[justinchuby]

Scribe consolidation of the graph-capture arc. Merges 7 inbox drops into decisions.md (15636->19014B), marks the V2-Lite _d1 fix #1181 MERGED and the graph-capture top lever FULLY LANDED (classifier #1171 + _d1 #1181 = 1.79x V2-Lite MoE capture, byte-identical, opt-in). Squad state only — no code changes.

Co-authored-by: Copilot 223556219+Copilot@users.noreply.github.com

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 79.86%. Comparing base (c9c7f64) to head (9c52ba3).
⚠️ Report is 20 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff            @@
##           main    #1187       +/-   ##
=========================================
+ Coverage      0   79.86%   +79.86%     
=========================================
  Files         0      359      +359     
  Lines         0   157760   +157760     
  Branches      0   157760   +157760     
=========================================
+ Hits          0   125999   +125999     
- Misses        0    27150    +27150     
- Partials      0     4611     +4611     

Flag	Coverage Δ
mlas	85.39% <ø> (?)
offline	79.75% <ø> (?)

Flags with carried forward coverage won't be shown. Click here to find out more.
see 359 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[github-actions]

🔴 Benchmark Regression Detected

Comparison of criterion micro-benchmarks: PR head vs merge-base, measured on the same runner in the same job (base first → PR second).

ℹ️ Absolute times are informational only — they vary with runner load. The % change column is the reliable signal because both sides ran under identical conditions.

Status	Scenario	Base	PR	Change
🔴	gather/medium_bf16_threads=1-internal/32768	2.34 µs	3.39 µs	+44.9%
🔴	gather/large_f32_threads=1-internal/131072	35.02 µs	46.45 µs	+32.6%
⚠️	gather/large_f16_threads=1-internal/131072	13.25 µs	16.71 µs	+26.1%
⚠️	grammar_masking/llguidance_compute_mask/32	78.43 µs	96.90 µs	+23.6%
⚠️	qwen3_sampling_processors/top_k_top_p_full_sort_baseline	5.60 ms	6.58 ms	+17.5%
✅	logit_processing/seven_processor_chain_per_step	317.42 µs	362.37 µs	+14.2%
✅	qwen3_sampling_processors/top_p_full_sort_after_top_k_baseline	3.73 ms	4.08 ms	+9.4%
✅	matmul/small_generic_f16_threads=1/1x256x256	31.96 µs	34.50 µs	+7.9%
✅	qwen3_sampling_processors/top_k_top_p_fast	697.59 µs	747.49 µs	+7.2%
✅	matmul/medium_generic_f32_threads=1/32x512x512	2.36 ms	2.48 ms	+5.2%
✅	qwen3_sampling_processors/top_k_full_sort_baseline	2.35 ms	2.47 ms	+5.1%
✅	matmul/small_generic_bf16_threads=8/1x256x256	36.14 µs	37.98 µs	+5.1%
✅	matmul/medium_generic_bf16_threads=1/32x512x512	532.12 µs	558.89 µs	+5.0%
✅	tokenization/decode_tokens_per_second	6.36 ms	6.64 ms	+4.4%
✅	kv_cache/alloc_dealloc_pages	41.44 µs	43.23 µs	+4.3%
✅	gather/medium_f16_threads=1-internal/32768	2.58 µs	2.69 µs	+4.3%
✅	matmul/medium_generic_f16_threads=1/32x512x512	30.58 µs	31.88 µs	+4.3%
✅	matmul/small_generic_bf16_threads=1/1x256x256	36.28 µs	37.41 µs	+3.1%
✅	matmul/small_generic_f32_threads=8/1x256x256	44.17 µs	45.16 µs	+2.2%
✅	qwen3_sampling_processors/top_p_fast_after_top_k	568.32 µs	580.58 µs	+2.2%
✅	sampling_latency/min_p_per_token	230.53 µs	235.30 µs	+2.1%
✅	matmul/medium_generic_f16_threads=8/32x512x512	30.13 µs	30.70 µs	+1.9%
✅	tokenization/encode_tokens_per_second	394.47 µs	401.90 µs	+1.9%
✅	gather/large_bf16_threads=1-internal/131072	15.50 µs	15.77 µs	+1.7%
✅	matmul/medium_generic_bf16_threads=8/32x512x512	403.87 µs	406.28 µs	+0.6%
✅	block_quantized_moe_cached_dense/mxfp4_cached_dense_expert_repeated_call/rows=1,H=256,I=256,E=4,top_k=1	175.11 µs	175.57 µs	+0.3%
✅	matmul/large_generic_bf16_threads=1/32x1024x1024	2.15 ms	2.09 ms	-3.0%
✅	sampling_latency/greedy_per_token	3.51 µs	3.39 µs	-3.7%
✅	qwen3_sampling_processors/top_k_partial_selection	161.00 µs	153.23 µs	-4.8%
✅	matmul/small_generic_f32_threads=1/1x256x256	40.37 µs	37.97 µs	-5.9%
✅	block_quantized_matmul_cached_dense/mxfp4_preexpanded_dense_oncelock_like_proxy/1x1024x1024	55.14 µs	51.31 µs	-6.9%
✅	block_quantized_moe_cached_dense/mxfp4_uncached_expert_dequant_each_call/rows=1,H=256,I=256,E=4,top_k=1	552.47 µs	510.68 µs	-7.6%
✅	matmul/medium_generic_f32_threads=8/32x512x512	1.11 ms	1.02 ms	-7.8%
✅	reduce_mean/large_f32_threads=1-internal/262144	990.85 µs	911.10 µs	-8.0%
✅	add/large_f32_threads=1-internal/4194304	683.89 µs	624.52 µs	-8.7%
✅	add/medium_f16_threads=1-internal/262144	107.64 µs	97.69 µs	-9.2%
✅	matmul/large_generic_f16_threads=8/32x1024x1024	97.10 µs	88.07 µs	-9.3%
✅	gather/medium_f32_threads=1-internal/32768	4.13 µs	3.73 µs	-9.6%
✅	add/large_f16_threads=1-internal/4194304	1.75 ms	1.58 ms	-9.7%
✅	add/medium_bf16_threads=1-internal/262144	108.77 µs	97.49 µs	-10.4%
✅	add/small_f16_threads=1-internal/1024	533.6 ns	476.0 ns	-10.8%
✅	matmul/large_generic_bf16_threads=8/32x1024x1024	1.61 ms	1.42 ms	-11.9%
✅	matmul/small_generic_f16_threads=8/1x256x256	35.45 µs	31.08 µs	-12.3%
✅	reduce_mean/medium_f32_threads=1-internal/65536	268.74 µs	234.46 µs	-12.8%
✅	sampling_latency/top_p_per_token	500.95 µs	430.33 µs	-14.1%
🟢	add/large_bf16_threads=1-internal/4194304	1.85 ms	1.56 ms	-15.7%
🟢	add/small_bf16_threads=1-internal/1024	511.9 ns	423.1 ns	-17.3%
🟢	sampling_latency/top_k_per_token	66.40 µs	54.43 µs	-18.0%
🟢	add/medium_f32_threads=1-internal/262144	28.06 µs	22.82 µs	-18.6%
🟢	matmul/large_generic_f32_threads=1/32x1024x1024	12.35 ms	10.00 ms	-19.1%
🟢	gather/small_f32_threads=1-internal/4096	777.6 ns	621.6 ns	-20.1%
🟢	reduce_mean/small_f32_threads=1-internal/4096	18.01 µs	13.98 µs	-22.4%
🟢	gather/small_f16_threads=1-internal/4096	601.8 ns	439.4 ns	-27.0%
🟢	gather/small_bf16_threads=1-internal/4096	610.8 ns	443.6 ns	-27.4%
🟢	matmul/large_generic_f16_threads=1/32x1024x1024	117.01 µs	82.65 µs	-29.4%
🟢	block_quantized_matmul_cached_dense/mxfp4_uncached_dequant_each_call/1x1024x1024	912.54 µs	612.85 µs	-32.8%
🟢	add/small_f32_threads=1-internal/1024	283.8 ns	188.3 ns	-33.6%
🟢	matmul/large_generic_f32_threads=8/32x1024x1024	6.69 ms	4.14 ms	-38.1%
🟢	block_quantized_matmul_cached_dense/mxfp4_cached_dense_repeated_call/1x1024x1024	93.53 µs	57.05 µs	-39.0%

Visual flags: ⚠️ ≥ 15% slower, 🔴 ≥ 30% slower — calibrated against measured runner noise (~27% worst-case on multi-threaded matmul)

Host info

CPU: Apple M1 (Virtual)
Cores: 3
OS: Darwin 25.5.0 arm64
Rust: rustc 1.97.1 (8bab26f4f 2026-07-14)
Load avg: { 5.71 4.59 5.91 }

What this cannot catch

• Regressions in code paths not covered by these benchmarks (e.g., end-to-end decode with a real model)
• Sub-threshold regressions that compound over multiple PRs
• Performance changes that only manifest under GPU execution
• Latency changes in the ORT integration path (these benchmarks exercise the native Rust kernels)