# Changelog

## Score

- CAI 36.6 → 35.5 (-1.1)
- Rubric changed (rubric-2026.11.2 → rubric-2026.11.4) — scores are not directly comparable.

## Lenses

- Code Health 51.8 → 48.5 (-3.3)
- Architecture 93.9 → 94.3 (+0.4)
- Maturity 74.6 → 77.4 (+2.8)
- Readiness 50.3 → 50.3 (0.0)
- Security 37.4 → 37.4 (0.0)
- Accessibility 21.4 → 19.4 (-2.0)
- Performance 100.0 → 100.0 (0.0)

## Resolved (145)

- Duplicated block (10 lines × 2) (archive/kt-sft/ktransformers/ktransformers_ext/triton/fp8gemm.py)
- Duplicated block (10 lines × 2) (archive/kt-sft/ktransformers/server/args.py)
- Duplicated block (10 lines × 2) (archive/kt-sft/ktransformers/server/config/config.py)
- Duplicated block (11 lines × 2) (archive/kt-sft/ktransformers/server/balance_serve/inference/model_runner.py)
- Duplicated block (11 lines × 2) (archive/kt-sft/ktransformers/util/utils.py)
- Duplicated block (12 lines × 2) (archive/csrc/ktransformers_ext/bench/bench_moe_amx.py)
- Duplicated block (12 lines × 2) (archive/kt-sft/ktransformers/server/backend/interfaces/transformers.py)
- Duplicated block (12 lines × 2) (archive/kt-sft/ktransformers/util/utils.py)
- Duplicated block (12 lines × 4) (archive/csrc/custom_marlin/utils/format24.py)
- Duplicated block (12 lines × 4) (archive/csrc/custom_marlin/utils/quant_utils.py)
- Duplicated block (13 lines × 2) (archive/kt-sft/ktransformers/server/balance_serve/inference/forward_batch.py)
- Duplicated block (13 lines × 4) (archive/csrc/custom_marlin/utils/format24.py)
- Duplicated block (14 lines × 2) (archive/kt-sft/ktransformers/operators/flashinfer_wrapper.py)
- Duplicated block (14 lines × 4) (archive/csrc/custom_marlin/utils/format24.py)
- Duplicated block (148 lines × 2) (archive/kt-sft/ktransformers/server/api/openai/endpoints/chat.py)
- Duplicated block (14–15 lines × 3) (kt-kernel/python/sft/backend.py)
- Duplicated block (15 lines × 2) (archive/kt-sft/ktransformers/operators/flashinfer_batch_prefill_wrapper.py)
- Duplicated block (15 lines × 2) (archive/kt-sft/ktransformers/util/custom_gguf.py)
- Duplicated block (15 lines × 2) (kt-kernel/python/sft/wrapper.py)
- Duplicated block (17 lines × 2) (archive/kt-sft/ktransformers/server/config/config.py)
- …and 125 more

## New (170)

- Duplicated block (10 lines × 4) (archive/kt-sft/ktransformers/operators/experts.py)
- Duplicated block (10–13 lines × 12) (archive/kt-sft/ktransformers/models/modeling_deepseek.py)
- Duplicated block (11 lines × 2) (archive/kt-sft/ktransformers/util/utils.py)
- Duplicated block (19 lines × 2) (archive/kt-sft/ktransformers/ktransformers_ext/triton/fp8gemm.py)
- Duplicated block (23–24 lines × 2) (kt-kernel/python/utils/loader.py)
- Duplicated block (35 lines × 4) (archive/kt-sft/ktransformers/operators/models.py)
- Duplicated block (46 lines × 2) (archive/kt-sft/ktransformers/util/custom_gguf.py)
- Duplicated block (47 lines × 2) (archive/kt-sft/ktransformers/ktransformers_ext/triton/fp8gemm.py)
- Duplicated block (5 lines × 4) (archive/kt-sft/ktransformers/server/balance_serve/inference/forward_batch.py)
- Duplicated block (7 lines × 3) (archive/kt-sft/ktransformers/util/custom_loader.py)
- Edited copies of one member (3 members, 15+ corresponding lines) (kt-kernel/python/sft/artifacts.py)
- Edited copies of one member (3 members, 25+ corresponding lines) (archive/ktransformers/operators/ascend/ascend_mlp.py)
- Edited copies of one member (3 members, 28+ corresponding lines) (kt-kernel/bench/bench_k2_write_buffer.py)
- Edited copies of one member (3 members, 4+ corresponding lines) (archive/kt-sft/ktransformers/util/custom_loader.py)
- Edited copies of one member (3 members, 48+ corresponding lines) (archive/kt-sft/ktransformers/configs/model_config/configuration_deepseek.py)
- Edited copies of one member (3 members, 52+ corresponding lines) (archive/csrc/ktransformers_ext/bench/bench_linear_torch.py)
- Edited copies of one member (4 members, 13+ corresponding lines) (kt-kernel/scripts/compare_weights.py)
- Edited copies of one member (4 members, 22+ corresponding lines) (archive/kt-sft/ktransformers/operators/cpuinfer.py)
- Edited copies of one member (4 members, 24+ corresponding lines) (archive/kt-sft/ktransformers/operators/cpuinfer.py)
- Edited copies of one member (4 members, 26+ corresponding lines) (archive/kt-sft/ktransformers/operators/experts.py)
- …and 150 more

## Changes since last survey

- 4 commits — 4 feature/other, 0 fixes

## By area

- doc/en — 1 commit
- kt-kernel/operators — 1 commit
- kt-kernel/python — 1 commit
- third_party/sglang — 1 commit

## Notable commits

- change: [build]: sync sglang submodule to 2c4085a01cf75cde927d35a259b4adaa7b07b3ab (#2270)
- change: [docs]: link community MXFP4 CPU expert weights for Ascend (#2257)
- change: [perf](kt-kernel): reuse BF16 LoRA recompute buffers (#2272)
- change: [perf](kt-kernel): share checkpoint indexes across pending MoE layers (#2231)

## Architecture

- Unchanged — 0 containers · 1 contexts · 0 edges
