discussions", "evidence": [ "gh api graphql repository(owner:anthropics,name:claude-code) issue/pullRequest/release/discussion totalCount and state counts", "gh api search/issues for agent workflow, subagent, worktree, task resume, background agent, plugins, and hooks" ], "note": "Observed 86,934 issues, 1,276 pull requests, 199 releases, and zero discussions (disabled) on 2026-08-25; targeted searches sampled the orchestration/recovery seams without pretending every issue body is pinned tree content." }, { "class": "fak_selfquery_witness", "evidence": [ "fak capabilities \"bounded concurrent coding agent fleet leases independent witness reconciliation\"", "fak capabilities \"agent task messaging cancellation resume recovery\"", "fak capabilities \"plugins specialist agents workflow composition\"", "fak index docs|leaves|verbs|claims \"ultracode workflow coordinator recovery claude code\"" ], "note": "Capability queries plus lexical tree/index fallback classify the candidates as owned, owned-with-follow-ons, or present at a different license/runtime boundary." }, { "class": "candidate_matrix", "evidence": [ "docs/research/inventory/anthropics-claude-code.json#non_tree_study.candidate_matrix" ] }, { "class": "issue_tracking", "evidence": [ "https://github.com/anthony-chaudhary/fak/issues/8995", "#8168", "#8334", "#8488", "#8490" ], "note": "No duplicate borrow issue was filed; the existing follow-ons remain the complete actionable set." } ], "issue_refs": [ "#8995", "#8168", "#8334", "#8488", "#8490" ] } }, { "repository": "langchain-ai/langgraph", "url": "https://github.com/langchain-ai/langgraph", "status": "studied", "priority": 17, "why": "Pregel task preparation, deterministic writes, bounded concurrency, checkpoint lineage, pending-write idempotency, and selective resume were studied as public implementation evidence. Fak should borrow selective recovery while retaining stronger ownership and independent effect proof.", "last_checked": "2026-08-25", "checked_revision": "f09cfe8ffc1eeffd68f4b628ed69c30f7cad229f", "stars_at_check": 40204, "last_push_at_check": "2026-08-21T20:04:35Z", "study_note": "docs/notes/CONCEPT-STUDY-ULTRACODE-WORKFLOWS-2026-08-21.md", "refresh_issue": "#8996", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/langchain-ai-langgraph.json", "indexed_revision": "f09cfe8ffc1eeffd68f4b628ed69c30f7cad229f", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 5, "candidate_count": 3, "filed_issue_count": 0, "completeness_critic": "exhaustive pinned tree map plus complete paged GitHub issue, pull-request, discussion, release, tag, branch, and local history read-back; fak self-query, candidate adjudication, and issue tracking are recorded under non_tree_study; skipped control directory: .git", "source_evidence": [ { "class": "open_closed_issues_prs_discussions", "evidence": [ "gh api --paginate --slurp repos/langchain-ai/langgraph/issues?state=all&per_page=100&sort=created&direction=asc", "gh api --paginate --slurp repos/langchain-ai/langgraph/pulls?state=all&per_page=100&sort=created&direction=asc", "gh api graphql repository(owner:langchain-ai,name:langgraph) discussions(first:100)", "gh api --paginate --slurp repos/langchain-ai/langgraph/releases?per_page=100", "gh api --paginate --slurp repos/langchain-ai/langgraph/tags?per_page=100", "gh api --paginate --slurp repos/langchain-ai/langgraph/branches?per_page=100" ], "note": "Complete 2026-08-25 read-back: 1,534 issues, 5,642 pull requests, zero discussions, 560 releases, 731 tags, and eight branches; tree evidence remains pinned to f09cfe8f." }, { "class": "fak_selfquery_witness", "evidence": [ "fak capabilities \"Pregel deterministic writes checkpoint lineage pending write idempotency selective resume independent effect proof\"", "fak capabilities \"durable execution interrupt resume checkpoint recovery task writes concurrency\"", "fak capabilities \"LangGraph graph orchestration state machine human in loop\"" ], "note": "Queries found native carry-forward, live-control, context-reuse, attribution, fleet-monitoring, and capability-floor surfaces; existing issues own witnessed resume and node observability." }, { "class": "candidate_matrix", "evidence": [ "docs/research/inventory/langchain-ai-langgraph.json#non_tree_study.candidate_matrix" ] }, { "class": "issue_tracking", "evidence": [ "https://github.com/anthony-chaudhary/fak/issues/8996", "https://github.com/anthony-chaudhary/fak/issues/8490", "https://github.com/anthony-chaudhary/fak/issues/8334", "https://github.com/anthony-chaudhary/fak/issues/8168" ], "note": "No duplicate borrow issue filed; #8334 remains the sole open LangGraph-derived follow-on." } ], "issue_refs": [ "#8996", "#8490", "#8334", "#8168" ] } }, { "repository": "a2aproject/A2A", "url": "https://github.com/a2aproject/A2A", "status": "studied", "priority": 18, "why": "Stable cross-vendor task, context, lifecycle, message, artifact, cancellation, authorization, and snapshot-before-resubscribe semantics were studied as a public standard. Fak should interoperate at this envelope and keep planning, leases, effects, and witness above it.", "last_checked": "2026-08-25", "checked_revision": "16ba52690519bf55b9388e34d4db356efa88aa51", "stars_at_check": 25448, "last_push_at_check": "2026-08-21T19:46:58Z", "study_note": "docs/notes/CONCEPT-STUDY-ULTRACODE-WORKFLOWS-2026-08-21.md", "refresh_issue": "#8997", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/a2aproject-a2a.json", "indexed_revision": "16ba52690519bf55b9388e34d4db356efa88aa51", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 10, "candidate_count": 4, "filed_issue_count": 0, "completeness_critic": "Exhaustive pinned tree and history map plus cutoff-filtered GitHub issues/PRs, current discussion denominator with historical limitation explicit, releases, roadmap/TODOs, Apache-2.0 provenance, FAK self-query, candidate decisions, and issue tracking; no dedicated in-tree tests directory and linked SDK repositories are explicit boundaries.", "source_evidence": [ { "class": "open_closed_issues_prs_discussions", "evidence": [ "gh api --paginate --slurp repos/a2aproject/A2A/issues?state=all&per_page=100&sort=created&direction=asc", "gh api --paginate --slurp repos/a2aproject/A2A/pulls?state=all&per_page=100&sort=created&direction=asc", "gh api graphql repository(owner:a2aproject,name:A2A) discussions(first:100)" ], "note": "At the 2026-08-18T17:09:55Z pin: 756 issues and 908 PRs by creation cutoff; all 359 current discussions denominator-checked on 2026-08-25 because GitHub cannot reconstruct historical discussion state." }, { "class": "fak_selfquery_witness", "evidence": [ "fak capabilities \"agent interoperability protocol task state message artifact streaming cancellation push notification authorization agent card\"", "fak capabilities \"A2A agent protocol task lifecycle context message artifact cancellation authorization streaming resubscribe\"", "fak capabilities \"A2A AgentCard resubscribe task lifecycle artifact\"" ], "note": "FAK owns lifecycle/cancellation/authorization/witness above transport and snapshot-before-resume internally; it does not claim a public A2A wire binding." }, { "class": "candidate_matrix", "evidence": [ "docs/research/inventory/a2aproject-a2a.json#non_tree_study.candidate_matrix" ] }, { "class": "issue_tracking", "evidence": [ "https://github.com/anthony-chaudhary/fak/issues/8997", "https://github.com/anthony-chaudhary/fak/issues/8936" ], "note": "No new borrow issue: two candidates are owned, and binding/conformance work remains a separately scoped deferred spine." } ], "issue_refs": [ "#8997", "#8936" ] } }, { "repository": "Mchicao/endless", "url": "https://github.com/Mchicao/endless", "status": "studied", "priority": 18, "why": "Experimental Codex open-turn continuation analogue: one typed wait_for_user_input rendezvous keeps repeated operator messages inside one provider turn. The narrow candidate is a capability-gated, policy-preserving matched experiment; the Bun TUI, Codex-specific host, parallel worker stack, and approval shortcuts are excluded.", "last_checked": "2026-08-25", "checked_revision": "bbf449ca624aca90d8cda0bc21100f2a73a56bf4", "stars_at_check": 2, "last_push_at_check": "2026-08-21T14:38:07Z", "study_note": "docs/notes/CONCEPT-STUDY-ENDLESS-2026-08-24.md", "refresh_issue": "#9004", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/mchicao-endless.json", "indexed_revision": "bbf449ca624aca90d8cda0bc21100f2a73a56bf4", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 7, "candidate_count": 3, "filed_issue_count": 0, "completeness_critic": "exhaustive pinned tree map plus complete GitHub issue, pull-request, discussion, release, tag, branch, and history read-back; fak self-query, candidate adjudication, and issue tracking are recorded under non_tree_study; skipped control directory: .git", "source_evidence": [ { "class": "open_closed_issues_prs_discussions", "evidence": [ "gh api --paginate --slurp repos/Mchicao/endless/issues?state=all&per_page=100&sort=created&direction=asc", "gh pr list --repo Mchicao/endless --state all --limit 1000 --json number,title,state,createdAt,closedAt,mergedAt,labels,url", "gh api graphql repository(owner:Mchicao,name:endless) discussions(first:100)", "gh api repos/Mchicao/endless/releases?per_page=100", "gh api repos/Mchicao/endless/tags?per_page=100" ], "note": "Exhaustive GitHub read-back returned zero issues, pull requests, discussions, releases, and tags." }, { "class": "fak_selfquery_witness", "evidence": [ "fak capabilities \"open provider turn wait for user input rendezvous session continuation policy capability\"", "fak capabilities \"operator input during one model turn pause resume control live session\"", "fak capabilities \"parallel worker terminal tui codex host approval bypass\"" ], "note": "Queries found native capability-floor, turn-tax, live-control, context-reuse, carry-forward, and fleet-monitoring surfaces; no equivalent typed in-provider-turn user rendezvous." }, { "class": "candidate_matrix", "evidence": [ "docs/research/inventory/mchicao-endless.json#non_tree_study.candidate_matrix" ] }, { "class": "issue_tracking", "evidence": [ "https://github.com/anthony-chaudhary/fak/issues/9004" ], "note": "All candidates were rejected or already owned; no borrow issue survived." } ], "issue_refs": [ "#9004" ] } }, { "repository": "openai/codex", "url": "https://github.com/openai/codex", "status": "studied", "priority": 19, "why": "The exhaustive provider-session inventory remains intact; the 2026-08-29 compatibility refresh adds stable/experimental schema separation, bidirectional current/released executor tests, configuration migration, build provenance, and unresolved mixed-version failures.", "last_checked": "2026-08-29", "checked_revision": "6478a751fde8884b2fdc76486fe23175a8e795d4", "stars_at_check": 119758, "last_push_at_check": "2026-08-29T10:54:29Z", "study_note": "docs/notes/CONCEPT-HARNESS-ABI-STABILITY-2026-08-29.md", "refresh_issue": "#10020", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/openai-codex.json", "indexed_revision": "2764e83626efe55f64e04d153fc99a157327f3c2", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 14, "candidate_count": 4, "filed_issue_count": 0, "completeness_critic": "Pinned local inventory walked all 6,613 non-.git regular files across 14 immediate subsystems; GitHub issues, PRs, discussions, releases, milestones, and repository metadata were audited separately at the observation date. Architecture evidence is workspace-derived because no canonical architecture document was found; hosted product docs and live-provider execution remain outside this source-level thread-boundary study. Four material candidates were explicitly kept or rejected, and no unmet FAK implementation follow-on remained.", "source_evidence": [ { "class": "open_closed_issues_prs_discussions", "evidence": [ "gh api graphql repository(owner:openai,name:codex) issue/PR/discussion totals observed 2026-08-26", "https://github.com/openai/codex/pull/40751", "https://github.com/openai/codex/pull/17499" ], "note": "Point-in-time GraphQL read-back covered open and closed issues, open/closed/merged PRs, and all discussions; targeted reset history found no open defect changing the conclusion." }, { "class": "fak_selfquery_witness", "evidence": [ "fak capabilities \"provider session reset thread boundary\"", "cmd/fak/provider_session_reset_test.go", "cmd/fak/provider_session_reset_more_test.go", "docs/_witnesses/issue-8218-codex-session-boundary.json" ], "note": "FAK already has provider-neutral boundary tests and a Codex session-boundary witness; no missing implementation seam was found." }, { "class": "candidate_matrix", "evidence": [ "docs/integrations/provider-session-reset.md#field-borrow-candidate-matrix-and-fak-decisions" ], "note": "Four material candidates were explicitly kept, rejected, or assigned monitor-only follow-on." }, { "class": "issue_tracking", "evidence": [ "https://github.com/anthony-chaudhary/fak/issues/8998", "https://github.com/anthony-chaudhary/fak/issues/8936" ], "note": "The refresh child and parent own the study; no new implementation issue survived adjudication." } ], "issue_refs": [ "#8998", "#8936" ] } }, { "repository": "google-gemini/gemini-cli", "url": "https://github.com/google-gemini/gemini-cli", "status": "studied", "priority": 20, "why": "The exhaustive provider-session inventory remains intact; the 2026-08-29 compatibility refresh adds stable hook DTO translators, extension update rollback, tolerant configuration behavior, and open event-spelling/unit-migration negative knowledge.", "last_checked": "2026-08-29", "checked_revision": "0bd1d439751478771c45d3d0895a6a9760554bf4", "stars_at_check": 106741, "last_push_at_check": "2026-08-29T01:56:42Z", "study_note": "docs/notes/CONCEPT-HARNESS-ABI-STABILITY-2026-08-29.md", "refresh_issue": "#10020", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/google-gemini-gemini-cli.json", "indexed_revision": "64b5b79a6dd89ea96e65cb761c23aae6c0b33ce4", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 18, "candidate_count": 3, "filed_issue_count": 0, "completeness_critic": "exhaustive local tree map plus pinned GitHub history/release/issue/PR/discussion metadata, roadmap and license audit, FAK self-query, candidate adjudication, and issue tracking; skipped dependency/cache/control directories: .git, packages/core/vendor", "source_evidence": [ { "class": "open_closed_issues_prs_discussions", "evidence": [ "gh api graphql -f query=", "gh api repos/google-gemini/gemini-cli/commits/64b5b79a6dd89ea96e65cb761c23aae6c0b33ce4", "gh api repos/google-gemini/gemini-cli/license", "gh api repos/google-gemini/gemini-cli/releases?per_page=5" ], "note": "14,309 issues, 12,659 pull requests, 662 discussions, and 601 releases at the observation cutoff; detailed state counts and method are in the map." }, { "class": "fak_selfquery_witness", "evidence": [ "fak capabilities \"provider session reset clear new session lifecycle event\"", "fak capabilities \"session checkpoint resume conversation persistence trace continuity\"", "fak capabilities \"MCP OAuth trust policy extensions hooks lifecycle security\"" ], "note": "Queries found FAK-owned portable-session, context-reuse, live-control, and capability-floor seams; no distinct unowned Gemini mechanism survived." }, { "class": "candidate_matrix", "evidence": [ "docs/research/inventory/google-gemini-gemini-cli.json#candidate-matrix" ] }, { "class": "issue_tracking", "evidence": [ "https://github.com/anthony-chaudhary/fak/issues/8999", "https://github.com/anthony-chaudhary/fak/issues/8220" ], "note": "Provider adaptation is already owned; remaining live dogfood is already tracked, so no new borrow issue was filed." } ], "issue_refs": [ "#8999", "#8220" ] } }, { "repository": "flashinfer-ai/flashinfer", "url": "https://github.com/flashinfer-ai/flashinfer", "status": "studied", "priority": 21, "why": "Production GPU-kernel control-plane analogue: reusable attention plans/workspace, correctness-gated profile tuning, JIT/artifact identity, and trace/replay mechanisms map directly to FAK compute HAL gaps.", "last_checked": "2026-08-28", "checked_revision": "93f4f2642e1b3680a52ebb51cf68e0fdad237796", "stars_at_check": 6273, "last_push_at_check": "2026-08-28T09:53:20Z", "study_note": "docs/notes/CONCEPT-STUDY-FLASHINFER-REFRESH-2026-08-28.md", "refresh_issue": "#9757", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/flashinfer-ai-flashinfer.json", "indexed_revision": "93f4f2642e1b3680a52ebb51cf68e0fdad237796", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 20, "candidate_count": 30, "filed_issue_count": 1, "completeness_critic": "Generated exhaustive main-tree inventory plus validated forge corpus, exact eight-commit delta, divergent v0.6.18rc10 branch/tag review, NOTICE/licenses/gitlinks, FAK self-query, and live issue outcome read-back. External private dashboards are unavailable and were not treated as evidence.", "source_evidence": [ { "class": "history_changelog_releases", "evidence": [ "docs/research/inventory/flashinfer-ai-flashinfer-corpus-2026-08-28.jsonl", "docs/research/flashinfer-2026-08-28/main-delta.json", "docs/research/flashinfer-2026-08-28/release-divergence.json" ], "note": "Validated forge corpus, exact eight-commit main delta, and divergent v0.6.18rc10 history." }, { "class": "open_closed_issues_prs_discussions", "evidence": [ "docs/research/inventory/flashinfer-ai-flashinfer-corpus-2026-08-28.jsonl", "gh api --paginate repos/flashinfer-ai/flashinfer/issues?state=all&per_page=100", "gh pr list --repo flashinfer-ai/flashinfer --state all --limit 1000 --json number,title,state,createdAt,closedAt,mergedAt,labels,url", "gh api graphql repository(owner:flashinfer-ai,name:flashinfer) discussions(first:100)" ], "note": "Complete accepted issue/pull/discussion pages at the recorded sequential cutoff." }, { "class": "roadmap_todos", "evidence": [ "docs/research/flashinfer-2026-08-28/source-completeness.json" ], "note": "No dedicated roadmap file; tracker and delta review carry current direction." }, { "class": "license_provenance", "evidence": [ "docs/research/flashinfer-2026-08-28/license-provenance.json" ] }, { "class": "fak_selfquery_witness", "evidence": [ "docs/research/flashinfer-2026-08-28/self-query.json#fak-selfquery-witness" ] }, { "class": "candidate_matrix", "evidence": [ "docs/research/flashinfer-2026-08-28/candidate-ledger.json#candidate-matrix" ] }, { "class": "issue_tracking", "evidence": [ "docs/research/flashinfer-2026-08-28/dedup-map.json#issue-tracking" ] } ], "issue_refs": [ "#9757", "#1479", "#3254", "#4180", "#8607", "#8657", "#9137", "#9139", "#9209" ] } }, { "repository": "sgl-project/mini-sglang", "url": "https://github.com/sgl-project/mini-sglang", "status": "studied", "priority": 21, "why": "Compact modern LLM-serving control loop studied for the integration of radix-prefix reuse, page-aware KV allocation, bounded chunked prefill, continuous decode, stable tensor-parallel request order, CUDA-graph replay, and user-visible latency/throughput measurement. The resulting map reuses fak radixkv and grounds the missing Qwen3.8 serving spine in existing issue #8395.", "last_checked": "2026-08-25", "checked_revision": "9a91cfafe754aa85daee49998176275667eb58f2", "stars_at_check": 4794, "last_push_at_check": "2026-05-17T12:37:42Z", "study_note": "docs/notes/CONCEPT-STUDY-MINI-SGLANG-2026-08-22.md", "refresh_issue": "#9000", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/sgl-project-mini-sglang.json", "indexed_revision": "9a91cfafe754aa85daee49998176275667eb58f2", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 6, "candidate_count": 6, "filed_issue_count": 0, "completeness_critic": "The generator walked all 121 files and 32 directories at the exact revision. The refresh separately paged all forge issues and pull requests, all releases, and all 171 commits through the revision timestamp; confirmed discussions disabled; inspected every repository TODO; verified MIT provenance; ran three candidate-specific fak self-queries; adjudicated six candidates; and reconciled issue tracking to #9000 and #8395. Dependency/cache/control directory skipped: .git only.", "source_evidence": [ { "class": "open_closed_issues_prs_discussions", "evidence": [ "gh api --paginate --slurp repos/sgl-project/mini-sglang/issues?state=all&per_page=100&sort=created&direction=asc", "gh pr list --repo sgl-project/mini-sglang --state all --limit 1000 --json number,title,state,createdAt,closedAt,mergedAt,labels,url", "gh api graphql repository(owner:sgl-project,name:mini-sglang) discussions(first:100)" ], "note": "Pinned-timestamp exhaustive read-back: 25 issues, 123 pull requests; discussions disabled." }, { "class": "fak_selfquery_witness", "evidence": [ "fak capabilities \"radix prefix cache page aware KV allocation chunked prefill continuous batching\"", "fak capabilities \"tensor parallel stable request order CUDA graph replay serving scheduler\"", "fak capabilities \"time to first token inter token latency throughput benchmark serving\"" ], "note": "Candidate-specific queries found native cache/context, routing, benchmarking, and capability-floor surfaces; serving work remains in #8395." }, { "class": "candidate_matrix", "evidence": [ "docs/research/inventory/sgl-project-mini-sglang.json#candidate-matrix" ] }, { "class": "issue_tracking", "evidence": [ "https://github.com/anthony-chaudhary/fak/issues/9000", "https://github.com/anthony-chaudhary/fak/issues/8395" ], "note": "No new borrow issue survived candidate adjudication." } ], "issue_refs": [ "#9000", "#8395" ] } }, { "repository": "fla-org/flash-linear-attention", "url": "https://github.com/fla-org/flash-linear-attention", "status": "studied", "priority": 22, "why": "FLA's MIT naive, chunked-prefill, and fused-recurrent GDN implementations were studied as an independent correctness oracle for fak's resident Qwen sequence work. The Python/PyTorch/Triton runtime and broad architecture catalog are excluded; packed multi-request state and graph capture remain watch items behind existing #8344 and #8391.", "last_checked": "2026-08-25", "checked_revision": "8f787def80f6b6862f4f8b84581810d3db537c2d", "stars_at_check": 5623, "last_push_at_check": "2026-08-25T20:24:58Z", "study_note": "docs/notes/CONCEPT-STUDY-FLASH-LINEAR-ATTENTION-2026-08-22.md", "refresh_issue": "#9001", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/fla-org-flash-linear-attention.json", "indexed_revision": "8f787def80f6b6862f4f8b84581810d3db537c2d", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "source_evidence": [ { "class": "open_closed_issues_prs_discussions", "evidence": [ "gh api --paginate repos/fla-org/flash-linear-attention/issues?state=all&per_page=100", "gh pr list --repo fla-org/flash-linear-attention --state all --limit 100 --json number,title,state,createdAt,updatedAt,url", "gh api --paginate repos/fla-org/flash-linear-attention/releases?per_page=100", "gh api repos/fla-org/flash-linear-attention/milestones?state=all", "gh api graphql repository discussions first 10" ], "note": "On 2026-08-25, audited 1,161 issues/PRs (90 open, 1,071 closed), 21 releases, three open milestones, and all six discussions." }, { "class": "fak_selfquery_witness", "evidence": [ "go run ./cmd/fak capabilities 'GDN gated delta net chunked prefill fused recurrent final state cu_seqlens resident sequence'", "go run ./cmd/fak-dev index docs 'GDN gated delta net chunked prefill fused recurrent final state cu_seqlens resident sequence'", "go run ./cmd/fak-dev index leaves 'GDN gated delta net chunked prefill fused recurrent final state cu_seqlens resident sequence'", "go run ./cmd/fak capabilities 'packed variable length multi request recurrent state offsets graph capture GDN'", "go run ./cmd/fak-dev index docs 'packed variable length multi request recurrent state offsets graph capture GDN'", "go run ./cmd/fak-dev index leaves 'packed variable length multi request recurrent state offsets graph capture GDN'", "go run ./cmd/fak capabilities 'Triton PyTorch linear attention architecture catalog TileLang ROCm AMD backend'" ], "note": "Self-query confirmed the resident GDN implementation and existing packed-state/graph follow-ons; it found no reason to import FLA runtime or backend machinery." }, { "class": "candidate_matrix", "evidence": [ "docs/notes/CONCEPT-STUDY-FLASH-LINEAR-ATTENTION-2026-08-22.md#refreshed-candidate-matrix-and-fak-decisions" ] }, { "class": "issue_tracking", "evidence": [ "https://github.com/anthony-chaudhary/fak/issues/9001", "https://github.com/anthony-chaudhary/fak/issues/8344", "https://github.com/anthony-chaudhary/fak/issues/8391" ] } ], "subsystem_count": 12, "candidate_count": 5, "filed_issue_count": 0, "issue_refs": [ "#9001", "#8344", "#8391" ], "completeness_critic": "739/739 tracked paths mapped at the pinned revision; README/design/runtime/tests/history/releases/issues/PRs/discussions/roadmap/license and FAK self-query were read; every candidate has an adopt/adapt/monitor/reject decision and every actionable follow-on is tied to an existing issue." } }, { "repository": "SemiAnalysisAI/InferenceX", "url": "https://github.com/SemiAnalysisAI/InferenceX", "status": "studied", "priority": 22, "why": "Continuous inference benchmark analogue: AgentX client/server lifecycle metrics, quality-constrained interactivity, scope-versioned measured power, artifact reuse and ingestion integrity map to FAK benchmark receipt gaps.", "last_checked": "2026-08-24", "checked_revision": "0b0138fd7de0a6f927f9769b19d594d01f586107", "stars_at_check": 1480, "last_push_at_check": "2026-08-24T18:19:19Z", "study_note": "docs/notes/BORROW-BENCHMARK-SERVING-METRICS-INFERENCEX-STUDY-2026-07-13.md", "refresh_issue": "#9003", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/semianalysisai-inferencex.json", "indexed_revision": "0b0138fd7de0a6f927f9769b19d594d01f586107", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 11, "candidate_count": 5, "filed_issue_count": 3, "completeness_critic": "exhaustive local tree map plus pinned-cutoff GitHub issue, pull-request, discussion, release, history, roadmap/TODO, and license/provenance read-back; candidate-specific fak self-query, adjudication, and issue tracking are recorded under non_tree_study; skipped dependency/cache/control directories: .git", "source_evidence": [ { "class": "open_closed_issues_prs_discussions", "evidence": [ "gh issue list --repo SemiAnalysisAI/InferenceX --state all --limit 1000 --json ...", "gh api --paginate repos/SemiAnalysisAI/InferenceX/pulls?state=all&per_page=100", "gh api repos/SemiAnalysisAI/InferenceX/discussions?per_page=100", "gh api repos/SemiAnalysisAI/InferenceX/releases?per_page=100" ], "note": "Pinned-cutoff read-back: 389 issues, 2,299 pull requests, all 4 discussions, and the sole release." }, { "class": "fak_selfquery_witness", "evidence": [ "fak capabilities \"benchmark serving metrics receipt time to first token inter token latency power energy quality\"", "fak capabilities \"artifact reuse ingestion integrity benchmark lineage source revision\"", "fak capabilities \"agentic continuous inference client server lifecycle judge task success\"" ], "note": "Candidate-specific self-query confirmed native receipt, artifact, routing, trajectory, and session-control surfaces; remaining applicable gaps are already tracked." }, { "class": "candidate_matrix", "evidence": [ "docs/research/inventory/semianalysisai-inferencex.json#non_tree_study.candidate_matrix" ] }, { "class": "issue_tracking", "evidence": [ "https://github.com/anthony-chaudhary/fak/issues/9003", "https://github.com/anthony-chaudhary/fak/issues/8773", "https://github.com/anthony-chaudhary/fak/issues/8774", "https://github.com/anthony-chaudhary/fak/issues/8775" ], "note": "No novel unowned borrow survived; applicable findings remain deduplicated to the three existing open follow-ons." } ], "issue_refs": [ "#9003", "#8773", "#8774", "#8775" ] } }, { "repository": "vllm-project/vllm", "url": "https://github.com/vllm-project/vllm", "status": "studied", "priority": 1, "why": "Primary high-throughput serving reference for scheduler, prefix-cache, memory-allocation, speculative-decode, and determinism tradeoffs. Deeply studied for engineering process failures (step overhead accretion, flaky hardware matrix CI, cache refcount leaks).", "last_checked": "2026-09-03", "checked_revision": "a56654d6de060495ff2db3b1d9ff0b187084d1a9", "stars_at_check": 90799, "last_push_at_check": "2026-09-02T19:43:08Z", "study_note": "docs/notes/CONCEPT-STUDY-PERFORMANCE-OSS-PROCESS-FAILURES-2026-09-03.md" }, { "repository": "sgl-project/sglang", "url": "https://github.com/sgl-project/sglang", "status": "candidate", "priority": 24, "why": "Primary serving reference for RadixAttention, HiCache/disaggregation, scheduler policy, and current request-state correctness. Earlier SGLang studies exist, but the live tree is moving fast; use this row for a current-source refresh instead of relying on July anchors.", "last_checked": "2026-08-25", "checked_revision": "edff717ef0106b4413b371bf7a05a5193ffeee85", "stars_at_check": 32437, "last_push_at_check": "2026-08-25T20:42:49Z" }, { "repository": "LMCache/LMCache", "url": "https://github.com/LMCache/LMCache", "status": "candidate", "priority": 25, "why": "Fast-moving KV-cache layer with vLLM/Dynamo integration, GPU transfer registration fixes, and operational cache lifecycle seams. Previous LMCache studies should be refreshed at this exact revision before any new L3/KV offload backlog is filed.", "last_checked": "2026-08-25", "checked_revision": "23cca67908e17b193eb8fab08ba1beb0115881cd", "stars_at_check": 11364, "last_push_at_check": "2026-08-25T18:39:04Z" }, { "repository": "NVIDIA/Model-Optimizer", "url": "https://github.com/NVIDIA/Model-Optimizer", "status": "candidate", "priority": 26, "why": "Apache-2.0 model optimization suite covering quantization, distillation, pruning, and speculative decoding export paths into TensorRT-LLM, vLLM, and peers. Candidate for Qwen3.8 quantization and borrowed-hot-path planning under #8923 without copying runtime code blindly.", "last_checked": "2026-08-25", "checked_revision": "73d778422388f0e849ecb180375d34ac445711ca", "stars_at_check": 3483, "last_push_at_check": "2026-08-25T20:22:29Z" }, { "repository": "vllm-project/speculators", "url": "https://github.com/vllm-project/speculators", "status": "studied", "priority": 27, "why": "Unified speculative-decoding algorithm library for vLLM, directly relevant to DFlash/DSpark/MTP candidate selection and the native Qwen3.8 speculation roadmap. No existing fak issue matched this exact source identity in the bounded dedupe scan.", "last_checked": "2026-08-25", "checked_revision": "0faffeb3bd547b4451a978d7aaf26a2f01b83d62", "stars_at_check": 762, "last_push_at_check": "2026-08-25T20:59:51Z", "study_note": "docs/notes/CONCEPT-STUDY-SPECULATORS-2026-08-25.md", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/vllm-project-speculators.json", "indexed_revision": "0faffeb3bd547b4451a978d7aaf26a2f01b83d62", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "source_evidence": [ { "class": "open_closed_issues_prs_discussions", "evidence": [ "gh issue list --repo vllm-project/speculators --state all --limit 20 --json number,title,state,createdAt,updatedAt,labels,url", "gh pr list --repo vllm-project/speculators --state all --limit 20 --json number,title,state,mergedAt,updatedAt,url,labels", "gh pr view 1038 --repo vllm-project/speculators --json number,title,state,mergedAt,body,url,files,commits", "gh release list --repo vllm-project/speculators --limit 20", "gh api graphql repository discussions first 10" ], "note": "Recent issues, PRs, merged loading fix, releases, tags, and discussions checked on 2026-08-25." }, { "class": "fak_selfquery_witness", "evidence": [ "go run ./cmd/fak capabilities 'DSpark confidence head cumulative acceptance product dynamic draft length thresholding per-position accept probability'", "go run ./cmd/fak-dev index docs 'DSpark confidence head cumulative acceptance product dynamic draft length thresholding per-position accept probability'", "go run ./cmd/fak-dev index leaves 'DSpark confidence head cumulative acceptance product dynamic draft length thresholding per-position accept probability'", "go run ./cmd/fak capabilities 'P-EAGLE parallel multi-token prediction COD sampling multi-depth speculative drafter one forward'", "go run ./cmd/fak-dev index docs 'P-EAGLE parallel multi-token prediction COD sampling multi-depth speculative drafter one forward'", "go run ./cmd/fak-dev index leaves 'P-EAGLE parallel multi-token prediction COD sampling multi-depth speculative drafter one forward'", "go run ./cmd/fak capabilities 'checkpoint tensor alias final norm wrong alias speculative acceptance regression shortest suffix resolver'" ], "note": "Self-query found per-position acceptance and broad speculation docs, but no calibrated DSpark controller, P-EAGLE-shaped draft witness, or speculator alias-drift canary." }, { "class": "candidate_matrix", "evidence": [ "docs/notes/CONCEPT-STUDY-SPECULATORS-2026-08-25.md#candidate-matrix" ] }, { "class": "issue_tracking", "evidence": [ "https://github.com/anthony-chaudhary/fak/issues/8938", "https://github.com/anthony-chaudhary/fak/issues/8939", "https://github.com/anthony-chaudhary/fak/issues/8940", "https://github.com/anthony-chaudhary/fak/issues/3197#issuecomment-5416891936", "https://github.com/anthony-chaudhary/fak/issues/8923#issuecomment-5416891900" ] } ], "subsystem_count": 10, "candidate_count": 7, "filed_issue_count": 3, "issue_refs": [ "#8938", "#8939", "#8940", "#3197", "#8923", "#8258", "#5154", "#4202" ], "completeness_critic": "exhaustive local map plus README/docs/source/tests/license read; recent open and closed issues, PRs including #1038, releases, tags, and discussions checked; no material source class left only asserted" } }, { "repository": "NVIDIA/kvpress", "url": "https://github.com/NVIDIA/kvpress", "status": "candidate", "priority": 28, "why": "Apache-2.0 KV-cache compression library with current entropy-gated chunk compression work. Existing issue #5328 already tracks the idea-scout lead; this row pins the current source for a real study pass before deciding whether #2240-style KV quant/compression gaps need refinement.", "last_checked": "2026-08-25", "checked_revision": "161705a68f64df329c88a6da5f20300a89aa7542", "stars_at_check": 1191, "last_push_at_check": "2026-08-18T10:31:20Z" }, { "repository": "nokia-applied-research/SPECTRA", "url": "https://github.com/nokia-applied-research/SPECTRA", "status": "studied", "priority": 28, "why": "MIT research artifact for activation-weighted low-rank KV transforms, reverse-water-filling mixed rates, and output-projection healing. The deep study separates its demonstrated memory/quality frontier from its unproven serving-throughput path and binds the surviving borrow to a fak-native Qwen3.8 oracle.", "last_checked": "2026-08-28", "checked_revision": "272032275106dc5944fbfa7091a1ceb403fa7e28", "stars_at_check": 1, "last_push_at_check": "2026-08-12T00:53:48Z", "study_note": "docs/notes/CONCEPT-STUDY-SPECTRA-2026-08-28.md", "inventory": { "mode": "exhaustive", "indexed_revision": "272032275106dc5944fbfa7091a1ceb403fa7e28", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 8, "candidate_count": 8, "filed_issue_count": 1, "source_evidence": [ { "class": "architecture_design", "evidence": [ "https://arxiv.org/abs/2608.07915v1" ], "note": "Version 1 source and PDF read; method, tables, appendices, calibration/evaluation envelope, and claim boundaries reconciled." }, { "class": "open_closed_issues_prs_discussions", "evidence": [ "gh issue list --state all --repo nokia-applied-research/SPECTRA", "gh pr list --state all --repo nokia-applied-research/SPECTRA", "gh api graphql repository discussions nokia-applied-research/SPECTRA" ], "note": "Projects v2 was not enumerable with the available token and is explicitly excluded from the absence claim." }, { "class": "fak_selfquery_witness", "evidence": [ "fak capabilities \"spectral latent KV reverse water-filling W_O healing fused latent attention\"", "internal/model/kvquant.go" ], "note": "Queries returned no exact card; raw code proves existing Q4+recovery and MLA strengths plus the absent transform/rate-allocation seam." }, { "class": "candidate_matrix", "evidence": [ "docs/notes/CONCEPT-STUDY-SPECTRA-2026-08-28.md#candidate-matrix" ] }, { "class": "issue_tracking", "evidence": [ "https://github.com/anthony-chaudhary/fak/issues/9785", "https://github.com/anthony-chaudhary/fak/issues/9789", "study_13eb7b60b128c49c8575e840bcf36ef0ae625c4a314fb5d36551794c9046c8c4" ] } ], "issue_refs": [ "#9785", "#9789", "#9141", "#9161", "#2236" ], "completeness_critic": "Paper source/PDF, all load-bearing code, README/demos/eval/bench/plot surfaces, all assets, exact tree, both commits, GitHub metadata and collaboration surfaces, license, FAK self-query, candidate dispositions, dedupe, and issue contract were covered; Projects v2 token scope, dependency repos, and the noncanonical fork are the declared exclusions." } }, { "repository": "novitalabs/pegaflow", "url": "https://github.com/novitalabs/pegaflow", "status": "candidate", "priority": 29, "why": "High-performance KV-cache storage with GPU offload, SSD caching, and cross-node sharing via RDMA for vLLM/SGLang. Candidate for cache-tier and route-by-reference comparisons, especially where FAK must distinguish native-owned KV from external-engine KV.", "last_checked": "2026-08-25", "checked_revision": "d3ed31a7a4ee9bc7cdc5f112f0a1eb26efe63675", "stars_at_check": 195, "last_push_at_check": "2026-08-19T06:57:29Z" }, { "repository": "jundot/omlx", "url": "https://github.com/jundot/omlx", "status": "candidate", "priority": 30, "why": "Apple-Silicon inference server with continuous batching and SSD caching, relevant to FAK's Metal/Qwen3.8 resident-decode and consumer-hardware serving envelope. Dedupe against the DFlash2 oMLX fork note before filing any Mac performance gap.", "last_checked": "2026-08-25", "checked_revision": "90ecf1c26dbed875e6ced82c4faa6e9250037f2d", "stars_at_check": 20631, "last_push_at_check": "2026-08-25T15:51:19Z" }, { "repository": "lightseekorg/tokenspeed", "url": "https://github.com/lightseekorg/tokenspeed", "status": "studied", "priority": 31, "why": "MIT speed-of-light inference engine for agentic workloads deeply studied across scheduler (C++ FSM control plane, overlap scheduling with in-flight tracking), modular kernel registry with capability gating and format signatures, and MLA kernels on Blackwell. Clean-room FAK borrows decompose into #10743 (FSM), #10748 (overlap scheduling), #10749 (declarative registry), and #10750 (selection objectives) under parent #10741.", "last_checked": "2026-09-02", "checked_revision": "b174a3186d9a6eb3192389afbb25611a976eefc7", "stars_at_check": 1967, "last_push_at_check": "2026-08-30T00:00:00Z", "study_note": "docs/notes/CONCEPT-STUDY-TOKENSPEED-2026-09-02.md", "refresh_issue": "#10741", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/lightseekorg-tokenspeed.json", "indexed_revision": "b174a3186d9a6eb3192389afbb25611a976eefc7", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 12, "candidate_count": 12, "filed_issue_count": 4, "completeness_critic": "Exhaustive fan-out across scheduler (C++ FSM + Python), kernel registry/selection, MLA, entrypoint, event loop, models/layers. Skipped third-party vendored code (CUTLASS, FlashInfer, DeepEP). All load-bearing subsystems opened.", "issue_refs": [ "#10741", "#10743", "#10748", "#10749", "#10750" ] } }, { "repository": "Luce-Org/lucebox", "url": "https://github.com/Luce-Org/lucebox", "status": "candidate", "priority": 32, "why": "Apache-2.0 speculative inference server for consumer and heterogeneous hardware. Candidate for the product boundary around speculative serving knobs on non-datacenter devices; compare to existing DFlash/DeepSpec/mlx-dspark notes before filing.", "last_checked": "2026-08-25", "checked_revision": "c994209a2c03c1bf3426924aba52b5edc05501ff", "stars_at_check": 2788, "last_push_at_check": "2026-08-25T16:48:25Z" }, { "repository": "apache/maka", "url": "https://github.com/apache/maka", "status": "studied", "priority": 33, "why": "Local-first agent workspace whose single Runtime Host authority, immutable runtime facts, source-bound context projections, graph admission, and typed effect uncertainty offer narrowly re-derived lessons for FAK; current ASF DISCLAIMER-WIP makes all absent/partial borrows inspire-only.", "last_checked": "2026-08-28", "checked_revision": "2c066316545e432459e327734c634dd2cb61fa40", "stars_at_check": 3889, "last_push_at_check": "2026-08-28T19:06:56Z", "study_note": "docs/notes/CONCEPT-STUDY-APACHE-MAKA-2026-08-28.md" }, { "repository": "superplanehq/superplane", "url": "https://github.com/superplanehq/superplane", "status": "candidate", "priority": 34, "why": "Open-source control plane for agentic engineering with current issue-intake analysis work. Candidate for queue/reconciler planning under #8873/#8909 and for reducing operator polling in performance-focused agent fleets.", "last_checked": "2026-08-25", "checked_revision": "541a64b87aa534d5cbbbccb369d069e75a06bce9", "stars_at_check": 5503, "last_push_at_check": "2026-08-25T20:35:54Z" }, { "repository": "kubernetes-sigs/agent-sandbox", "url": "https://github.com/kubernetes-sigs/agent-sandbox", "status": "candidate", "priority": 35, "why": "Kubernetes-native isolated, stateful singleton workloads for AI agent runtimes and RL. Candidate for sandbox/runtime process isolation comparisons; dedupe against existing agent-sandbox governance and supervision issues before filing.", "last_checked": "2026-08-25", "checked_revision": "3444bfb35a42e933752d3c6624a24e00a3e53d6b", "stars_at_check": 3622, "last_push_at_check": "2026-08-25T20:37:46Z" }, { "repository": "ARahim3/mlx-dspark", "url": "https://github.com/ARahim3/mlx-dspark", "status": "candidate", "priority": 36, "why": "Prior July study filed #4199 from mlx-dspark's yield-per-cost control discipline. The current repo has moved substantially and now advertises Qwen3.8/Apple-Silicon serving additions, so any further speculative-decoding borrow should refresh against this exact revision.", "last_checked": "2026-08-25", "checked_revision": "89f4bb1629230833bb2d05653fa83052d5802b01", "stars_at_check": 601, "last_push_at_check": "2026-08-25T20:32:23Z", "study_note": "docs/notes/study-mlx-dspark-borrow-scout-2026-07-10.md" }, { "repository": "ollama/ollama", "url": "https://github.com/ollama/ollama", "status": "studied", "priority": 37, "why": "Hybrid production runtime: Go owns model graphs and orchestration while MLX/llama.cpp owns tensor execution; the exact boundary is a key native-inference comparison.", "last_checked": "2026-08-28", "checked_revision": "f96e7aa0513b9973a0ccc71be414c2ecb9d65b1a", "stars_at_check": 179656, "last_push_at_check": "2026-08-28T19:46:41Z", "study_note": "docs/notes/CONCEPT-STUDY-GO-INFERENCE-RUNTIMES-2026-08-28.md", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/go-inference-runtimes-github-2026-08-28.json", "indexed_revision": "f96e7aa0513b9973a0ccc71be414c2ecb9d65b1a", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 1, "candidate_count": 1, "filed_issue_count": 1, "completeness_critic": "Shared 23-repository manifest records the pinned runtime tree boundary, implementation/test anchors, activity and license/provenance, fak self-query, candidate dispositions, explicit exclusions, query ledger, and a family-level completeness critique.", "issue_refs": [ "#9846" ] } }, { "repository": "mudler/LocalAI", "url": "https://github.com/mudler/LocalAI", "status": "studied", "priority": 38, "why": "Go-governed backend marketplace: API, lifecycle, coalescing and supervision in Go over declared foreign inference workers.", "last_checked": "2026-08-28", "checked_revision": "0cdc31dcb30cf1464a7402b2065e4c9bc3fe3646", "stars_at_check": 48724, "last_push_at_check": "2026-08-28T20:04:54Z", "study_note": "docs/notes/CONCEPT-STUDY-GO-INFERENCE-RUNTIMES-2026-08-28.md", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/go-inference-runtimes-github-2026-08-28.json", "indexed_revision": "0cdc31dcb30cf1464a7402b2065e4c9bc3fe3646", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 1, "candidate_count": 1, "filed_issue_count": 1, "completeness_critic": "Shared 23-repository manifest records the pinned runtime tree boundary, implementation/test anchors, activity and license/provenance, fak self-query, candidate dispositions, explicit exclusions, query ledger, and a family-level completeness critique.", "issue_refs": [ "#9846" ] } }, { "repository": "ardanlabs/kronk", "url": "https://github.com/ardanlabs/kronk", "status": "studied", "priority": 39, "why": "Embeddable Go model lifecycle, batching and queueing over llama.cpp, whisper.cpp and diffusion native engines.", "last_checked": "2026-08-28", "checked_revision": "6fadebdfeb9e8c40dd3fd237fca2e8253cfcf51e", "stars_at_check": 766, "last_push_at_check": "2026-08-28T18:20:04Z", "study_note": "docs/notes/CONCEPT-STUDY-GO-INFERENCE-RUNTIMES-2026-08-28.md", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/go-inference-runtimes-github-2026-08-28.json", "indexed_revision": "6fadebdfeb9e8c40dd3fd237fca2e8253cfcf51e", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 1, "candidate_count": 1, "filed_issue_count": 1, "completeness_critic": "Shared 23-repository manifest records the pinned runtime tree boundary, implementation/test anchors, activity and license/provenance, fak self-query, candidate dispositions, explicit exclusions, query ledger, and a family-level completeness critique.", "issue_refs": [ "#9846" ] } }, { "repository": "born-ml/born", "url": "https://github.com/born-ml/born", "status": "studied", "priority": 40, "why": "Emerging native Go tensor and LLaMA runtime with Go/assembly CPU and Go-owned WebGPU/WGSL execution.", "last_checked": "2026-08-28", "checked_revision": "b1237ec5135d1108f56af6dfd5068638068ad8d3", "stars_at_check": 124, "last_push_at_check": "2026-08-25T11:51:06Z", "study_note": "docs/notes/CONCEPT-STUDY-GO-INFERENCE-RUNTIMES-2026-08-28.md", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/go-inference-runtimes-github-2026-08-28.json", "indexed_revision": "b1237ec5135d1108f56af6dfd5068638068ad8d3", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 1, "candidate_count": 1, "filed_issue_count": 1, "completeness_critic": "Shared 23-repository manifest records the pinned runtime tree boundary, implementation/test anchors, activity and license/provenance, fak self-query, candidate dispositions, explicit exclusions, query ledger, and a family-level completeness critique.", "issue_refs": [ "#9846" ] } }, { "repository": "townsendmerino/goinfer", "url": "https://github.com/townsendmerino/goinfer", "status": "studied", "priority": 41, "why": "Emerging native Go decoder runtime owning loaders, quantization, CPU execution and optional cgo-free accelerator paths.", "last_checked": "2026-08-28", "checked_revision": "bf6bdc3afc633ac8c118a7a982fc00c4ff8d2bdc", "stars_at_check": 8, "last_push_at_check": "2026-08-28T17:53:44Z", "study_note": "docs/notes/CONCEPT-STUDY-GO-INFERENCE-RUNTIMES-2026-08-28.md", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/go-inference-runtimes-github-2026-08-28.json", "indexed_revision": "bf6bdc3afc633ac8c118a7a982fc00c4ff8d2bdc", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 1, "candidate_count": 1, "filed_issue_count": 1, "completeness_critic": "Shared 23-repository manifest records the pinned runtime tree boundary, implementation/test anchors, activity and license/provenance, fak self-query, candidate dispositions, explicit exclusions, query ledger, and a family-level completeness critique.", "issue_refs": [ "#9846" ] } }, { "repository": "rostamlabs/rembed", "url": "https://github.com/rostamlabs/rembed", "status": "studied", "priority": 42, "why": "Focused native Go/assembly embedding runtime with SafeTensors/Hugging Face model support.", "last_checked": "2026-08-28", "checked_revision": "19d673b357bd8c244aa1a6d17d83769224fa98c9", "stars_at_check": 2, "last_push_at_check": "2026-08-26T21:24:58Z", "study_note": "docs/notes/CONCEPT-STUDY-GO-INFERENCE-RUNTIMES-2026-08-28.md", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/go-inference-runtimes-github-2026-08-28.json", "indexed_revision": "19d673b357bd8c244aa1a6d17d83769224fa98c9", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 1, "candidate_count": 1, "filed_issue_count": 1, "completeness_critic": "Shared 23-repository manifest records the pinned runtime tree boundary, implementation/test anchors, activity and license/provenance, fak self-query, candidate dispositions, explicit exclusions, query ledger, and a family-level completeness critique.", "issue_refs": [ "#9846" ] } }, { "repository": "magomedcoder/gogguf", "url": "https://github.com/magomedcoder/gogguf", "status": "studied", "priority": 43, "why": "Experimental pure-Go quantized GGUF CPU runtime with broad model claims and an explicit CGo CUDA boundary.", "last_checked": "2026-08-28", "checked_revision": "c157545050743f73356da7ea9db7532e9129013e", "stars_at_check": 4, "last_push_at_check": "2026-08-27T08:55:09Z", "study_note": "docs/notes/CONCEPT-STUDY-GO-INFERENCE-RUNTIMES-2026-08-28.md", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/go-inference-runtimes-github-2026-08-28.json", "indexed_revision": "c157545050743f73356da7ea9db7532e9129013e", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 1, "candidate_count": 1, "filed_issue_count": 1, "completeness_critic": "Shared 23-repository manifest records the pinned runtime tree boundary, implementation/test anchors, activity and license/provenance, fak self-query, candidate dispositions, explicit exclusions, query ledger, and a family-level completeness critique.", "issue_refs": [ "#9846" ] } }, { "repository": "nlpodyssey/spago", "url": "https://github.com/nlpodyssey/spago", "status": "studied", "priority": 44, "why": "Paused but substantial native Go/assembly tensor, autograd and NLP framework reference.", "last_checked": "2026-08-28", "checked_revision": "3130dda657565c0b6b8e494070d8c35a037eee4e", "stars_at_check": 1850, "last_push_at_check": "2025-04-01T10:17:42Z", "study_note": "docs/notes/CONCEPT-STUDY-GO-INFERENCE-RUNTIMES-2026-08-28.md", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/go-inference-runtimes-github-2026-08-28.json", "indexed_revision": "3130dda657565c0b6b8e494070d8c35a037eee4e", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 1, "candidate_count": 1, "filed_issue_count": 1, "completeness_critic": "Shared 23-repository manifest records the pinned runtime tree boundary, implementation/test anchors, activity and license/provenance, fak self-query, candidate dispositions, explicit exclusions, query ledger, and a family-level completeness critique.", "issue_refs": [ "#9846" ] } }, { "repository": "gorgonia/gorgonia", "url": "https://github.com/gorgonia/gorgonia", "status": "studied", "priority": 45, "why": "Mature native Go graph compiler and VM substrate, distinct from pretrained-model serving runtimes.", "last_checked": "2026-08-28", "checked_revision": "d7a3ce27c9a1ffbee531d7850485229584db8d97", "stars_at_check": 5930, "last_push_at_check": "2024-08-12T04:53:17Z", "study_note": "docs/notes/CONCEPT-STUDY-GO-INFERENCE-RUNTIMES-2026-08-28.md", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/go-inference-runtimes-github-2026-08-28.json", "indexed_revision": "d7a3ce27c9a1ffbee531d7850485229584db8d97", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 1, "candidate_count": 1, "filed_issue_count": 1, "completeness_critic": "Shared 23-repository manifest records the pinned runtime tree boundary, implementation/test anchors, activity and license/provenance, fak self-query, candidate dispositions, explicit exclusions, query ledger, and a family-level completeness critique.", "issue_refs": [ "#9846" ] } }, { "repository": "zerfoo/zerfoo", "url": "https://github.com/zerfoo/zerfoo", "status": "studied", "priority": 46, "why": "Broad native Go GGUF/model/runtime implementation whose rapid feature claims need independent operating-envelope verification.", "last_checked": "2026-08-28", "checked_revision": "a6c9dcbdc0117f929269c4d90bf4204c74afa372", "stars_at_check": 5, "last_push_at_check": "2026-08-24T01:03:45Z", "study_note": "docs/notes/CONCEPT-STUDY-GO-INFERENCE-RUNTIMES-2026-08-28.md", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/go-inference-runtimes-github-2026-08-28.json", "indexed_revision": "a6c9dcbdc0117f929269c4d90bf4204c74afa372", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 1, "candidate_count": 1, "filed_issue_count": 1, "completeness_critic": "Shared 23-repository manifest records the pinned runtime tree boundary, implementation/test anchors, activity and license/provenance, fak self-query, candidate dispositions, explicit exclusions, query ledger, and a family-level completeness critique.", "issue_refs": [ "#9846" ] } }, { "repository": "goccy/go-llama", "url": "https://github.com/goccy/go-llama", "status": "studied", "priority": 47, "why": "Static Go execution produced by compiling llama.cpp to Wasm and translating it to Go; provenance differs from Go-authored kernels.", "last_checked": "2026-08-28", "checked_revision": "6b608ae6947a4f6a27b2f05f85cb0c56ac3253b9", "stars_at_check": 150, "last_push_at_check": "2026-08-28T12:30:56Z", "study_note": "docs/notes/CONCEPT-STUDY-GO-INFERENCE-RUNTIMES-2026-08-28.md", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/go-inference-runtimes-github-2026-08-28.json", "indexed_revision": "6b608ae6947a4f6a27b2f05f85cb0c56ac3253b9", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 1, "candidate_count": 1, "filed_issue_count": 1, "completeness_critic": "Shared 23-repository manifest records the pinned runtime tree boundary, implementation/test anchors, activity and license/provenance, fak self-query, candidate dispositions, explicit exclusions, query ledger, and a family-level completeness critique.", "issue_refs": [ "#9846" ] } }, { "repository": "gomlx/gomlx", "url": "https://github.com/gomlx/gomlx", "status": "studied", "priority": 48, "why": "Active Go model/graph framework with separable pure-Go, OpenXLA/PJRT and ONNX Runtime execution routes.", "last_checked": "2026-08-28", "checked_revision": "f847c57c9a4e10c1c66c737bbc44625fd4d1e538", "stars_at_check": 1624, "last_push_at_check": "2026-08-28T11:27:46Z", "study_note": "docs/notes/CONCEPT-STUDY-GO-INFERENCE-RUNTIMES-2026-08-28.md", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/go-inference-runtimes-github-2026-08-28.json", "indexed_revision": "f847c57c9a4e10c1c66c737bbc44625fd4d1e538", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 1, "candidate_count": 1, "filed_issue_count": 1, "completeness_critic": "Shared 23-repository manifest records the pinned runtime tree boundary, implementation/test anchors, activity and license/provenance, fak self-query, candidate dispositions, explicit exclusions, query ledger, and a family-level completeness critique.", "issue_refs": [ "#9846" ] } }, { "repository": "knights-analytics/hugot", "url": "https://github.com/knights-analytics/hugot", "status": "studied", "priority": 49, "why": "Go transformer pipeline framework whose ORT, XLA and GoMLX-Go routes make per-backend evidence mandatory.", "last_checked": "2026-08-28", "checked_revision": "e08591494e5383c534e1c3cef92ac9d766e461b0", "stars_at_check": 638, "last_push_at_check": "2026-08-27T08:43:42Z", "study_note": "docs/notes/CONCEPT-STUDY-GO-INFERENCE-RUNTIMES-2026-08-28.md", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/go-inference-runtimes-github-2026-08-28.json", "indexed_revision": "e08591494e5383c534e1c3cef92ac9d766e461b0", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 1, "candidate_count": 1, "filed_issue_count": 1, "completeness_critic": "Shared 23-repository manifest records the pinned runtime tree boundary, implementation/test anchors, activity and license/provenance, fak self-query, candidate dispositions, explicit exclusions, query ledger, and a family-level completeness critique.", "issue_refs": [ "#9846" ] } }, { "repository": "yalue/onnxruntime_go", "url": "https://github.com/yalue/onnxruntime_go", "status": "studied", "priority": 50, "why": "Principal general ONNX Runtime Go binding in the corpus; CGo owns API crossing while ORT owns all computation.", "last_checked": "2026-08-28", "checked_revision": "8ce7fa5882bb3ac038bacdd53bfe02da3691899f", "stars_at_check": 704, "last_push_at_check": "2026-08-19T14:03:58Z", "study_note": "docs/notes/CONCEPT-STUDY-GO-INFERENCE-RUNTIMES-2026-08-28.md", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/go-inference-runtimes-github-2026-08-28.json", "indexed_revision": "8ce7fa5882bb3ac038bacdd53bfe02da3691899f", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 1, "candidate_count": 1, "filed_issue_count": 1, "completeness_critic": "Shared 23-repository manifest records the pinned runtime tree boundary, implementation/test anchors, activity and license/provenance, fak self-query, candidate dispositions, explicit exclusions, query ledger, and a family-level completeness critique.", "issue_refs": [ "#9846" ] } }, { "repository": "sugarme/gotch", "url": "https://github.com/sugarme/gotch", "status": "studied", "priority": 51, "why": "Go/libtorch binding that inherits PyTorch model and kernel breadth while accepting CGo and native ABI burden.", "last_checked": "2026-08-28", "checked_revision": "db661550e5a88582a40830e54bb1777e6afd32e7", "stars_at_check": 653, "last_push_at_check": "2025-02-25T22:41:10Z", "study_note": "docs/notes/CONCEPT-STUDY-GO-INFERENCE-RUNTIMES-2026-08-28.md", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/go-inference-runtimes-github-2026-08-28.json", "indexed_revision": "db661550e5a88582a40830e54bb1777e6afd32e7", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 1, "candidate_count": 1, "filed_issue_count": 1, "completeness_critic": "Shared 23-repository manifest records the pinned runtime tree boundary, implementation/test anchors, activity and license/provenance, fak self-query, candidate dispositions, explicit exclusions, query ledger, and a family-level completeness critique.", "issue_refs": [ "#9846" ] } }, { "repository": "oramasearch/onnx-go", "url": "https://github.com/oramasearch/onnx-go", "status": "studied", "priority": 52, "why": "ONNX decoder and graph importer with pluggable execution backends; format ownership is separate from computation.", "last_checked": "2026-08-28", "checked_revision": "5befeb870169e8cd10af28c1552ed14662f82390", "stars_at_check": 906, "last_push_at_check": "2024-09-02T01:46:08Z", "study_note": "docs/notes/CONCEPT-STUDY-GO-INFERENCE-RUNTIMES-2026-08-28.md", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/go-inference-runtimes-github-2026-08-28.json", "indexed_revision": "5befeb870169e8cd10af28c1552ed14662f82390", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 1, "candidate_count": 1, "filed_issue_count": 1, "completeness_critic": "Shared 23-repository manifest records the pinned runtime tree boundary, implementation/test anchors, activity and license/provenance, fak self-query, candidate dispositions, explicit exclusions, query ledger, and a family-level completeness critique.", "issue_refs": [ "#9846" ] } }, { "repository": "hybridgroup/yzma", "url": "https://github.com/hybridgroup/yzma", "status": "studied", "priority": 53, "why": "Current no-CGo llama.cpp binding reference: purego improves distribution while foreign native libraries retain compute ownership.", "last_checked": "2026-08-28", "checked_revision": "c4fed7865c4c5cb116d7ae9105a1765bc0398803", "stars_at_check": 573, "last_push_at_check": "2026-08-28T19:22:08Z", "study_note": "docs/notes/CONCEPT-STUDY-GO-INFERENCE-RUNTIMES-2026-08-28.md", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/go-inference-runtimes-github-2026-08-28.json", "indexed_revision": "c4fed7865c4c5cb116d7ae9105a1765bc0398803", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 1, "candidate_count": 1, "filed_issue_count": 1, "completeness_critic": "Shared 23-repository manifest records the pinned runtime tree boundary, implementation/test anchors, activity and license/provenance, fak self-query, candidate dispositions, explicit exclusions, query ledger, and a family-level completeness critique.", "issue_refs": [ "#9846" ] } }, { "repository": "go-skynet/go-llama.cpp", "url": "https://github.com/go-skynet/go-llama.cpp", "status": "studied", "priority": 54, "why": "Historical high-level CGo llama.cpp binding with a stale pinned native engine but useful API/provenance lessons.", "last_checked": "2026-08-28", "checked_revision": "6a8041ef6b46d4712afc3ae791d1c2d73da0ad1c", "stars_at_check": 937, "last_push_at_check": "2026-08-28T11:42:24Z", "study_note": "docs/notes/CONCEPT-STUDY-GO-INFERENCE-RUNTIMES-2026-08-28.md", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/go-inference-runtimes-github-2026-08-28.json", "indexed_revision": "6a8041ef6b46d4712afc3ae791d1c2d73da0ad1c", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 1, "candidate_count": 1, "filed_issue_count": 1, "completeness_critic": "Shared 23-repository manifest records the pinned runtime tree boundary, implementation/test anchors, activity and license/provenance, fak self-query, candidate dispositions, explicit exclusions, query ledger, and a family-level completeness critique.", "issue_refs": [ "#9846" ] } }, { "repository": "benedoc-inc/onnxer", "url": "https://github.com/benedoc-inc/onnxer", "status": "studied", "priority": 55, "why": "Emerging purego ORT/GenAI binding with multi-runtime isolation and mixed file-level licensing signals.", "last_checked": "2026-08-28", "checked_revision": "d3f86f897e42e0f5544494d62bfe5d8ead6370fd", "stars_at_check": 4, "last_push_at_check": "2026-08-07T13:52:28Z", "study_note": "docs/notes/CONCEPT-STUDY-GO-INFERENCE-RUNTIMES-2026-08-28.md", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/go-inference-runtimes-github-2026-08-28.json", "indexed_revision": "d3f86f897e42e0f5544494d62bfe5d8ead6370fd", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 1, "candidate_count": 1, "filed_issue_count": 1, "completeness_critic": "Shared 23-repository manifest records the pinned runtime tree boundary, implementation/test anchors, activity and license/provenance, fak self-query, candidate dispositions, explicit exclusions, query ledger, and a family-level completeness critique.", "issue_refs": [ "#9846" ] } }, { "repository": "dianlight/gollama.cpp", "url": "https://github.com/dianlight/gollama.cpp", "status": "studied", "priority": 56, "why": "Generated purego llama.cpp binding lineage reference; Yzma is the more material current descendant.", "last_checked": "2026-08-28", "checked_revision": "6376d0d4e0a2d59f6f048a7c1ca121a5c1129604", "stars_at_check": 38, "last_push_at_check": "2026-08-28T12:10:54Z", "study_note": "docs/notes/CONCEPT-STUDY-GO-INFERENCE-RUNTIMES-2026-08-28.md", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/go-inference-runtimes-github-2026-08-28.json", "indexed_revision": "6376d0d4e0a2d59f6f048a7c1ca121a5c1129604", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 1, "candidate_count": 1, "filed_issue_count": 1, "completeness_critic": "Shared 23-repository manifest records the pinned runtime tree boundary, implementation/test anchors, activity and license/provenance, fak self-query, candidate dispositions, explicit exclusions, query ledger, and a family-level completeness critique.", "issue_refs": [ "#9846" ] } }, { "repository": "nikolaydubina/llama2.go", "url": "https://github.com/nikolaydubina/llama2.go", "status": "studied", "priority": 57, "why": "Archived pure-Go Llama-2/llama2.c implementation retained as a compact educational reference.", "last_checked": "2026-08-28", "checked_revision": "836881ddaeb0d96d8c11efe3cd59dbc5a8ea15d2", "stars_at_check": 192, "last_push_at_check": "2024-11-30T03:17:10Z", "study_note": "docs/notes/CONCEPT-STUDY-GO-INFERENCE-RUNTIMES-2026-08-28.md", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/go-inference-runtimes-github-2026-08-28.json", "indexed_revision": "836881ddaeb0d96d8c11efe3cd59dbc5a8ea15d2", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 1, "candidate_count": 1, "filed_issue_count": 1, "completeness_critic": "Shared 23-repository manifest records the pinned runtime tree boundary, implementation/test anchors, activity and license/provenance, fak self-query, candidate dispositions, explicit exclusions, query ledger, and a family-level completeness critique.", "issue_refs": [ "#9846" ] } }, { "repository": "adalkiran/llama-nuts-and-bolts", "url": "https://github.com/adalkiran/llama-nuts-and-bolts", "status": "studied", "priority": 58, "why": "Source-addressable pure-Go Llama 3.1 educational implementation over official weights.", "last_checked": "2026-08-28", "checked_revision": "ccbe25e9fdc0f2ca80c6a5de9d7eb8c82554bf89", "stars_at_check": 316, "last_push_at_check": "2024-08-20T15:10:00Z", "study_note": "docs/notes/CONCEPT-STUDY-GO-INFERENCE-RUNTIMES-2026-08-28.md", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/go-inference-runtimes-github-2026-08-28.json", "indexed_revision": "ccbe25e9fdc0f2ca80c6a5de9d7eb8c82554bf89", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 1, "candidate_count": 1, "filed_issue_count": 1, "completeness_critic": "Shared 23-repository manifest records the pinned runtime tree boundary, implementation/test anchors, activity and license/provenance, fak self-query, candidate dispositions, explicit exclusions, query ledger, and a family-level completeness critique.", "issue_refs": [ "#9846" ] } }, { "repository": "divy-sh/llama3-go", "url": "https://github.com/divy-sh/llama3-go", "status": "studied", "priority": 59, "why": "Unlicensed experimental pure-Go Llama-family implementation; discovery-only and DO-NOT-USE.", "last_checked": "2026-08-28", "checked_revision": "a68a81fe918bd20f43cce845cdcc362b16bce236", "stars_at_check": 2, "last_push_at_check": "2026-07-18T07:54:33Z", "study_note": "docs/notes/CONCEPT-STUDY-GO-INFERENCE-RUNTIMES-2026-08-28.md", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/go-inference-runtimes-github-2026-08-28.json", "indexed_revision": "a68a81fe918bd20f43cce845cdcc362b16bce236", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 1, "candidate_count": 1, "filed_issue_count": 1, "completeness_critic": "Shared 23-repository manifest records the pinned runtime tree boundary, implementation/test anchors, activity and license/provenance, fak self-query, candidate dispositions, explicit exclusions, query ledger, and a family-level completeness critique.", "issue_refs": [ "#9846" ] } }, { "repository": "JetBrains/go-modern-guidelines", "url": "https://github.com/JetBrains/go-modern-guidelines", "status": "studied", "priority": 60, "why": "Version-aware, progressively disclosed modern-Go guidance for coding agents, with semantics-safe adaptation fences.", "last_checked": "2026-08-28", "checked_revision": "40781f167719913666fe2a7dc1c77ea6f256df0a", "stars_at_check": 2558, "last_push_at_check": "2026-08-19T17:39:58Z", "study_note": "docs/notes/CONCEPT-STUDY-GO-MODERN-GUIDELINES-2026-08-28.md" }, { "repository": "TencentCloud/TencentDB-Agent-Memory", "url": "https://github.com/TencentCloud/TencentDB-Agent-Memory", "status": "studied", "priority": 61, "why": "Layered agent memory, replay-safe extraction, governed memory/skill bridges, graph recall, and cross-harness adapters; four FAK-native correctness gaps survived dedupe.", "last_checked": "2026-08-28", "checked_revision": "5299c00aaf65481703c180fd69df066d11254eb7", "stars_at_check": 25025, "last_push_at_check": "2026-08-27T06:49:32Z", "study_note": "docs/notes/CONCEPT-STUDY-TENCENTDB-AGENT-MEMORY-2026-08-28.md", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/tencentcloud-tencentdb-agent-memory.json", "indexed_revision": "5299c00aaf65481703c180fd69df066d11254eb7", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 11, "candidate_count": 11, "filed_issue_count": 4, "completeness_critic": "The local map walked 995 files across 11 subsystems; three isolated study lanes covered runtime, tests, docs, roadmap, releases, forge governance, and provenance. Private advisories/reviews, expired Actions, image SBOMs, external provenance repositories, and a transitive license audit remain unavailable or out of scope; retained candidates were independently reopened before filing.", "issue_refs": [ "#9851", "#9865", "#9866", "#9867", "#9868", "#2349", "#2621", "#2624" ], "source_evidence": [ { "class": "open_closed_issues_prs_discussions", "evidence": [ "gh issue list --repo TencentCloud/TencentDB-Agent-Memory --state all --limit 1000 --json number", "gh pr list --repo TencentCloud/TencentDB-Agent-Memory --state all --limit 1000 --json number", "gh api graphql repository(owner:TencentCloud,name:TencentDB-Agent-Memory) discussions(first:100)", "gh api repos/TencentCloud/TencentDB-Agent-Memory/releases?per_page=100", "docs/notes/CONCEPT-STUDY-TENCENTDB-AGENT-MEMORY-2026-08-28.md#source-and-forge-coverage" ], "note": "Public forge coverage included 320 issues, 858 pull requests, 14 releases, discussions, and targeted comment/review/timeline reads." }, { "class": "fak_selfquery_witness", "evidence": [ "fak capabilities 'layered agent memory persona distillation'", "fak capabilities 'memory asset loadout ACL'", "fak capabilities 'versioned skill asset session pin'", "fak capabilities 'wiki link graph memory recall'", "docs/notes/CONCEPT-STUDY-TENCENTDB-AGENT-MEMORY-2026-08-28.md#fak-on-axis-witness" ], "note": "Capabilities, index surfaces, raw source, and issue searches were ablated into PRESENT/PARTIAL/ABSENT findings." }, { "class": "candidate_matrix", "evidence": [ "docs/notes/CONCEPT-STUDY-TENCENTDB-AGENT-MEMORY-2026-08-28.md#candidate-matrix", "fak study search --limit 20 TencentDB" ], "note": "Eleven candidates have on-axis status, disposition, and evidence; direct reuse and unproven headline numbers are excluded." }, { "class": "issue_tracking", "evidence": [ "https://github.com/anthony-chaudhary/fak/issues/9851", "https://github.com/anthony-chaudhary/fak/issues/9865", "https://github.com/anthony-chaudhary/fak/issues/9866", "https://github.com/anthony-chaudhary/fak/issues/9867", "https://github.com/anthony-chaudhary/fak/issues/9868" ], "note": "The tracker and four bounded implementation issues were independently read back after creation." } ] } }, { "repository": "cactus-compute/needle", "url": "https://github.com/cactus-compute/needle", "status": "studied", "priority": 62, "why": "Tiny-device tool calling, negative-case environment suites, custom quantized artifacts, and model/runtime boundary lessons; one attributed Qwen3.8 corpus gap survived FAK self-query.", "last_checked": "2026-08-28", "checked_revision": "ee221ce7c13579d9809209b979a9b7a50936614c", "stars_at_check": 9575, "last_push_at_check": "2026-08-28T13:50:22Z", "study_note": "docs/notes/CONCEPT-STUDY-NEEDLE-2026-08-28.md", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/cactus-compute-needle.json", "indexed_revision": "ee221ce7c13579d9809209b979a9b7a50936614c", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 10, "candidate_count": 11, "filed_issue_count": 1, "completeness_critic": "All 58 tracked files were mapped across ten load-bearing subsystems; focused runtime tests passed 45 with one native-engine skip; 283 commits, 10 tags, all 43 issues, all 55 PR states/conversations, Releases, roadmap signals, and current-tree license were covered. GitHub Projects V2, external native/model artifacts, separate runtimes, every PR diff/test, Actions logs, PyPI contents, and asset authorship remain unavailable or out of scope and therefore support no runtime/performance/binary-license claims.", "issue_refs": [ "#9852", "#9864" ], "source_evidence": [ { "class": "open_closed_issues_prs_discussions", "evidence": [ "gh api repos/cactus-compute/needle/issues?state=all&per_page=100", "gh api repos/cactus-compute/needle/pulls?state=all&per_page=100", "gh api repos/cactus-compute/needle/releases", "docs/notes/CONCEPT-STUDY-NEEDLE-2026-08-28.md#source-and-history-ledger", "gh api graphql repository discussions cactus-compute/needle" ], "note": "Public forge read-back found 43 issues, 55 pull requests, zero GitHub Releases, disabled Discussions, and sparse review evidence at the observation cutoff." }, { "class": "fak_selfquery_witness", "evidence": [ "fak capabilities confidence calibrated tool call threshold", "fak capabilities frozen environment suite critical negation invalid parallel", "fak capabilities model artifact topology quantization tokenizer kv numerics", "fak capabilities native library checksum signature revision", "docs/notes/CONCEPT-STUDY-NEEDLE-2026-08-28.md#fak-on-axis-witness" ], "note": "Capability/index queries plus raw package search found descriptor, provenance, mixed-precision, KV-budget, and exact-model acceptance mechanisms already present; only the six-class corpus breadth remained partial." }, { "class": "candidate_matrix", "evidence": [ "docs/research/inventory/cactus-compute-needle.json#candidate_matrix", "docs/notes/CONCEPT-STUDY-NEEDLE-2026-08-28.md#candidate-matrix", "study_ca43928a48a872230ea6bdbfd676249cb26d8d22fd095441b9c792919ad8b49c" ], "note": "Eleven candidates have on-axis FAK status, portfolio/license disposition, evidence, and disconfirming boundaries." }, { "class": "issue_tracking", "evidence": [ "https://github.com/anthony-chaudhary/fak/issues/9852", "https://github.com/anthony-chaudhary/fak/issues/9864" ], "note": "Tracker and sole surviving implementation issue were independently read back; the candidate passed the live, dedupe-armed issue contract as dispatchable." } ] } }, { "repository": "diepquynh/ultracode", "url": "https://github.com/diepquynh/ultracode", "status": "candidate", "priority": 63, "why": "Newly discovered named Ultracode plugin with repository inventory, multi-role workflow stages, approval gates, and an explicit quality-versus-token-cost posture directly relevant to the FAK Ultracode effectiveness audit.", "last_checked": "2026-08-29", "checked_revision": "63f0d6e4dbf62258190e938ee4ebf9b93278f119", "stars_at_check": 4, "last_push_at_check": "2026-08-29T19:44:24Z", "study_note": "docs/notes/CONCEPT-STUDY-ULTRACODE-OSS-VARIANTS-2026-08-29.md" }, { "repository": "hesreallyhim/ultracode-workflows", "url": "https://github.com/hesreallyhim/ultracode-workflows", "status": "candidate", "priority": 64, "why": "Named Ultracode workflow library with deterministic scripts, resumable checkpoints, adversarial verification, arithmetic coverage, plan previews, and runtime observation; it was absent from the earlier source ledger and is directly comparable on effectiveness controls.", "last_checked": "2026-08-29", "checked_revision": "9b5404d11b885b28380d3eb17471ef7b17601b5e", "stars_at_check": 2, "last_push_at_check": "2026-08-07T06:48:37Z", "study_note": "docs/notes/CONCEPT-STUDY-ULTRACODE-OSS-VARIANTS-2026-08-29.md" }, { "repository": "Suraj1235/open-dynamic-workflows", "url": "https://github.com/Suraj1235/open-dynamic-workflows", "status": "candidate", "priority": 65, "why": "Open-source script-as-orchestrator engine for Ultracode-style workflows across Codex, OpenCode, MCP clients, and local models, with daemon and embedded execution paths that may inform FAK portability and control-plane boundaries.", "last_checked": "2026-08-29", "checked_revision": "972bb98494ea23f907df88850024bd7022b099d4", "stars_at_check": 33, "last_push_at_check": "2026-07-09T16:35:47Z", "study_note": "docs/notes/CONCEPT-STUDY-ULTRACODE-OSS-VARIANTS-2026-08-29.md" }, { "repository": "Caleb0796/codex-ultracode-kit", "url": "https://github.com/Caleb0796/codex-ultracode-kit", "status": "candidate", "priority": 66, "why": "Small Codex-specific Ultracode kit combining an in-session skill with a deterministic external fan-out harness and skeptic verification, directly overlapping FAK's Codex adapter and effectiveness-envelope questions.", "last_checked": "2026-08-29", "checked_revision": "8176113ed18b0df1ef1a2717e2eb556aaf132018", "stars_at_check": 0, "last_push_at_check": "2026-07-04T05:40:28Z", "study_note": "docs/notes/CONCEPT-STUDY-ULTRACODE-OSS-VARIANTS-2026-08-29.md" }, { "repository": "KangarooKi/Ultracode", "url": "https://github.com/KangarooKi/Ultracode", "status": "candidate", "priority": 67, "why": "Independent local coding CLI named UltraCode with task graphs, permissions, hooks, background commands, MCP, memory, worktrees, and subagents; useful as a product-surface variant but license provenance requires explicit review.", "last_checked": "2026-08-29", "checked_revision": "4a7d1b6eb10e4e020b628b7fbaa21c1837b31edd", "stars_at_check": 5, "last_push_at_check": "2026-06-01T01:19:52Z", "study_note": "docs/notes/CONCEPT-STUDY-ULTRACODE-OSS-VARIANTS-2026-08-29.md" }, { "repository": "toel1234/oh-my-opencode", "url": "https://github.com/toel1234/oh-my-opencode", "status": "candidate", "priority": 68, "why": "Ultrawork lineage with multi-model orchestration, background agents, opt-in team mode, intent routing, progress tooling, and completion enforcement; adjacent rather than name-identical, but relevant to the OSS variant inventory.", "last_checked": "2026-08-29", "checked_revision": "18d7643bc1eb4f3e99e0e80e7b8cca9fe199ff1a", "stars_at_check": 0, "last_push_at_check": "2026-05-10T20:12:47Z", "study_note": "docs/notes/CONCEPT-STUDY-ULTRACODE-OSS-VARIANTS-2026-08-29.md" }, { "repository": "QuintinShaw/pi-dynamic-workflows", "url": "https://github.com/QuintinShaw/pi-dynamic-workflows", "status": "candidate", "priority": 69, "why": "High-signal dynamic-workflow runtime with batch-scoped cancellation, nested caps, deterministic state deltas, prefix resume, and terminal child draining; its directional authoring study lacks a committed matched provider artifact.", "last_checked": "2026-08-29", "checked_revision": "2f85c9eb350319e14a2ff16b98986a30b3c8bb4d", "stars_at_check": 458, "last_push_at_check": "2026-08-29T15:48:04Z", "study_note": "docs/notes/CONCEPT-STUDY-ULTRACODE-OSS-VARIANTS-2026-08-29.md" }, { "repository": "code-yeongyu/oh-my-openagent", "url": "https://github.com/code-yeongyu/oh-my-openagent", "status": "candidate", "priority": 70, "why": "Current Ultrawork lineage with an observable stop goal, atomic todos, cleanup, real-surface QA scenarios, and bounded reviewer resubmission; repository-level license metadata is inconclusive and requires a full provenance pass before reuse.", "last_checked": "2026-08-29", "checked_revision": "b4e0094c04df49360271fd9e449681d0c386eb35", "stars_at_check": 68507, "last_push_at_check": "2026-08-29T16:59:42Z", "study_note": "docs/notes/CONCEPT-STUDY-ULTRACODE-OSS-VARIANTS-2026-08-29.md" }, { "repository": "Yeachan-Heo/oh-my-claudecode", "url": "https://github.com/Yeachan-Heo/oh-my-claudecode", "status": "candidate", "priority": 71, "why": "Historical Ultrawork lineage whose retirement and stop-hook echo-loop reports provide negative evidence for live-goal-derived persistence and immediate cancellation semantics.", "last_checked": "2026-08-29", "checked_revision": "adf4bf3280c8a8d7b932d5c11aef84ba22d6a11d", "stars_at_check": 38876, "last_push_at_check": "2026-08-29T16:39:27Z", "study_note": "docs/notes/CONCEPT-STUDY-ULTRACODE-OSS-VARIANTS-2026-08-29.md" }, { "repository": "mlcommons/inference", "url": "https://github.com/mlcommons/inference", "status": "studied", "priority": 72, "why": "Apache-2.0 LoadGen study for reusable traffic generation, frozen workload and RNG identity, raw result logging, bounded peak search, and independent performance-validity checks. Relevant to FAK's cross-layer sweep evidence contract, not as a universal runner or algorithm. Refresh when the pinned LoadGen peak-search or validity semantics change, #10189 or #10079 lands, or a mapped FAK surface adds a finding class the proposed folds cannot represent.", "last_checked": "2026-08-29", "checked_revision": "cfb0df14b21a3898521891f021a1c6aadec2ab2c", "stars_at_check": 1622, "last_push_at_check": "2026-08-29T21:11:11Z", "study_note": "docs/notes/CONCEPT-CROSS-LAYER-SWEEPS-2026-08-29.md", "refresh_issue": "#10192" }, { "repository": "anthony-chaudhary/metrics-service", "url": "https://github.com/anthony-chaudhary/metrics-service", "status": "studied", "priority": 73, "why": "Private repository studied for aligned typed partial fleet evidence, component readiness with recovery actions, and registry-driven dashboard and alert drift auditing. No project license grant was found at the pinned revision, so follow-ons are clean-room adaptations only; refresh if licensing, release provenance, or the pinned mechanisms materially change.", "last_checked": "2026-08-29", "checked_revision": "df58aad21de603992fc7b4900ac814610092c6c3", "stars_at_check": 0, "last_push_at_check": "2026-05-12T01:20:46Z", "study_note": "docs/research/metrics-service-study-2026-08-29.md", "refresh_issue": "#10287", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/anthony-chaudhary-metrics-service.json", "indexed_revision": "df58aad21de603992fc7b4900ac814610092c6c3", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 6, "candidate_count": 6, "filed_issue_count": 3, "source_evidence": [ { "class": "open_closed_issues_prs_discussions", "evidence": [ "gh issue list --state all --repo anthony-chaudhary/metrics-service", "gh pr list --state all --repo anthony-chaudhary/metrics-service", "gh release list --repo anthony-chaudhary/metrics-service", "gh api graphql repository discussions anthony-chaudhary/metrics-service" ], "note": "The pinned private repository had three historical pull requests and no issues; tags continued beyond the latest GitHub Release, and tags were unsigned." }, { "class": "fak_selfquery_witness", "evidence": [ "internal/metrics/device_spine_scrape.go:122-175", "internal/metrics/device_spine_federate.go:55-72", "internal/serverproduct/contract.go:97-235", "internal/serverproduct/encoding.go:16-92", "internal/metrics/openmetrics.go:13-344", "internal/metrics/parity_dashboard.go:198-370", "fak capabilities anthony-chaudhary/metrics-service" ], "note": "Exact FAK source anchors decided PRESENT/PARTIAL/ABSENT dispositions and bounded the clean-room follow-ons." }, { "class": "candidate_matrix", "evidence": [ "docs/research/metrics-service-study-2026-08-29.md#candidate-dispositions", "docs/notes/candidate-matrix.md" ] }, { "class": "issue_tracking", "evidence": [ "https://github.com/anthony-chaudhary/fak/issues/10287", "https://github.com/anthony-chaudhary/fak/issues/10312", "https://github.com/anthony-chaudhary/fak/issues/10313", "https://github.com/anthony-chaudhary/fak/issues/10314" ], "note": "The tracker and three independently shippable follow-ons carry the immutable study receipt and pinned clean-room provenance constraint." } ], "issue_refs": [ "#10287", "#10312", "#10313", "#10314" ], "completeness_critic": "Covered the committed tree, runtime/config/collector/storage/API lifecycle, operator UI and Grafana surfaces, tracked tests and fixtures, complete local history and tags, releases, pull requests, issues and discussions, provenance and license searches, current FAK overlap, candidate dispositions, deduplication, and delegated reads independently rechecked by the coordinator. Untracked target files, dependency-by-dependency license audit, external production deployments, and private runtime logs are outside the decision denominator." } }, { "repository": "anthony-chaudhary/Benchmark", "url": "https://github.com/anthony-chaudhary/Benchmark", "status": "studied", "priority": 74, "why": "Pinned clean-room study of benchmark orchestration and evidence mechanisms. FAK already covers typed metrics, sinks, raw integrity, preflight, publication admission, and execution/export separation; watcher-safe snapshots, correlation receipts, and parser-completeness receipts survive as bounded follow-ons. Refresh if licensing, methodology, or the pinned mechanisms materially change.", "last_checked": "2026-08-29", "checked_revision": "b60f2f05e23ec38ae4f50ca19717c2f615ed419a", "stars_at_check": 0, "last_push_at_check": "2026-08-19T00:33:36Z", "study_note": "docs/notes/CONCEPT-STUDY-BENCH-2026-08-29.md", "refresh_issue": "#10263", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/anthony-chaudhary-benchmark.json", "indexed_revision": "b60f2f05e23ec38ae4f50ca19717c2f615ed419a", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 44, "candidate_count": 9, "filed_issue_count": 3, "source_evidence": [ { "class": "open_closed_issues_prs_discussions", "evidence": [ "gh issue list --state all --repo anthony-chaudhary/Benchmark", "gh pr list --state all --repo anthony-chaudhary/Benchmark", "gh release list --repo anthony-chaudhary/Benchmark", "gh api graphql repository discussions anthony-chaudhary/Benchmark" ], "note": "Forge census was captured separately from the mechanical local-tree inventory." }, { "class": "fak_selfquery_witness", "evidence": [ "internal/cacheobs/metricspec.go:77-169", "internal/findingsink/findingsink.go:137-168", "internal/auditreceipt/auditreceipt.go:59-243", "internal/benchmarkartifact", "internal/armbench/manifest.go", "internal/benchcatalog/ingest.go:168-314", "internal/benchauthority/validate.go:23-178", "internal/bench/ctxadmission.go:341-490", "internal/bench/setupparity.go:510-517", "fak capabilities anthony-chaudhary/Benchmark" ], "note": "Exact FAK seams decided PRESENT/PARTIAL dispositions and prevented duplicate abstractions." }, { "class": "candidate_matrix", "evidence": [ "docs/notes/CONCEPT-STUDY-BENCH-2026-08-29.md#candidate-dispositions", "study_36e9161928991db39b7aebd463809d8ca37e9ac9d66ba074888c4bd45edd6882", "docs/notes/candidate-matrix.md" ] }, { "class": "issue_tracking", "evidence": [ "https://github.com/anthony-chaudhary/fak/issues/10263", "https://github.com/anthony-chaudhary/fak/issues/10302", "https://github.com/anthony-chaudhary/fak/issues/10303", "https://github.com/anthony-chaudhary/fak/issues/10304" ], "note": "The parent and three bounded follow-ons carry pinned clean-room provenance and the durable receipt." } ], "issue_refs": [ "#10263", "#10302", "#10303", "#10304" ], "completeness_critic": "Covered the committed tree, architecture and runtime flow, tests and fixtures, local history and tags, forge issues/PRs/discussions metadata, provenance and license evidence, current FAK overlap, candidate dispositions, deduplication, issue filing, and durable receipt read-back. Inventory totals are a mechanical local-tree denominator that includes broad environment material, not project-owned source totals. Private deployments, external runtime logs, dependency-by-dependency license review, producer authentication, and comparative effectiveness remain outside the denominator." } }, { "repository": "ActaClad/plumbline", "url": "https://github.com/ActaClad/plumbline", "status": "studied", "priority": 76, "why": "Apache-2.0 Python reliability and architecture analyzer for LLM/agent systems. Pinned study retained four bounded adaptations\u2014SARIF projection, labeled precision/recall corpora, stable finding baseline deltas, and one bounded cross-function value-flow rule\u2014while rejecting a duplicate Python analyzer framework. Refresh after a post-v0.0.2 release, a material engine/reporter change, or a witnessed determinism fix.", "last_checked": "2026-08-31", "checked_revision": "cae5eeaa244830ef88bd47e18d39b7b7761df41e", "stars_at_check": 26, "last_push_at_check": "2026-08-27T07:03:08Z", "study_note": "docs/research/plumbline-study-2026-08-31.md", "refresh_issue": "#10452", "inventory": { "mode": "exhaustive", "map_path": "docs/research/plumbline-study-2026-08-31.md", "indexed_revision": "cae5eeaa244830ef88bd47e18d39b7b7761df41e", "source_classes": [ "runtime_source", "tests_fixtures", "readme_docs", "history_changelog_releases", "roadmap_todos", "open_closed_issues_prs_discussions", "license_provenance", "fak_selfquery_witness" ], "subsystem_count": 8, "candidate_count": 7, "filed_issue_count": 4, "source_evidence": [ { "class": "runtime_source", "evidence": [ "https://github.com/ActaClad/plumbline/tree/cae5eeaa244830ef88bd47e18d39b7b7761df41e/src/plumbline" ], "note": "AST, value derivation, taint/evidence, rules, scoring, reporters, adapters, and skill export inspected." }, { "class": "tests_fixtures", "evidence": [ "python -m pytest -q (with PYTHONPATH=src): 481 passed, 2 failed" ], "note": "Failures recorded in the study note; current determinism claim is not treated as fully witnessed." }, { "class": "readme_docs", "evidence": [ "docs/research/plumbline-study-2026-08-31.md" ], "note": "README, ADRs/specs, backlog, changelog, CI and integration docs inspected." }, { "class": "history_changelog_releases", "evidence": [ "git -C _scratch/study-plumbline/repo log --reverse" ], "note": "125-commit implementation sequence inspected." }, { "class": "roadmap_todos", "evidence": [ "https://github.com/ActaClad/plumbline/releases" ], "note": "v0.0.1 and v0.0.2 inspected." }, { "class": "open_closed_issues_prs_discussions", "evidence": [ "https://github.com/ActaClad/plumbline/issues?q=is%3Aissue", "https://github.com/ActaClad/plumbline/pulls?q=is%3Apr", "gh api graphql repository discussions ActaClad/plumbline" ], "note": "No non-PR issues at check time; PR history is Dependabot-only." }, { "class": "license_provenance", "evidence": [ "https://github.com/ActaClad/plumbline/blob/cae5eeaa244830ef88bd47e18d39b7b7761df41e/LICENSE", "https://github.com/ActaClad/plumbline/blob/cae5eeaa244830ef88bd47e18d39b7b7761df41e/pyproject.toml" ], "note": "Apache-2.0 and Python packaging inspected." }, { "class": "fak_selfquery_witness", "evidence": [ "docs/research/plumbline-study-2026-08-31.md", "#10460", "#10461", "#10462", "#10464", "study_b61c432e9a8f2ad5574036e224b8fd957aa482a9c588e0fe295396ed532edad6", "fak capabilities ActaClad/plumbline" ], "note": "Every surviving candidate has a status, disposition, issue, and durable receipt." } ], "issue_refs": [ "#10460", "#10461", "#10462", "#10464" ], "completeness_critic": "One bounded follow-up remains intentionally deferred: no individual Python-specific rule was ported because no FAK corpus currently proves a miss; the source-analysis issue requires selecting and witnessing one concrete Go rule before implementation." } }, { "repository": "warpfront/hipfire", "url": "https://github.com/warpfront/hipfire", "status": "studied", "priority": 88, "why": "RDNA-native Rust inference engine with strict physical admission, session-safe prefix reuse, transactional adaptive KV, fail-closed replay, architecture-isolated kernels, and unusually strong route/hash-bound evidence and negative-result retention. Reuse existing #9587 and #8395 seams; watch emerging device rollback and mesh work rather than crediting it as shipped.", "last_checked": "2026-09-01", "checked_revision": "0f1628241f751d9ee44922b5a1ca40aa756a82e0", "stars_at_check": 578, "last_push_at_check": "2026-09-01T08:23:08Z", "study_note": "docs/notes/CONCEPT-STUDY-HIPFIRE-2026-09-01.md", "refresh_issue": "#10591", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/warpfront-hipfire.json", "indexed_revision": "0f1628241f751d9ee44922b5a1ca40aa756a82e0", "source_classes": [ "runtime_source", "architecture_design", "tests_fixtures", "roadmap_todos", "readme_docs", "open_closed_issues_prs_discussions", "license_provenance", "fak_selfquery_witness" ], "subsystem_count": 12, "candidate_count": 7, "filed_issue_count": 0, "source_evidence": [ { "class": "runtime_source", "evidence": [ "crates/hipfire-runtime/src/admission.rs:86@0f1628241f751d9ee44922b5a1ca40aa756a82e0", "crates/hipfire-runtime/src/session_table.rs:135@0f1628241f751d9ee44922b5a1ca40aa756a82e0", "crates/rdna-compute/src/replay.rs:68@0f1628241f751d9ee44922b5a1ca40aa756a82e0" ], "note": "1,200 tracked files inventoried; runtime admission, cache/session ownership, replay, graph, slot, scratch, dispatch and tests deep-read." }, { "class": "architecture_design", "evidence": [ "kernels/src/attention_flash_q8_0_reduce_gated_mq_rotate.gfx1100.hip:11@0f1628241f751d9ee44922b5a1ca40aa756a82e0", "kernels/src/attention_dflash_wmma_m32_kstg_FAILED.hip:6@0f1628241f751d9ee44922b5a1ca40aa756a82e0" ], "note": "1,034 tracked files inventoried; architecture isolation and retained failed-kernel evidence sampled deeply." }, { "class": "tests_fixtures", "evidence": [ "docs/BENCHMARKS.md:25@0f1628241f751d9ee44922b5a1ca40aa756a82e0", "docs/GOLDEN-REDLINE.md:24@0f1628241f751d9ee44922b5a1ca40aa756a82e0" ], "note": "311 benchmark files, 28 tests, and five Redline corpus files inventoried; measurement/admission and identity contracts deep-read." }, { "class": "roadmap_todos", "evidence": [ "autoresearch/ar/driver.py:119@0f1628241f751d9ee44922b5a1ca40aa756a82e0", "tools/change_gate/selector.py:115@0f1628241f751d9ee44922b5a1ca40aa756a82e0" ], "note": "Autoresearch, agent rules, CI, reaping and change-gate surfaces inventoried." }, { "class": "open_closed_issues_prs_discussions", "evidence": [ "https://github.com/warpfront/hipfire/commit/0f1628241f751d9ee44922b5a1ca40aa756a82e0", "https://github.com/warpfront/hipfire/releases/tag/v0.2.1", "https://github.com/warpfront/hipfire/pull/671", "gh api graphql repository discussions warpfront/hipfire", "gh issue list --state all --repo warpfront/hipfire" ], "note": "History, 21 releases, issue/PR states, and selected architecture discussions inspected; post-pin PRs are labeled emerging." }, { "class": "license_provenance", "evidence": [ "LICENSE:1@0f1628241f751d9ee44922b5a1ca40aa756a82e0", "NOTICE:1@0f1628241f751d9ee44922b5a1ca40aa756a82e0", "PRIOR-ART.md:1@0f1628241f751d9ee44922b5a1ca40aa756a82e0", ".gitmodules:1@0f1628241f751d9ee44922b5a1ca40aa756a82e0" ], "note": "MIT/Apache bundle and attribution machinery inspected; pinned submodule license remains unverified." }, { "class": "fak_selfquery_witness", "evidence": [ "docs/notes/CONCEPT-STUDY-HIPFIRE-2026-09-01.md", "#9587", "#8395", "#10591", "fak capabilities warpfront/hipfire" ], "note": "Every candidate has status, disposition, on-axis FAK witness, and explicit issue outcome." } ], "issue_refs": [ "#9587", "#8395", "#10591" ], "completeness_critic": "Major tracked source classes were exhaustively inventoried, but the 1,086-document investigation corpus was clustered rather than read line-by-line; older benchmark identity and statistical rigor are uneven; selected active PRs lacked formal review; and the pinned submodule license was not locally verified." } }, { "repository": "varjoranta/turboquant-vllm", "url": "https://github.com/varjoranta/turboquant-vllm", "status": "studied", "priority": 25, "why": "TurboQuant+ KV cache and 3-bit weight compression plugin for vLLM. Source of sparse top-k MoE dequant into shared scratch pool (8.4x decode speedup at bs=1 on Qwen3-30B-A3B) and served-benchmark validation of shape-gain norm correction + asymmetric K4/V3 bits.", "last_checked": "2026-09-02", "checked_revision": "c8a7e0a73b2b9bb93dc66c9380dceab985a0fbc5", "stars_at_check": 80, "last_push_at_check": "2026-07-14T06:43:07Z", "study_note": "docs/notes/CONCEPT-STUDY-TURBOQUANT-VLLM-2026-09-02.md", "refresh_issue": "#10709" }, { "repository": "portkey-ai/gateway", "url": "https://github.com/portkey-ai/gateway", "status": "studied", "priority": 75, "why": "Multi-provider AI gateway with unified OpenAI-compatible API, 1600+ models across 45+ providers, declarative provider normalization, routing strategies (fallback/loadbalance/conditional), guardrails plugin architecture, multi-backend caching, OTLP observability, and MCP gateway. Direct seam overlap with fak's routing, policy, caching, and observability surfaces.", "last_checked": "2026-09-02", "checked_revision": "669825cbe89ee51569918b8f78a9db486fd69dd4", "stars_at_check": 0, "last_push_at_check": "2026-09-02T00:00:00Z", "study_note": "docs/notes/CONCEPT-STUDY-PORTKEY-AI-GATEWAY-2026-09-02.md", "inventory": { "mode": "exhaustive", "map_path": "docs/notes/CONCEPT-STUDY-PORTKEY-AI-GATEWAY-2026-09-02.md", "indexed_revision": "669825cbe89ee51569918b8f78a9db486fd69dd4", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 8, "candidate_count": 14, "filed_issue_count": 9, "source_evidence": [ { "class": "runtime_source", "evidence": [ "https://github.com/portkey-ai/gateway/tree/669825cbe89ee51569918b8f78a9db486fd69dd4" ], "note": "Shallow clone at pinned revision; 766 tracked files across 8 load-bearing subsystems deep-read via parallel agents" }, { "class": "architecture_design", "evidence": [ "docs/notes/CONCEPT-STUDY-PORTKEY-AI-GATEWAY-2026-09-02.md#candidate-matrix" ], "note": "Core routing engine, provider abstraction, caching, guardrails, observability, MCP gateway all mapped to source" }, { "class": "license_provenance", "evidence": [ "https://github.com/portkey-ai/gateway/blob/669825cbe89ee51569918b8f78a9db486fd69dd4/LICENSE" ], "note": "MIT license compatible with Apache-2.0; direct port/adapt permitted with attribution" }, { "class": "fak_selfquery_witness", "evidence": [ "fak capabilities \"multi-provider routing\"", "fak capabilities \"conditional routing\"", "fak capabilities \"guardrails plugin architecture\"", "fak capabilities \"cache backend\"", "fak capabilities \"OTLP spans\"", "fak capabilities \"provider normalization\"" ], "note": "Capability/index queries plus raw source search decided PRESENT/PARTIAL/ABSENT dispositions" }, { "class": "candidate_matrix", "evidence": [ "docs/notes/CONCEPT-STUDY-PORTKEY-AI-GATEWAY-2026-09-02.md#candidate-matrix" ], "note": "14 candidates with on-axis status, disposition, and evidence" }, { "class": "completeness_critic", "evidence": [ "docs/notes/CONCEPT-STUDY-PORTKEY-AI-GATEWAY-2026-09-02.md#completeness-critic" ], "note": "All 8 load-bearing subsystems deeply read; tests, cookbooks, deployments, PR history justified as out of kernel-scope" }, { "class": "issue_tracking", "evidence": [ "https://github.com/anthony-chaudhary/fak/issues/10692", "https://github.com/anthony-chaudhary/fak/issues/10693", "https://github.com/anthony-chaudhary/fak/issues/10694", "https://github.com/anthony-chaudhary/fak/issues/10695", "https://github.com/anthony-chaudhary/fak/issues/10696", "https://github.com/anthony-chaudhary/fak/issues/10697", "https://github.com/anthony-chaudhary/fak/issues/10698", "https://github.com/anthony-chaudhary/fak/issues/10699", "https://github.com/anthony-chaudhary/fak/issues/10700" ], "note": "9 issues filed and verified independently" } ], "issue_refs": [ "#10692", "#10693", "#10694", "#10695", "#10696", "#10697", "#10698", "#10699", "#10700" ], "completeness_critic": "All 8 load-bearing subsystems (core routing, provider abstraction, caching, guardrails, observability, MCP, transformation, streaming) deeply read via fan-out agents. Tests, cookbooks, deployment configs, and git history justified as non-kernel-relevant. No material kernel-relevant subsystem unopened." } }, { "repository": "eunomia-bpf/agentsight", "url": "https://github.com/eunomia-bpf/agentsight", "status": "studied", "priority": 77, "why": "eBPF-based boundary tracing for AI agents: SSL/TLS uprobes for plaintext LLM traffic, process/file/network tracing via tracepoints, multi-protocol HTTP/SSE/WebSocket parsing, session-centric live top, agent-native session parsers, materialized view with row sinks. Direct overlap with fak's gateway, observability, and session surfaces.", "last_checked": "2026-09-02", "checked_revision": "934f441eff8ca210807333633f47b2efcb8cd020", "stars_at_check": 0, "last_push_at_check": "2026-09-02T00:00:00Z", "study_note": "docs/notes/CONCEPT-STUDY-agentsight-2026-09-02.md", "inventory": { "mode": "exhaustive", "map_path": "docs/research/inventory/eunomia-bpf-agentsight.json", "indexed_revision": "934f441eff8ca210807333633f47b2efcb8cd020", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 14, "candidate_count": 14, "filed_issue_count": 4, "source_evidence": [ { "class": "runtime_source", "evidence": [ "bpf/sslsniff.bpf.c:1-590@934f441", "bpf/process.bpf.c:1-327@934f441", "ext/analysis/src/analyzers/http_parser.rs:1-1184@934f441", "ext/analysis/src/analyzers/sse_processor.rs:1-1052@934f441", "ext/analysis/src/view/mod.rs:1-1038@934f441", "ext/session/src/parser.rs:1-5273@934f441" ], "note": "Shallow clone at pinned revision; 14 load-bearing subsystems deep-read via parallel agents" }, { "class": "architecture_design", "evidence": [ "docs/design/materialized-view-architecture.md@934f441", "docs/design/session-centric-top.md@934f441", "docs/design/paper.txt:1-300@934f441" ], "note": "Boundary tracing rationale, session-centric top, academic paper all mapped to source" }, { "class": "license_provenance", "evidence": [ "https://github.com/eunomia-bpf/agentsight/blob/934f441eff8ca210807333633f47b2efcb8cd020/LICENSE" ], "note": "MIT license compatible with Apache-2.0; direct port/adapt permitted with attribution" }, { "class": "fak_selfquery_witness", "evidence": [ "fak capabilities \"SSL/TLS interception\"", "fak capabilities \"materialized view\"", "fak capabilities \"session discovery\"", "fak capabilities \"token usage normalization\"", "fak capabilities \"process tree\"", "fak capabilities \"OTel GenAI export\"", "fak capabilities \"bounded retention\"" ], "note": "Capability/index queries plus raw source search decided PRESENT/PARTIAL/ABSENT/DIVERGENT dispositions" }, { "class": "candidate_matrix", "evidence": [ "docs/notes/CONCEPT-STUDY-agentsight-2026-09-02.md#candidate-borrows", "study_8e02e83cb64d8494cf8fe3d320cf13be15d6d1b9f9a296f4f52eaba07f124f0d", "docs/notes/candidate-matrix.md" ], "note": "14 candidates with on-axis status, disposition, evidence, and durable study receipt" }, { "class": "completeness_critic", "evidence": [ "docs/notes/CONCEPT-STUDY-agentsight-2026-09-02.md#what-was-read-fan-out-coverage" ], "note": "All 14 load-bearing subsystems deeply read; no material kernel-relevant subsystem unopened" }, { "class": "issue_tracking", "evidence": [ "https://github.com/anthony-chaudhary/fak/issues/10736", "https://github.com/anthony-chaudhary/fak/issues/10737", "https://github.com/anthony-chaudhary/fak/issues/10738", "https://github.com/anthony-chaudhary/fak/issues/10739", "https://github.com/anthony-chaudhary/fak/issues/10740" ], "note": "Parent study tracker and 4 independently dispatchable child issues filed and verified" } ], "issue_refs": [ "#10736", "#10737", "#10738", "#10739", "#10740" ], "completeness_critic": "All 14 load-bearing subsystems (eBPF SSL uprobes, process tracepoints, HTTP/2+SSE+WS parsers, token normalization, materialized view, session parsing, process correlation, container resolution, OTel export, bounded retention, session-centric top, paper framing) deeply read via fan-out agents. Tests, CI, docker, frontend, experimental browsertrace justified as non-kernel-relevant. No material kernel-relevant subsystem unopened." } }, { "repository": "ovg-project/kvcached", "url": "https://github.com/ovg-project/kvcached", "status": "studied", "priority": 15, "why": "Virtualized elastic KV cache bringing CUDA VMM (cuMemAddressReserve/cuMemMap) to LLM serving on shared GPUs. Deeply studied for virtual-physical capacity decoupling, page-aligned eviction, zero-copy prefix sharing, page-granular CPU offloading, and compound-page layout negative knowledge.", "last_checked": "2026-09-02", "checked_revision": "60cad949389af6bbf1d65c4eddf325113df5a9eb", "stars_at_check": 1281, "last_push_at_check": "2026-08-23T05:24:49Z", "study_note": "docs/notes/CONCEPT-STUDY-KVCACHED-2026-09-02.md", "refresh_issue": "#10720", "inventory": { "mode": "exhaustive", "map_path": "docs/notes/CONCEPT-STUDY-KVCACHED-2026-09-02.md", "indexed_revision": "60cad949389af6bbf1d65c4eddf325113df5a9eb", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 6, "candidate_count": 8, "filed_issue_count": 4, "completeness_critic": "Exhaustive exploration across kvcached/ Python core, csrc/ CUDA/C++ VMM driver layer, engine_integration/ runtime patches, controller/ router and sleep manager, examples, benchmarks, tests, and docs; 4 independent issues filed (#10720, #10721, #10722, #10723) under epic #2236; durable receipt study_7f18852854b61725455e62ba7e29fbbc289a2e1775ca418021c0cfcfee7703ff recorded.", "issue_refs": [ "#10720", "#10721", "#10722", "#10723" ] } }, { "repository": "julianmb/q38rocm", "url": "https://github.com/julianmb/q38rocm", "status": "studied", "priority": 1, "why": "Dedicated single-model deployment package for Qwen 3.8 27B ROCmFP4 on AMD Strix Halo (Ryzen AI Max+ 395). Hardware-aligned block quantization (block 32), MTP speculative decoding, asymmetric TurboQuant KV cache, prompt-cache coexistence, and tuning/benchmark tooling.", "last_checked": "2026-09-02", "checked_revision": "6c142530031c923ece1adf4d8c9c824b7369fef8", "stars_at_check": 185, "last_push_at_check": "2026-09-02T18:30:00Z", "study_note": "docs/notes/CONCEPT-STUDY-q38rocm-2026-09-02.md", "refresh_issue": "#10730", "inventory": { "mode": "exhaustive", "map_path": "docs/notes/CONCEPT-STUDY-q38rocm-2026-09-02.md", "indexed_revision": "6c142530031c923ece1adf4d8c9c824b7369fef8", "source_classes": [ "readme_docs", "architecture_design", "runtime_source", "tests_fixtures", "history_changelog_releases", "open_closed_issues_prs_discussions", "roadmap_todos", "license_provenance", "fak_selfquery_witness", "candidate_matrix", "completeness_critic", "issue_tracking" ], "subsystem_count": 8, "candidate_count": 12, "filed_issue_count": 6, "completeness_critic": "Exhaustive exploration across quantization pipeline, MTP speculative decoding, KV cache and prompt caching, benchmarks and quality eval, NPU hybrid pipeline, engine build and provenance, server launcher, Docker integration, and upstream tracking; 6 independent issues filed (#10730, #10731, #10732, #10733, #10734, #10735) under epics #6221, #2236, #10193; durable receipt study_6777fd46608e1a1293a3f111c4a4f17a9aab28c16431e689f0e011cffc965cb6 recorded.", "issue_refs": [ "#10730", "#10731", "#10732", "#10733", "#10734", "#10735" ] } }, { "repository": "aic0d3r/qwen38-strix-halo-harness", "url": "https://github.com/aic0d3r/qwen38-strix-halo-harness", "status": "studied", "priority": 1, "why": "Local coding-agent harness for AMD Strix Halo APUs: runs Qwen3.8-27B (256k ctx) + Flash-Next 125B on 70-80W, bundling 6 pi extensions (compaction, repomap, branch-summary, triage, commit, /tune), Ling-3.0-tiny aux model on loopback :8090, deep-context ubatch throttling, and playtest-graded game verification benchmark.", "last_checked": "2026-09-05", "checked_revision": "e34ecf3f4b08c4f499be6eda86618f2d0518adfb", "stars_at_check": 2, "last_push_at_check": "2026-09-05T00:00:00Z", "study_note": "docs/notes/CONCEPT-STUDY-STRIX-HALO-ECOSYSTEM-2026-09-05.md", "inventory": { "mode": "exhaustive", "indexed_revision": "e34ecf3f4b08c4f499be6eda86618f2d0518adfb", "subsystem_count": 5, "candidate_count": 5, "filed_issue_count": 0, "issue_refs": [] } }, { "repository": "Hal0ai/hal0", "url": "https://github.com/Hal0ai/hal0", "status": "studied", "priority": 1, "why": "24/7 self-hosted home/office AI appliance platform for unified-memory APUs. Manages container slots via systemd Quadlets, FastFlowLM NPU trio coresidence (chat, STT, embeddings on AMDXDNA), GpuArbiter exclusive LLM vs ComfyUI time-multiplexing, hybrid cgroup memory max-pooling, and 2-tier administrative MCP approval queue.", "last_checked": "2026-09-05", "checked_revision": "108b366d545e7c3a8c13bf2dd2cb4fc74c59458f", "stars_at_check": 69, "last_push_at_check": "2026-09-05T00:00:00Z", "study_note": "docs/notes/CONCEPT-STUDY-STRIX-HALO-ECOSYSTEM-2026-09-05.md", "inventory": { "mode": "exhaustive", "indexed_revision": "108b366d545e7c3a8c13bf2dd2cb4fc74c59458f", "subsystem_count": 5, "candidate_count": 5, "filed_issue_count": 1, "issue_refs": [ "#11748" ] } }, { "repository": "Nathanw1014/strix-halo-llamacpp", "url": "https://github.com/Nathanw1014/strix-halo-llamacpp", "status": "studied", "priority": 1, "why": "Performance-tuned Vulkan/HIP llama.cpp fork for AMD Strix Halo (gfx1151): dequant-once quantized Flash Attention (+47% prefill), f16 KV contiguization eliminating the 16-channel LPDDR5X stride penalty (2.69x speedup to 190 t/s at 64k), bytes-aware command buffer submission gating solving >137k watchdog crashes, and dense Wave32 retiling.", "last_checked": "2026-09-05", "checked_revision": "45bec945dd4c7944fba45ebd02e6b713b342d0ba", "stars_at_check": 162, "last_push_at_check": "2026-09-04T00:00:00Z", "study_note": "docs/notes/CONCEPT-STUDY-STRIX-HALO-ECOSYSTEM-2026-09-05.md", "inventory": { "mode": "exhaustive", "indexed_revision": "45bec945dd4c7944fba45ebd02e6b713b342d0ba", "subsystem_count": 4, "candidate_count": 4, "filed_issue_count": 1, "issue_refs": [ "#11746" ] } }, { "repository": "julianmb/halofpx", "url": "https://github.com/julianmb/halofpx", "status": "studied", "priority": 1, "why": "Turnkey FastAPI model serving daemon and CLI for AMD Strix Halo: auto-detects gfx1151/gfx1201, defaults to Mesa RADV Wave64 cooperative matrices (KHR_coopmat) over ROCm, fixes MTP speculative decoding prompt-cache coexistence via empty-spec decoupling patch, and audits Linux TTM/GTT kernel parameters (strix_doctor.py).", "last_checked": "2026-09-05", "checked_revision": "8790f91dedc1b6eecab5c361d410b312da726342", "stars_at_check": 52, "last_push_at_check": "2026-09-05T00:00:00Z", "study_note": "docs/notes/CONCEPT-STUDY-STRIX-HALO-ECOSYSTEM-2026-09-05.md", "inventory": { "mode": "exhaustive", "indexed_revision": "8790f91dedc1b6eecab5c361d410b312da726342", "subsystem_count": 4, "candidate_count": 3, "filed_issue_count": 0, "issue_refs": [] } }, { "repository": "otheru-ai/ember", "url": "https://github.com/otheru-ai/ember", "status": "studied", "priority": 1, "why": "High-performance C99/C++ inference server for DeepSeek-V4-Flash on AMD Strix Halo (gfx1151): stable C backend ABI, cross-session asynchronous GPU target verification + NPU/CPU DSpark drafting (1.48x throughput), full-prompt cut-point prefix caching (194 ms warm TTFT), and discriminated union DSML EBNF grammar compilation via xgrammar.", "last_checked": "2026-09-05", "checked_revision": "c8a55a851ad245b82b20a624b9a56c243005c30a", "stars_at_check": 21, "last_push_at_check": "2026-09-05T00:00:00Z", "study_note": "docs/notes/CONCEPT-STUDY-STRIX-HALO-ECOSYSTEM-2026-09-05.md", "inventory": { "mode": "exhaustive", "indexed_revision": "c8a55a851ad245b82b20a624b9a56c243005c30a", "subsystem_count": 5, "candidate_count": 4, "filed_issue_count": 1, "issue_refs": [ "#11747" ] } }, { "repository": "kyuz0/amd-strix-halo-toolboxes", "url": "https://github.com/kyuz0/amd-strix-halo-toolboxes", "status": "studied", "priority": 1, "why": "Canonical community container toolbox for AMD Strix Halo: pre-built OCI containers for Vulkan RADV and ROCm 10.0 on gfx1151, llama.cpp host buffer fixes (#25992), gguf-vram-estimator.py, and gpu-workload-watch autonomous systemd power and 100% fan duty governor.", "last_checked": "2026-09-05", "checked_revision": "687d9b6c63093ca7bc925dcfc60bf10e5a55449c", "stars_at_check": 1913, "last_push_at_check": "2026-09-05T00:00:00Z", "study_note": "docs/notes/CONCEPT-STUDY-STRIX-HALO-ECOSYSTEM-2026-09-05.md", "inventory": { "mode": "exhaustive", "indexed_revision": "687d9b6c63093ca7bc925dcfc60bf10e5a55449c", "subsystem_count": 4, "candidate_count": 3, "filed_issue_count": 1, "issue_refs": [ "#11749" ] } } ] }