Add AgentFEM runtime validation and handoff

Adds checksum-verified DENIM runtime evidence, exact software identities, documented tangent boundary, and a data/model handoff for downstream research teams.

AGENTFEM_RUNTIME_SHA256SUMS

@@ -0,0 +1,5 @@
+54b80792286cdc6fc42394be0a31e539227f12ede63d24ac08f192cd6931da29  README.md
+5124cd03d84ee846f5f874fffab7df616ab69a64a08288066890380ee1d1aa2b  artifacts/t2_denim_agentfem_v1/RUNTIME_VALIDATION.md
+3c7b642c1888abe566bb3beec7cbabc36466fb6dbab2b62d193f14a248841174  artifacts/t2_denim_agentfem_v1/runtime_validation.json
+3d1726dcc00bf9e888647632e7dfe4b58771ece931461cd988b614098f1746d3  docs/T2_ALL_DATA_STAGE_SUMMARY.md
+830607c710ec01a03e416f0f5cf216d0828506b2d3f437c1216ed692a37b596a  docs/T2_DATA_MODEL_HANDOFF.md

README.md

@@ -98,6 +98,26 @@ Expanded-DENIM RMSE on matched 61/121/481/961-state histories was
 - Full metrics: `artifacts/t2_denim_boundary_v1/model_metrics.json`
 - All-data usage: `docs/T2_ALL_DATA_STAGE_SUMMARY.md`
 
+## AgentFEM runtime validation
+
+The expanded DENIM checkpoint is also available as a checksum-authenticated
+`safetensors` bundle in the
+[AgentFEM-DENIM model repository](https://huggingface.co/HaomingLuo/AgentFEM-DENIM).
+The safe bundle reproduces the legacy implementation on the fixed 121-step
+path to a maximum stress difference of `2.68e-7 Pa` and identical final PEEQ.
+
+Serial and two-rank AgentFEM implicit plastic-bar runs both completed 4/4
+increments, with a reported maximum-stress difference of `5.96e-8 Pa` between
+the two executions. The current plastic automatic-differentiation tangent has
+a documented 0.05–0.80% discrepancy against fixed-old-state finite
+differences over the audited plastic states. The demonstrated global cases
+converge; an exact consistent plastic tangent remains outside the present
+claim boundary.
+
+- Machine evidence: `artifacts/t2_denim_agentfem_v1/runtime_validation.json`
+- Validation note: `artifacts/t2_denim_agentfem_v1/RUNTIME_VALIDATION.md`
+- Data/model handoff: `docs/T2_DATA_MODEL_HANDOFF.md`
+
 ## Multiaxial v2 at a glance
 
 - eight path families, including non-proportional, rotating and random

artifacts/t2_denim_agentfem_v1/RUNTIME_VALIDATION.md

@@ -0,0 +1,49 @@
+# DENIM × AgentFEM runtime validation
+
+## Result
+
+The expanded 918-parameter DENIM checkpoint was migrated from the trusted
+legacy PyTorch file to a checksum-authenticated `safetensors` bundle and loaded
+through the generic AgentFEM learned-constitutive contract.
+
+The 121-step material-point result agrees with the legacy implementation to a
+maximum stress difference of `2.68e-7 Pa`; final PEEQ is identical. The bundle
+produces a maximum absolute stress of `292.547138 MPa`, final PEEQ of
+`0.0095270883`, and maximum yield residual of `0.401 Pa` on the frozen runtime
+path.
+
+## Global implicit execution
+
+| Case | Processes | Prescribed displacement | Maximum absolute stress | Increments | Result |
+|---|---:|---:|---:|---:|---|
+| Elastic bar | 1 | 0.001 m | 190.000 MPa | 4/4 | converged |
+| Plastic bar | 1 | 0.008 m | 500.441 MPa | 4/4 | converged |
+| Plastic bar | 2 | 0.008 m | 500.441 MPa | 4/4 | converged |
+
+The serial and two-rank plastic results differ by `5.96e-8 Pa` in reported
+maximum stress and have identical maximum displacement.
+
+## Tangent boundary
+
+The elastic tangent agrees with fixed-old-state central differences to
+`9.78e-14` relative error. Through the fixed-iteration plastic return map, the
+current automatic-differentiation tangent differs from finite differences by
+approximately 0.05–0.80% across the audited plastic states. The global tests
+above converge without cutback, while the stricter `2e-5` material-point
+tangent criterion is not met in plastic loading.
+
+The runtime evidence therefore supports deployable implicit execution and
+reproducible solver integration. It does not support a claim of a
+mathematically exact consistent plastic tangent or production UMAT
+certification. Full machine-readable evidence is in `runtime_validation.json`.
+
+## Fixed identity
+
+- AgentFEM: `40c965225fc3b8d12a93a68e0d6b01dd35d47d82`
+- AgentFEM-learning: `06d454e4ee66a7ee4f4c8a705d5958ff9577d18c`
+- source model revision: `5629df0a23a3d1ed43e9de2150e3d33cb979fdc1`
+- dataset revision: `c84f416e5a71daa157e406c50afc3fc73509b9ca`
+- safe weights SHA-256: `2ed35af21f3c11d6f2237485871ac39d63b084eeb47b6c341175ab6ccb0f960a`
+
+All 34 T2 project regressions, 18 AgentFEM learned-material tests and 11
+AgentFEM-learning provider tests passed.

artifacts/t2_denim_agentfem_v1/runtime_validation.json

@@ -0,0 +1,132 @@
+{
+  "schema": "agentfem.physics-data.denim-runtime-validation",
+  "schema_version": "1.0.0",
+  "status": "passed_with_documented_tangent_boundary",
+  "validated_at": "2026-09-25T13:30:00+08:00",
+  "software": {
+    "agentfem_version": "0.3.7.dev0",
+    "agentfem_commit": "40c965225fc3b8d12a93a68e0d6b01dd35d47d82",
+    "agentfem_learning_version": "0.1.0a1",
+    "agentfem_learning_commit": "06d454e4ee66a7ee4f4c8a705d5958ff9577d18c",
+    "python": "3.11.15",
+    "pytorch": "2.13.0",
+    "numpy": "2.4.6",
+    "h5py": "3.16.0",
+    "mpi_vendor": "MPICH 5.0.1"
+  },
+  "identity": {
+    "model_repo": "HaomingLuo/AgentFEM-DENIM",
+    "source_model_revision": "5629df0a23a3d1ed43e9de2150e3d33cb979fdc1",
+    "dataset_repo": "HaomingLuo/AgentFEM-Material-Loading-Memory",
+    "dataset_revision": "c84f416e5a71daa157e406c50afc3fc73509b9ca",
+    "model_version": "1.1.0",
+    "architecture": "denim.v1",
+    "parameter_count": 918,
+    "weights_format": "safetensors",
+    "weights_sha256": "2ed35af21f3c11d6f2237485871ac39d63b084eeb47b6c341175ab6ccb0f960a",
+    "manifest_sha256": "5c3636278999f1b92c5b8b94fc8500e8dd270b5f7285dbd53fc5d0a3b6976105"
+  },
+  "dataset_audit": {
+    "complete_trajectories": 2660,
+    "t2_v1": 1008,
+    "t2_multiaxial_v2": 1024,
+    "denim_closure_v1": 128,
+    "denim_boundary_v1": 500,
+    "t2_regression_tests_passed": 34,
+    "t2_regression_tests_failed": 0,
+    "boundary_release_checksums_verified": true
+  },
+  "safe_bundle_equivalence": {
+    "path_steps": 121,
+    "maximum_absolute_stress_difference_pa": 2.682209014892578e-7,
+    "stress_rmse_pa": 2.090951128498144e-8,
+    "final_peeq_difference": 0.0,
+    "maximum_absolute_stress_mpa": 292.54713784451934,
+    "final_peeq": 0.009527088265532677,
+    "maximum_yield_residual_pa": 0.40072572231292725
+  },
+  "tangent_audit": {
+    "method": "central_difference_fixed_old_state_vs_autodiff",
+    "relative_step": 1e-6,
+    "strict_acceptance_tolerance": 2e-5,
+    "points": [
+      {
+        "equivalent_path_strain": 0.001,
+        "regime": "elastic",
+        "relative_error": 9.777786913369087e-14,
+        "strictly_accepted": true
+      },
+      {
+        "equivalent_path_strain": 0.0015,
+        "regime": "near_yield",
+        "relative_error": 0.000463908560930518,
+        "strictly_accepted": false
+      },
+      {
+        "equivalent_path_strain": 0.002,
+        "regime": "plastic",
+        "relative_error": 0.00151667819950217,
+        "strictly_accepted": false
+      },
+      {
+        "equivalent_path_strain": 0.003,
+        "regime": "plastic",
+        "relative_error": 0.0034972718149479777,
+        "strictly_accepted": false
+      },
+      {
+        "equivalent_path_strain": 0.006,
+        "regime": "plastic",
+        "relative_error": 0.008029543517786931,
+        "strictly_accepted": false
+      }
+    ],
+    "interpretation": "The current autodiff tangent is exact in the elastic regime and approximate through the fixed-iteration plastic return map. The observed plastic-point discrepancy is 0.05-0.80 percent over this audit. Global solves below completed without cutback, but the artifact is not claimed to provide a mathematically exact consistent plastic tangent."
+  },
+  "global_implicit_validation": {
+    "elastic_serial": {
+      "prescribed_displacement_m": 0.001,
+      "maximum_absolute_stress_pa": 190000000.00000003,
+      "accepted_increments": 4,
+      "attempted_increments": 4,
+      "converged": true
+    },
+    "plastic_serial": {
+      "prescribed_displacement_m": 0.008,
+      "maximum_absolute_stress_pa": 500440565.1864772,
+      "accepted_increments": 4,
+      "attempted_increments": 4,
+      "converged": true
+    },
+    "plastic_mpi2": {
+      "prescribed_displacement_m": 0.008,
+      "maximum_absolute_stress_pa": 500440565.1864771,
+      "accepted_increments": 4,
+      "attempted_increments": 4,
+      "converged": true
+    },
+    "serial_mpi2_maximum_stress_difference_pa": 5.960464477539063e-8,
+    "serial_mpi2_displacement_difference_m": 0.0,
+    "applicability_status": "in_domain"
+  },
+  "software_regression": {
+    "agentfem_tests_passed": 18,
+    "agentfem_learning_tests_passed": 11,
+    "failures": 0
+  },
+  "claim_boundary": {
+    "supported": [
+      "lossless legacy-to-safetensors checkpoint migration",
+      "offline checksum-verified loading",
+      "material-point and batch inference",
+      "state commit and result evidence through AgentFEM",
+      "serial and two-rank implicit finite-element execution"
+    ],
+    "not_supported": [
+      "production UMAT certification",
+      "experimental material calibration",
+      "cross-material universality",
+      "mathematically exact consistent plastic tangent"
+    ]
+  }
+}

docs/T2_ALL_DATA_STAGE_SUMMARY.md

@@ -53,5 +53,21 @@ DENIM保留J2骨架、关联流动和隐式积分，只学习缺失的硬化与
 4. 后续优先研究长历史稳定性、内部状态监督减少和物理不完备程度阶梯；不继续海选
    大量普通时序网络。
 
+## 数据与模型产品状态
+
+截至2026-09-25，2,660条轨迹的本地质量门和34项T2回归测试全部通过。扩展DENIM
+已转换为带清单、状态定义、适用域和SHA-256的`safetensors`离线模型包，并通过
+AgentFEM `40c9652`与AgentFEM-learning `06d454e`通用学习型本构接口验证。
+
+- 安全模型包与原始实现的应力最大差：`2.68e-7 Pa`；
+- 121步终态PEEQ差：0；
+- 串行与双进程塑性杆：均4/4增量收敛；
+- 串行与双进程最大应力差：`5.96e-8 Pa`；
+- 塑性点自动微分切线相对有限差分差异：约0.05%–0.80%。
+
+切线差异作为当前公开能力边界保留。它没有改变数据、权重或材料点应力精度，但
+限制了“严格一致算法切线”的表述。完整证据见
+`artifacts/t2_denim_agentfem_v1/runtime_validation.json`。
+
 模型与完整机器指标见`artifacts/t2_denim_boundary_v1/model_metrics.json`，对比图见
 `artifacts/t2_denim_boundary_v1/boundary_model_comparison.png`。

docs/T2_DATA_MODEL_HANDOFF.md

@@ -0,0 +1,91 @@
+# AgentFEM material-memory data and DENIM model handoff
+
+## Public products
+
+- Dataset: `HaomingLuo/AgentFEM-Material-Loading-Memory`
+- Model: `HaomingLuo/AgentFEM-DENIM`
+
+The dataset contains 2,660 complete material histories. A sample is always one
+ordered trajectory; time points are not counted as independent samples.
+
+| Release | Trajectories | Primary use |
+|---|---:|---|
+| T2 loading-memory v1 | 1,008 | loading-history baseline and proportional cyclic response |
+| T2 multiaxial OOD v2 | 1,024 | ID, unseen-path and parameter-OOD comparison |
+| DENIM closure v1 | 128 | incomplete internal-state closure |
+| DENIM boundary v1 | 500 | path, amplitude, long-history and resolution boundaries |
+
+These releases form one evidence chain and retain separate protocols. Pooling
+all 2,660 trajectories as exchangeable supervised samples would introduce
+conflicting constitutive targets and invalidate the frozen tests.
+
+## Model product
+
+The recommended checkpoint is `denim-expanded`, version `1.1.0`. It has 918
+trainable parameters and retains a small-strain J2 skeleton with two learned
+memory channels. The reference material uses three memory channels and hidden
+tabulated hardening, so the task measures closure under incomplete material
+state and evolution knowledge.
+
+Published stress RMSE values are:
+
+| Evaluation | DENIM expanded | Incomplete J2 | GRU |
+|---|---:|---:|---:|
+| Frozen held-out paths | 0.714 MPa | 59.541 MPa | 98.260 MPa |
+| New path OOD | 0.743 MPa | 58.177 MPa | 96.959 MPa |
+| Amplitude OOD | 2.294 MPa | 69.315 MPa | 95.896 MPa |
+| Long history | 10.906 MPa | 54.394 MPa | 90.757 MPa |
+
+Long-history state drift is the clearest current boundary. It should remain in
+future comparisons rather than being replaced by additional ordinary-path
+tests.
+
+## Runtime package
+
+The model repository includes an `agentfem_bundle/` directory containing:
+
+- `model.json`: immutable architecture, state, parameter and applicability contract;
+- `weights.safetensors`: runtime weights without pickle deserialization;
+- `SHA256SUMS`: authenticated bundle files;
+- `README.md`: offline-use notes.
+
+The bundle is tied to exact dataset and source-model revisions. AgentFEM owns
+the global Newton process, material-state commit/rollback, checkpointing and
+result evidence; AgentFEM-learning owns the PyTorch provider and DENIM adapter.
+
+## Reproducible evidence
+
+The current package has four evidence levels:
+
+1. dataset integrity and constitutive quality gates;
+2. material-point equivalence between legacy and safe bundles;
+3. path, amplitude, long-history and discretization boundary metrics;
+4. serial and two-rank AgentFEM implicit execution.
+
+The plastic automatic-differentiation tangent is currently approximate to
+about 0.05–0.80% against fixed-old-state finite differences over the audited
+states. The demonstrated global cases converge, but an exact consistent
+plastic tangent remains an open numerical improvement.
+
+## Material for a university research team
+
+The public assets already provide the experimental matrix needed to organize
+a computational study:
+
+- fixed data splits and model identities;
+- black-box, weak-physics, white-box and incomplete-physics comparisons;
+- architecture and parameter-count evidence;
+- OOD and long-history capability boundaries;
+- energy, yield, incompressibility and state diagnostics;
+- finite-element deployment evidence and a documented tangent limitation;
+- complete generation, training and validation code.
+
+The main method proposition available for further theoretical analysis is:
+
+> Learn a transferable, incrementally integrable material-memory closure when
+> the evolution law and internal-state description are incomplete.
+
+Further academic work may analyze identifiability, reduced supervision,
+implicit differentiation, error propagation and constitutive stability. The
+published claims remain limited to the synthetic fixed-material protocols and
+the verified software/runtime versions recorded above.