Publish gemma-4-26B-A4B-it distribution-fidelity artifact (part 4)

gemma-4-26B-A4B-it-AWQ-4bit/attribution.json

@@ -9160,10 +9160,10 @@
     "routing_mode": "teacher_ids_student_weights",
     "routing_trace_manifest": "attribution/routing/routing-manifest.json",
     "routing_trace_protocol_version": 2,
-    "routing_trace_sha256": "0fcde10d9c35222c9efe3362b3439a790aec49caa70d829bd455bfac62b3ce9a",
+    "routing_trace_sha256": "c4a98d392c393b111c0d5f9eb554dd056b12706c68b7670e5a8cc2cce1b1c5b3",
     "score_from": 0,
     "timings": {
-      "score_forward": 167.04280401812866
+      "score_forward": 167.53983125044033
     },
     "token_sha256": "81d66409b44937fadcb5361f42b497543b6cfd377d7fb85c7dabdcd7afd57f32",
     "top1_agreement": 0.716593643136297,
@@ -10067,11 +10067,11 @@
     "routing_mode": "student_natural",
     "routing_trace_manifest": "attribution/routing/routing-manifest.json",
     "routing_trace_protocol_version": 2,
-    "routing_trace_sha256": "0fcde10d9c35222c9efe3362b3439a790aec49caa70d829bd455bfac62b3ce9a",
+    "routing_trace_sha256": "c4a98d392c393b111c0d5f9eb554dd056b12706c68b7670e5a8cc2cce1b1c5b3",
     "score_from": 0,
     "timings": {
-      "control_forward": 165.96417924482375,
-      "score_forward": 156.3145539299585
+      "control_forward": 165.76927155489102,
+      "score_forward": 156.034483961761
     },
     "token_sha256": "81d66409b44937fadcb5361f42b497543b6cfd377d7fb85c7dabdcd7afd57f32",
     "top1_agreement": 0.6932445601286435,

gemma-4-26B-A4B-it-AWQ-4bit/compliance.json

@@ -9161,10 +9161,10 @@
       "routing_mode": "teacher_ids_student_weights",
       "routing_trace_manifest": "attribution/routing/routing-manifest.json",
       "routing_trace_protocol_version": 2,
-      "routing_trace_sha256": "0fcde10d9c35222c9efe3362b3439a790aec49caa70d829bd455bfac62b3ce9a",
+      "routing_trace_sha256": "c4a98d392c393b111c0d5f9eb554dd056b12706c68b7670e5a8cc2cce1b1c5b3",
       "score_from": 0,
       "timings": {
-        "score_forward": 167.04280401812866
+        "score_forward": 167.53983125044033
       },
       "token_sha256": "81d66409b44937fadcb5361f42b497543b6cfd377d7fb85c7dabdcd7afd57f32",
       "top1_agreement": 0.716593643136297,
@@ -10068,11 +10068,11 @@
       "routing_mode": "student_natural",
       "routing_trace_manifest": "attribution/routing/routing-manifest.json",
       "routing_trace_protocol_version": 2,
-      "routing_trace_sha256": "0fcde10d9c35222c9efe3362b3439a790aec49caa70d829bd455bfac62b3ce9a",
+      "routing_trace_sha256": "c4a98d392c393b111c0d5f9eb554dd056b12706c68b7670e5a8cc2cce1b1c5b3",
       "score_from": 0,
       "timings": {
-        "control_forward": 165.96417924482375,
-        "score_forward": 156.3145539299585
+        "control_forward": 165.76927155489102,
+        "score_forward": 156.034483961761
       },
       "token_sha256": "81d66409b44937fadcb5361f42b497543b6cfd377d7fb85c7dabdcd7afd57f32",
       "top1_agreement": 0.6932445601286435,
@@ -10104,8 +10104,9 @@
     ],
     "kld_vocab_size": 262144,
     "kv_cache_dtype": "bfloat16",
-    "laws_version": 14,
+    "laws_version": 15,
     "model_runner_v2": false,
+    "numerics_digest": "251a9225b37415b91fc58d975b0670fc4b481e6f5a816b663573c42e161438f8",
     "reference_config_sha256": "ed0c1eb3633de771906e9ba004a44cc5635bcc06ee2062077c3d2e88a50707d3",
     "rows": 768,
     "score_from": 0,
@@ -10114,11 +10115,10 @@
     "suite_id": "gemma4-31b-it-fidelity-1024x2048-v1",
     "tensor_parallel_size": 1,
     "token_sha256": "81d66409b44937fadcb5361f42b497543b6cfd377d7fb85c7dabdcd7afd57f32",
-    "torch": "2.13.0+cu132",
-    "vllm_commit": "a2ae11e29e34294cfaa4d768f1a1d8d8199d1d61"
+    "torch": "2.13.0+cu132"
   },
   "compliant": true,
-  "evaluated_at": "2026-09-09T14:19:34.359704+00:00",
+  "evaluated_at": "2026-09-12T03:26:23.943396+00:00",
   "failed_laws": [],
   "findings": [
     {
@@ -10146,13 +10146,13 @@
       "title": "Real vocabulary"
     },
     {
-      "detail": "all bound fields present; manifest b5ab5ede01f46551b1932e913620d9554df5f97dd7c5e2bcca05f61db68725ef",
+      "detail": "all bound fields present; manifest fbc447717db774ab70ec4c9af8b197f0152475cfa119698fb5c2ee50b5dd1c07",
       "law": 5,
       "status": "pass",
       "title": "Manifest binding"
     },
     {
-      "detail": "torch 2.13.0+cu132, vLLM 0.1.dev20446+gb2bc9171d @ a2ae11e29e34, driver 580.173.02, 4 GPU(s)",
+      "detail": "torch 2.13.0+cu132, vLLM 0.1.dev20446+gb2bc9171d @ 60071d1ab732, driver 580.173.02, 4 GPU(s)",
       "law": 6,
       "status": "pass",
       "title": "Provenance"
@@ -10194,7 +10194,7 @@
       "title": "Reusable reference"
     },
     {
-      "detail": "QxQ 1.02626820, BxQ 0.84088786, QxQ - BxQ +0.18538034; natural selections changed at 62.737% of (token, layer) choices; trace 0fcde10d9c35222c",
+      "detail": "QxQ 1.02626820, BxQ 0.84088786, QxQ - BxQ +0.18538034; natural selections changed at 62.737% of (token, layer) choices; trace c4a98d392c393b11",
       "law": 14,
       "status": "pass",
       "title": "Routed-model intervention"
@@ -10224,7 +10224,7 @@
       "title": "Recorded deviation"
     }
   ],
-  "laws_version": 14,
+  "laws_version": 15,
   "mean_kld": 1.0262681996551193,
   "nondeterminism_floor": 0.0,
   "overridden_laws": [],

gemma-4-26B-A4B-it-AWQ-4bit/manifest.json

@@ -873,10 +873,11 @@
       "NVIDIA RTX PRO 6000 Blackwell Workstation Edition"
     ],
     "nccl": "2.29.7",
+    "numerics_digest": "251a9225b37415b91fc58d975b0670fc4b481e6f5a816b663573c42e161438f8",
     "platform": "Linux-6.8.0-137-generic-x86_64-with-glibc2.39",
     "python": "3.12.3",
     "torch": "2.13.0+cu132",
-    "vllm_commit": "a2ae11e29e34294cfaa4d768f1a1d8d8199d1d61",
+    "vllm_commit": "60071d1ab73229321712254b1350dcaaac1c125a",
     "vllm_dirty_digest": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
     "vllm_tree_dirty": false
   },

gemma-4-26B-A4B-it-AWQ-4bit/report.json

@@ -9156,10 +9156,10 @@
     "routing_mode": "teacher_ids_student_weights",
     "routing_trace_manifest": "/media/fmodels2/Local-Inference-Lab/work/gemma4-26b-a4b/captures/gemma-4-26B-A4B-it-ref-v0-tp1-analysis-c2048-s0-routing/routing-manifest.json",
     "routing_trace_protocol_version": 2,
-    "routing_trace_sha256": "0fcde10d9c35222c9efe3362b3439a790aec49caa70d829bd455bfac62b3ce9a",
+    "routing_trace_sha256": "c4a98d392c393b111c0d5f9eb554dd056b12706c68b7670e5a8cc2cce1b1c5b3",
     "score_from": 0,
     "timings": {
-      "score_forward": 167.04280401812866
+      "score_forward": 167.53983125044033
     },
     "top1_agreement": 0.716593643136297,
     "topk_agreement": {
@@ -9175,7 +9175,7 @@
     "token_sha256": "81d66409b44937fadcb5361f42b497543b6cfd377d7fb85c7dabdcd7afd57f32",
     "reference_config_sha256": "ed0c1eb3633de771906e9ba004a44cc5635bcc06ee2062077c3d2e88a50707d3"
   },
-  "capture_manifest_sha256": "b5ab5ede01f46551b1932e913620d9554df5f97dd7c5e2bcca05f61db68725ef",
+  "capture_manifest_sha256": "fbc447717db774ab70ec4c9af8b197f0152475cfa119698fb5c2ee50b5dd1c07",
   "confidence_buckets": [
     {
       "frac": 0.07217434558703795,
@@ -9214,6 +9214,9 @@
     }
   ],
   "context_length": 2048,
+  "declared_expert_activation_quant": [
+    "unquantized"
+  ],
   "deployed_mean_kld": 1.0262681996551193,
   "depth_buckets": [
     {
@@ -9257,6 +9260,7 @@
   "mean_kld_reverse": 1.0424924595989462,
   "median_kld": 0.2431836575269699,
   "model_runner_v2": false,
+  "moe_backend_named": null,
   "num_positions": 1572096,
   "num_rows": 768,
   "nvfp4_dense_scale_inspection": {
@@ -18425,11 +18429,11 @@
     "routing_mode": "student_natural",
     "routing_trace_manifest": "/media/fmodels2/Local-Inference-Lab/work/gemma4-26b-a4b/captures/gemma-4-26B-A4B-it-ref-v0-tp1-analysis-c2048-s0-routing/routing-manifest.json",
     "routing_trace_protocol_version": 2,
-    "routing_trace_sha256": "0fcde10d9c35222c9efe3362b3439a790aec49caa70d829bd455bfac62b3ce9a",
+    "routing_trace_sha256": "c4a98d392c393b111c0d5f9eb554dd056b12706c68b7670e5a8cc2cce1b1c5b3",
     "score_from": 0,
     "timings": {
-      "control_forward": 165.96417924482375,
-      "score_forward": 156.3145539299585
+      "control_forward": 165.76927155489102,
+      "score_forward": 156.034483961761
     },
     "top1_agreement": 0.6932445601286435,
     "topk_agreement": {
@@ -18553,10 +18557,11 @@
       "NVIDIA RTX PRO 6000 Blackwell Workstation Edition"
     ],
     "nccl": "2.29.7",
+    "numerics_digest": "251a9225b37415b91fc58d975b0670fc4b481e6f5a816b663573c42e161438f8",
     "platform": "Linux-6.8.0-137-generic-x86_64-with-glibc2.39",
     "python": "3.12.3",
     "torch": "2.13.0+cu132",
-    "vllm_commit": "a2ae11e29e34294cfaa4d768f1a1d8d8199d1d61",
+    "vllm_commit": "60071d1ab73229321712254b1350dcaaac1c125a",
     "vllm_dirty_digest": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
     "vllm_tree_dirty": false
   },
@@ -18659,15 +18664,15 @@
     }
   },
   "timings": {
-    "bxq_repeat_forward": 165.66919468762353,
-    "bxq_score_forward": 167.04280401812866,
-    "decompose_head": 257.8311422751285,
-    "qxq_control_forward": 165.96417924482375,
-    "qxq_control_repeat_forward": 165.65947952121496,
-    "qxq_natural_repeat_forward": 156.47465047379956,
-    "routing_compare": 7.6060543968342245,
-    "score_forward": 156.3145539299585,
-    "student_load": 24.379473275039345
+    "bxq_repeat_forward": 166.0894009261392,
+    "bxq_score_forward": 167.53983125044033,
+    "decompose_head": 262.916086464189,
+    "qxq_control_forward": 165.76927155489102,
+    "qxq_control_repeat_forward": 165.4896008549258,
+    "qxq_natural_repeat_forward": 156.22356016235426,
+    "routing_compare": 5.909322227817029,
+    "score_forward": 156.034483961761,
+    "student_load": 24.73130771610886
   },
   "top1_agreement": 0.6932445601286435,
   "topk_agreement": {

gemma-4-26B-A4B-it-AWQ-4bit/report.md

@@ -18,7 +18,7 @@ Reverse direction, KLD(candidate || reference): 1.04249246.
 | Candidate weights SHA-256 | f1f4146421ee33ad |
 | Suite | gemma4-31b-it-fidelity-1024x2048-v1 |
 | Suite token SHA-256 | 81d66409b44937fa |
-| Capture manifest SHA-256 | b5ab5ede01f46551 |
+| Capture manifest SHA-256 | fbc447717db774ab |
 | Tokenizer | None |
 | Scored vocabulary | 262144 |
 | Declared vocabulary | 262144 |
@@ -31,16 +31,19 @@ Reverse direction, KLD(candidate || reference): 1.04249246.
 | Prefix caching | False |
 | max_num_seqs | 1 |
 | vLLM | 0.1.dev20446+gb2bc9171d |
-| vLLM commit | a2ae11e29e34 |
+| vLLM commit | 60071d1ab732 |
 | vLLM dirty digest | e3b0c44298fc1c14 |
+| Numerics digest | 251a9225b37415b9 |
 | Compiled extensions | f2fbc7537b0f01f6 |
 | FlashInfer | 0.6.17 |
 | torch | 2.13.0+cu132 |
 | Driver | 580.173.02 |
 | GPUs | NVIDIA RTX PRO 6000 Blackwell Workstation Edition, NVIDIA RTX PRO 6000 Blackwell Workstation Edition, NVIDIA RTX PRO 6000 Blackwell Workstation Edition, NVIDIA RTX PRO 6000 Blackwell Workstation Edition |
-| Laws version | 14 |
+| Laws version | 15 |
 | Partition | analysis |
 
+The commit says when this number was taken. The numerics digest says what took it: a hash of the runtime and the scorer, which moves only when code that can change a logit changes. Comparability is bounded by the digest, so a result stays current across commits that cannot reach a number, and two results carrying the same digest were computed by the same code whatever their commits say.
+
 ## Routed-model intervention: QxQ and BxQ
 
 **The first axis is routing source; the second Q is the same unchanged quantized candidate in both runs.** QxQ uses the student's natural expert IDs. BxQ forces the BF16 teacher's expert IDs while the student computes its own gating weights for those experts.
@@ -52,7 +55,7 @@ Reverse direction, KLD(candidate || reference): 1.04249246.
 | QxQ - BxQ | paired routing-intervention delta; not additive attribution | +0.18538034 |  |  |
 | Exact repeat | two QxQ, two forced-natural, two BxQ samples | certified |  |  |
 
-Natural QxQ selections changed from the teacher at **62.7369%** of (token, layer) choices. BxQ protocol 5; routing trace `0fcde10d9c35222c`.
+Natural QxQ selections changed from the teacher at **62.7369%** of (token, layer) choices. BxQ protocol 5; routing trace `c4a98d392c393b11`.
 
 Natural control used exact-repeat certification: max position delta 0.000e+00 / 0.000e+00; absolute mean delta 0.000e+00 / 0.000e+00; natural route-repeat flips 0.000%.
 
@@ -123,6 +126,15 @@ The same expert weights rounded through each scheme, which is the cell that disc
 
 Full cells, digests, and the deployed checkpoint's inspection are in [attribution.json](attribution.json).
 
+## Expert kernels: declared against built
+
+| Property | Value |
+|---|---|
+| Declared for its experts | `unquantized` |
+| Expert implementation built | `MarlinExperts` |
+| Expert kernel built | `FusedMoEKernel` |
+| Expert layers carrying an activation scale | 0 of 30 |
+
 ## Routing divergence
 
 This is the observed selection divergence in the normal QxQ run, not the BxQ intervention and not a QDQ router-weight cell. It compares the student's natural expert IDs with the teacher's recorded IDs over 30 routed layers, 8 experts per token.
@@ -230,15 +242,15 @@ Zero baseline: reference against itself scored **0.00000000** over 2047 position
 | 2 | Determinism | PASS | eager enforced; prefix caching off, max_num_seqs=1 |
 | 3 | Frozen input | PASS | token hash matches suite gemma4-31b-it-fidelity-1024x2048-v1 [analysis]: 81d66409b44937fa |
 | 4 | Real vocabulary | PASS | scored 262144 real tokens of 262144 declared (0 padding rows) |
-| 5 | Manifest binding | PASS | all bound fields present; manifest b5ab5ede01f46551b1932e913620d9554df5f97dd7c5e2bcca05f61db68725ef |
-| 6 | Provenance | PASS | torch 2.13.0+cu132, vLLM 0.1.dev20446+gb2bc9171d @ a2ae11e29e34, driver 580.173.02, 4 GPU(s) |
+| 5 | Manifest binding | PASS | all bound fields present; manifest fbc447717db774ab70ec4c9af8b197f0152475cfa119698fb5c2ee50b5dd1c07 |
+| 6 | Provenance | PASS | torch 2.13.0+cu132, vLLM 0.1.dev20446+gb2bc9171d @ 60071d1ab732, driver 580.173.02, 4 GPU(s) |
 | 7 | Storage integrity | PASS | hidden storage with bitwise-exact replay |
 | 8 | Head transparency | PASS | trunk 1.02626820, deployed 1.02626820, delta -2.4355184535806984e-11 |
 | 9 | Tail and depth disclosure | PASS | mean 1.02626820, median 0.24318366, max 43.06172943, 4 depth buckets |
 | 10 | Comparability | PASS | comparability key fully resolved |
 | 11 | Freeze before qualification | NOT_APPLICABLE | partition is 'analysis' |
 | 12 | Reusable reference | PASS | suite, reference, head, and checksums present; published reference matches the scored capture on every bound field |
-| 14 | Routed-model intervention | PASS | QxQ 1.02626820, BxQ 0.84088786, QxQ - BxQ +0.18538034; natural selections changed at 62.737% of (token, layer) choices; trace 0fcde10d9c35222c |
+| 14 | Routed-model intervention | PASS | QxQ 1.02626820, BxQ 0.84088786, QxQ - BxQ +0.18538034; natural selections changed at 62.737% of (token, layer) choices; trace c4a98d392c393b11 |
 | 15 | Domain disclosure | PASS | 10 domains disclosed; weakest other_multilingual at 1.36233317, strongest code_docs_issues at 0.74656348, spread 1.8x |
 | 16 | Candidate weight binding | PASS | scored weights f1f4146421ee33ad as inspected |
 | 17 | Substitution disclosure | PASS | every scored layer used the checkpoint's own quantization parameters |
@@ -252,7 +264,7 @@ Zero baseline: reference against itself scored **0.00000000** over 2047 position
 | Platform | Linux-6.8.0-137-generic-x86_64-with-glibc2.39 |
 | Python | 3.12.3 |
 | vLLM | 0.1.dev20446+gb2bc9171d |
-| vLLM commit | a2ae11e29e34294cfaa4d768f1a1d8d8199d1d61 |
+| vLLM commit | 60071d1ab73229321712254b1350dcaaac1c125a |
 | torch | 2.13.0+cu132 |
 | torch CUDA runtime | 13.2 |
 | cuDNN | 9.20.0 (92000) |
@@ -260,7 +272,7 @@ Zero baseline: reference against itself scored **0.00000000** over 2047 position
 | CUDA arch list | sm_75, sm_80, sm_86, sm_90, sm_100, sm_120 |
 | nvcc | Cuda compilation tools, release 13.0, V13.0.88 |
 | gcc | gcc (Ubuntu 13.3.0-6ubuntu2~24.04.1) 13.3.0 |
-| glibc / ldd | ldd (Ubuntu GLIBC 2.39-0ubuntu8.8) 2.39 |
+| glibc / ldd | ldd (Ubuntu GLIBC 2.39-0ubuntu8.9) 2.39 |
 | NVIDIA driver | 580.173.02 |
 | float32 matmul precision | highest |
 | TF32 (matmul / cuDNN) | False / True |
@@ -287,6 +299,11 @@ Credential values are never published. Set but redacted: `HF_TOKEN`.
 | `CUDA_HOME` | `/usr/local/cuda-13.0` |
 | `HF_TOKEN` | `<redacted>` |
 | `NCCL_DETERMINISTIC` | `1` |
+| `PYTORCH_NVML_BASED_CUDA_CHECK` | `1` |
+| `TORCHINDUCTOR_CACHE_DIR` | `/tmp/torchinductor_phaedawg` |
+| `TORCHINDUCTOR_COMPILE_THREADS` | `1` |
+| `TRITON_CACHE_AUTOTUNING` | `1` |
+| `TRITON_PTXAS_BLACKWELL_PATH` | `/usr/local/cuda-13.0/bin/ptxas` |
 | `VLLM_BATCH_INVARIANT` | `1` |
 | `VLLM_MARLIN_USE_ATOMIC_ADD` | `0` |
 | `VLLM_MOE_USE_DEEP_GEMM` | `0` |
@@ -298,25 +315,25 @@ Paths are relative to the artifact root, the same paths `checksums.txt` uses. Ve
 | Path | Size | What it is |
 |---|---|---|
 | `gemma-4-26B-A4B-it-AWQ-4bit/report.md` | 22.06 KiB | This document. |
-| `gemma-4-26B-A4B-it-AWQ-4bit/report.json` | 566.64 KiB | Every statistic behind it, machine-readable: per-bucket means, percentiles, agreement rates, and the phase timings. |
-| `gemma-4-26B-A4B-it-AWQ-4bit/manifest.json` | 87.35 KiB | The capture manifest this result is bound to (Law 5). Scoring refuses to run if the live configuration differs from it. |
-| `gemma-4-26B-A4B-it-AWQ-4bit/compliance.json` | 378.57 KiB | The law-by-law receipt, including the comparability key. |
-| `baselines/self-kld.json` | 6.89 KiB | The zero-baseline proof required by Law 1: the reference scored against a capture of itself. |
+| `gemma-4-26B-A4B-it-AWQ-4bit/report.json` | 566.81 KiB | Every statistic behind it, machine-readable: per-bucket means, percentiles, agreement rates, and the phase timings. |
+| `gemma-4-26B-A4B-it-AWQ-4bit/manifest.json` | 87.44 KiB | The capture manifest this result is bound to (Law 5). Scoring refuses to run if the live configuration differs from it. |
+| `gemma-4-26B-A4B-it-AWQ-4bit/compliance.json` | 378.60 KiB | The law-by-law receipt, including the comparability key. |
+| `baselines/self-kld.json` | 7.05 KiB | The zero-baseline proof required by Law 1: the reference scored against a capture of itself. |
 | `suite/suite-manifest.json` | 382.61 KiB | The frozen evaluation input's identity: token hashes per context and per partition, sources, strata, and the analysis/qualification split. |
 | `suite/tokens` | 17.32 MiB in 1024 files | The token IDs themselves. These are the evaluation input, not a description of it; retokenizing source text does not reproduce them. |
 | `suite/sources.json` | 737.48 KiB | Per-context provenance: dataset, revision, licence, source unit, and the deterministic token offset chosen within the document. |
 | `suite/validation/capability-overlap.json` | 1.48 KiB | The benchmark-contamination scan and every document it blocked. |
-| `reference/manifest.json` | 87.35 KiB | The reference capture's own manifest: geometry, vocabulary, storage mode, and a hash for every tensor file. |
+| `reference/manifest.json` | 87.44 KiB | The reference capture's own manifest: geometry, vocabulary, storage mode, and a hash for every tensor file. |
 | `reference` | 11.62 GiB in 771 files | The reusable reference distributions. Pass this directory as `--reference-logits` to score a new candidate against the same reference without loading the reference checkpoint. |
 | `reference/lm_head.safetensors` | 1.38 GiB | The reference language-model head, which turns the stored hidden states back into reference logits. |
-| `environment/runtime.json` | 5.08 KiB | Machine-readable provenance: torch, CUDA, cuDNN, NCCL, driver, devices, and the captured environment variables. |
+| `environment/runtime.json` | 5.42 KiB | Machine-readable provenance: torch, CUDA, cuDNN, NCCL, driver, devices, and the captured environment variables. |
 | `environment/summary.md` | 1.10 KiB | The same provenance as prose, plus an index of every captured file. |
 | `environment/toolchain-nvcc.txt` | 243 B | `nvcc --version` verbatim; `toolchain-gcc.txt` and `toolchain-ldd.txt` sit beside it. |
 | `environment/gpu-smi-query.txt` | 54.73 KiB | `nvidia-smi -q` verbatim: ECC state, persistence mode, clocks, and throttle reasons, any of which can move a bitwise result. |
 | `environment/pip-freeze.txt` | 4.47 KiB | Every installed package version in the scoring environment. |
 | `environment/models` | 28.70 KiB in 7 files | Checkpoint fingerprints: file listing, sizes, config and tokenizer hashes, and `config.json` verbatim for each model scored. |
 | `checksums.txt` | 1.13 MiB | `sha256sum --check` compatible over every other file here. This is authoritative for integrity (Law 12). |
-| `LAWS.md` | 38.37 KiB | The laws this artifact was produced under, including the override procedure. |
+| `LAWS.md` | 40.20 KiB | The laws this artifact was produced under, including the override procedure. |
 
 ## Scope
 

gemma-4-26B-A4B-it-FP8-dynamic/attribution/gemma-4-26B-A4B-it-qdq-experts-attention-dense_mlp-fp8_per_channel.json

@@ -1,5 +1,5 @@
 {
-  "capture_manifest_sha256": "b5ab5ede01f46551b1932e913620d9554df5f97dd7c5e2bcca05f61db68725ef",
+  "capture_manifest_sha256": "fbc447717db774ab70ec4c9af8b197f0152475cfa119698fb5c2ee50b5dd1c07",
   "confidence_buckets": [
     {
       "frac": 0.07217434558703795,
@@ -38,6 +38,7 @@
     }
   ],
   "context_length": 2048,
+  "declared_expert_activation_quant": [],
   "depth_buckets": [
     {
       "depth_hi": 511,
@@ -71,6 +72,7 @@
   "mean_kld_reverse": 0.6055071098612179,
   "median_kld": 0.08827311545610428,
   "model_runner_v2": false,
+  "moe_backend_named": null,
   "num_positions": 1572096,
   "num_rows": 768,
   "nvfp4_dense_scale_inspection": {
@@ -8638,10 +8640,11 @@
       "NVIDIA RTX PRO 6000 Blackwell Workstation Edition"
     ],
     "nccl": "2.29.7",
+    "numerics_digest": "251a9225b37415b91fc58d975b0670fc4b481e6f5a816b663573c42e161438f8",
     "platform": "Linux-6.8.0-137-generic-x86_64-with-glibc2.39",
     "python": "3.12.3",
     "torch": "2.13.0+cu132",
-    "vllm_commit": "a2ae11e29e34294cfaa4d768f1a1d8d8199d1d61",
+    "vllm_commit": "60071d1ab73229321712254b1350dcaaac1c125a",
     "vllm_dirty_digest": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
     "vllm_tree_dirty": false
   },
@@ -8744,9 +8747,9 @@
     }
   },
   "timings": {
-    "routing_compare": 7.93826940888539,
-    "score_forward": 164.40250429091975,
-    "student_load": 22.570553320925683
+    "routing_compare": 8.198981787543744,
+    "score_forward": 165.1197206461802,
+    "student_load": 23.07501409109682
   },
   "top1_agreement": 0.7729655186451718,
   "topk_agreement": {

gemma-4-26B-A4B-it-FP8-dynamic/attribution/gemma-4-26B-A4B-it-qdq-experts-fp8_per_channel.json

@@ -1,5 +1,5 @@
 {
-  "capture_manifest_sha256": "b5ab5ede01f46551b1932e913620d9554df5f97dd7c5e2bcca05f61db68725ef",
+  "capture_manifest_sha256": "fbc447717db774ab70ec4c9af8b197f0152475cfa119698fb5c2ee50b5dd1c07",
   "confidence_buckets": [
     {
       "frac": 0.07217434558703795,
@@ -38,6 +38,7 @@
     }
   ],
   "context_length": 2048,
+  "declared_expert_activation_quant": [],
   "depth_buckets": [
     {
       "depth_hi": 511,
@@ -71,6 +72,7 @@
   "mean_kld_reverse": 0.46927067081799145,
   "median_kld": 0.05512994900345802,
   "model_runner_v2": false,
+  "moe_backend_named": null,
   "num_positions": 1572096,
   "num_rows": 768,
   "nvfp4_dense_scale_inspection": {
@@ -8638,10 +8640,11 @@
       "NVIDIA RTX PRO 6000 Blackwell Workstation Edition"
     ],
     "nccl": "2.29.7",
+    "numerics_digest": "251a9225b37415b91fc58d975b0670fc4b481e6f5a816b663573c42e161438f8",
     "platform": "Linux-6.8.0-137-generic-x86_64-with-glibc2.39",
     "python": "3.12.3",
     "torch": "2.13.0+cu132",
-    "vllm_commit": "a2ae11e29e34294cfaa4d768f1a1d8d8199d1d61",
+    "vllm_commit": "60071d1ab73229321712254b1350dcaaac1c125a",
     "vllm_dirty_digest": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
     "vllm_tree_dirty": false
   },
@@ -8744,9 +8747,9 @@
     }
   },
   "timings": {
-    "routing_compare": 7.074598795268685,
-    "score_forward": 165.17221873020753,
-    "student_load": 22.120783646125346
+    "routing_compare": 8.236040574032813,
+    "score_forward": 166.57450863486156,
+    "student_load": 22.73681427212432
   },
   "top1_agreement": 0.8040806668295066,
   "topk_agreement": {

gemma-4-26B-A4B-it-FP8-dynamic/attribution/gemma-4-26B-A4B-it-qdq-experts-mxfp8.json

@@ -1,5 +1,5 @@
 {
-  "capture_manifest_sha256": "b5ab5ede01f46551b1932e913620d9554df5f97dd7c5e2bcca05f61db68725ef",
+  "capture_manifest_sha256": "fbc447717db774ab70ec4c9af8b197f0152475cfa119698fb5c2ee50b5dd1c07",
   "confidence_buckets": [
     {
       "frac": 0.07217434558703795,
@@ -38,6 +38,7 @@
     }
   ],
   "context_length": 2048,
+  "declared_expert_activation_quant": [],
   "depth_buckets": [
     {
       "depth_hi": 511,
@@ -71,6 +72,7 @@
   "mean_kld_reverse": 0.4616639767880551,
   "median_kld": 0.05473179370164871,
   "model_runner_v2": false,
+  "moe_backend_named": null,
   "num_positions": 1572096,
   "num_rows": 768,
   "nvfp4_dense_scale_inspection": {
@@ -8638,10 +8640,11 @@
       "NVIDIA RTX PRO 6000 Blackwell Workstation Edition"
     ],
     "nccl": "2.29.7",
+    "numerics_digest": "251a9225b37415b91fc58d975b0670fc4b481e6f5a816b663573c42e161438f8",
     "platform": "Linux-6.8.0-137-generic-x86_64-with-glibc2.39",
     "python": "3.12.3",
     "torch": "2.13.0+cu132",
-    "vllm_commit": "a2ae11e29e34294cfaa4d768f1a1d8d8199d1d61",
+    "vllm_commit": "60071d1ab73229321712254b1350dcaaac1c125a",
     "vllm_dirty_digest": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
     "vllm_tree_dirty": false
   },
@@ -8744,9 +8747,9 @@
     }
   },
   "timings": {
-    "routing_compare": 8.242558104451746,
-    "score_forward": 166.26376615883783,
-    "student_load": 22.620598616078496
+    "routing_compare": 8.28942439565435,
+    "score_forward": 166.19212660426274,
+    "student_load": 22.985856047831476
   },
   "top1_agreement": 0.8053471289285132,
   "topk_agreement": {

gemma-4-26B-A4B-it-FP8-dynamic/attribution/gemma-4-26B-A4B-it-qdq-experts-nvfp4.json

@@ -1,5 +1,5 @@
 {
-  "capture_manifest_sha256": "b5ab5ede01f46551b1932e913620d9554df5f97dd7c5e2bcca05f61db68725ef",
+  "capture_manifest_sha256": "fbc447717db774ab70ec4c9af8b197f0152475cfa119698fb5c2ee50b5dd1c07",
   "confidence_buckets": [
     {
       "frac": 0.07217434558703795,
@@ -38,6 +38,7 @@
     }
   ],
   "context_length": 2048,
+  "declared_expert_activation_quant": [],
   "depth_buckets": [
     {
       "depth_hi": 511,
@@ -71,6 +72,7 @@
   "mean_kld_reverse": 0.6656885178338737,
   "median_kld": 0.10631347447633743,
   "model_runner_v2": false,
+  "moe_backend_named": null,
   "num_positions": 1572096,
   "num_rows": 768,
   "nvfp4_dense_scale_inspection": {
@@ -8638,10 +8640,11 @@
       "NVIDIA RTX PRO 6000 Blackwell Workstation Edition"
     ],
     "nccl": "2.29.7",
+    "numerics_digest": "251a9225b37415b91fc58d975b0670fc4b481e6f5a816b663573c42e161438f8",
     "platform": "Linux-6.8.0-137-generic-x86_64-with-glibc2.39",
     "python": "3.12.3",
     "torch": "2.13.0+cu132",
-    "vllm_commit": "a2ae11e29e34294cfaa4d768f1a1d8d8199d1d61",
+    "vllm_commit": "60071d1ab73229321712254b1350dcaaac1c125a",
     "vllm_dirty_digest": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
     "vllm_tree_dirty": false
   },
@@ -8744,9 +8747,9 @@
     }
   },
   "timings": {
-    "routing_compare": 8.273550726938993,
-    "score_forward": 163.96420371392742,
-    "student_load": 22.59668383700773
+    "routing_compare": 8.257802481763065,
+    "score_forward": 163.95397236617282,
+    "student_load": 22.78515542205423
   },
   "top1_agreement": 0.7592182665689627,
   "topk_agreement": {

gemma-4-26B-A4B-it-FP8-dynamic/attribution/routing/routing-manifest.json

@@ -1,5 +1,5 @@
 {
-  "capture_manifest_sha256": "b5ab5ede01f46551b1932e913620d9554df5f97dd7c5e2bcca05f61db68725ef",
+  "capture_manifest_sha256": "fbc447717db774ab70ec4c9af8b197f0152475cfa119698fb5c2ee50b5dd1c07",
   "file_hashes": {
     "routing_00000.safetensors": "b15d371f91aff64ce1f741b0fa20fec537834f00753b7e789418d5fbd24ab207",
     "routing_00001.safetensors": "91d03a3baff046afb21017604984267d9abd885408acc1cbcdce2e66784d94c5",
@@ -838,10 +838,11 @@
       "NVIDIA RTX PRO 6000 Blackwell Workstation Edition"
     ],
     "nccl": "2.29.7",
+    "numerics_digest": "251a9225b37415b91fc58d975b0670fc4b481e6f5a816b663573c42e161438f8",
     "platform": "Linux-6.8.0-137-generic-x86_64-with-glibc2.39",
     "python": "3.12.3",
     "torch": "2.13.0+cu132",
-    "vllm_commit": "a2ae11e29e34294cfaa4d768f1a1d8d8199d1d61",
+    "vllm_commit": "60071d1ab73229321712254b1350dcaaac1c125a",
     "vllm_dirty_digest": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
     "vllm_tree_dirty": false
   },