Paper II Zenodo DOI + sesion 32 v2 records (+9, total 88)

- Paper II citation added: 10.5281/zenodo.19960573
- 9 new records:
- Pythia-410M, Pythia-1.4B, StarCoder2-3B (bf16/4-bit pairs, R2-direction rule)
- Mistral-7B base + Instruct (4-bit, F9 RLHF pair, Δγ=-0.133)
- Qwen2.5-7B base (4-bit, completes Qwen RLHF pair)
- Coverage now: 88 records, 35 models, 13 families
- R²-direction rule panel now n=8 paired (7/8 sign-correct; StarCoder2-3B is the new outlier)

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

README.md

@@ -27,8 +27,10 @@ configs:
 
 > **First public dataset of attention-decay exponent γ measurements
 > across transformer LLMs.**
-> Companion to the paper *Predicting How Transformers Attend*
-> (Marín 2026, [Zenodo DOI 10.5281/zenodo.19826343](https://zenodo.org/records/19826343)).
+> Companion to the *Thermodynamic Attention Framework* (TAF) papers by
+> Carles Marín (2026):
+> - Paper I: [10.5281/zenodo.19826343](https://doi.org/10.5281/zenodo.19826343) — *Predicting How Transformers Attend*
+> - Paper II: [10.5281/zenodo.19960573](https://doi.org/10.5281/zenodo.19960573) — *A Six-Axis Decomposition with the Learned Imprint, Sink-Dominated Precision Boundaries, Bimodal Phase Structure, and Honest Revisions*
 
 ## What it is
 
@@ -49,14 +51,15 @@ where θ is the RoPE base frequency and T is the evaluation context length.
 
 ## Coverage
 
-- **33 models** across 13 families (Pythia, Qwen, Llama, Mistral, Gemma, Phi, OLMo, OLMoE, DeepSeek, StarCoder2, CodeLlama, GPT-J, SmolLM2, Falcon, Yi)
-- **79 records** total
+- **35 models** across 13 families (Pythia, Qwen, Llama, Mistral, Gemma, Phi, OLMo, OLMoE, DeepSeek, StarCoder2, CodeLlama, GPT-J, SmolLM2, Falcon, Yi)
+- **88 records** total
 - **2 corpora**: real text (`real_text`, MongoDB English episodes) + random tokens (`random_tokens`)
 - **2 precisions**: 4-bit NF4 (BitsAndBytes) + bfloat16
 - **Includes random-init controls** (E2 falsifier on Pythia 70M/410M/1B with random Gaussian init, no pretraining) — establishes that the slope ν = ∂γ/∂log₁₀(P) ≈ −1/(2π) is genuinely a *training imprint*, not architecture artifact.
 - **Pythia-70M training trajectory** (9 checkpoints × 2 corpora = 18 records, sesion 32) — within-model γ across `step1000` → `step143000`. **Honest null result**: trajectory does NOT converge to ν = −1/(2π); imprint constant emerges across-models, not within-model.
 - **Pythia-31m high-n robustness** (n=60 prompts × 2 corpora = 2 records) — tightens CI on smallest pythia anchor.
 - **Yi-9B random_tokens** (n=30) — fills 9B class gap in family panel.
+- **R²-direction rule extension** (sesion 32 v2, 2026-05-02): 6 new bf16/4-bit paired measurements (Pythia-410M, Pythia-1.4B, StarCoder2-3B, Mistral-7B base + Instruct, Qwen2.5-7B base). Brings R²-direction rule panel from $n=5$ to $n=8$ paired (7/8 sign-correct; StarCoder2-3B is the new outlier).
 
 ## Schema
 
@@ -160,6 +163,18 @@ This release closes that gap.
   doi       = {10.5281/zenodo.19826343},
   url       = {https://zenodo.org/records/19826343}
 }
+
+@article{marin2026taf2,
+  author    = {Mar{\'\i}n, Carles},
+  title     = {Predicting How Transformers Attend, Part II: A Six-Axis
+               Decomposition with the Learned Imprint $\nu = -1/(2\pi)$,
+               Sink-Dominated Precision Boundaries, Bimodal Phase Structure,
+               and Honest Revisions},
+  year      = {2026},
+  publisher = {Zenodo},
+  doi       = {10.5281/zenodo.19960573},
+  url       = {https://doi.org/10.5281/zenodo.19960573}
+}
 ```
 
 ## Acknowledgements
@@ -193,6 +208,10 @@ GitHub repo — refutations are welcome.
   - Pythia-70M ν trajectory (9 ckpts × 2 corpora = 18) — within-model null documented
   - Yi-9B random_tokens (1) — 9B class gap filled
   - Pythia-31m high-n robustness (2, n=60 each) — tightened CI on smallest anchor
+- 2026-05-02: ★ **Paper II released on Zenodo** (DOI [10.5281/zenodo.19960573](https://doi.org/10.5281/zenodo.19960573)) + 9 new records (88 total):
+  - 3 bf16/4-bit pairs (Pythia-410M, Pythia-1.4B, StarCoder2-3B) — R²-direction rule extension
+  - Mistral-7B base + Instruct (4-bit) — F9 RLHF pair, finds Δγ_RLHF = −0.133
+  - Qwen2.5-7B base (4-bit) — completes the Qwen GQA RLHF pair
 - 2026-05-01: ★ **TAF v0.5 machine-verified consistency** — 15 algebraic identities
   of TAF critical exponents formally proven via dual-tool approach:
   - **Sage Groebner basis** (algebraic decision in PolynomialRing(ℚ))

summary.csv

@@ -78,3 +78,12 @@ EleutherAI/pythia-70m-deduped@step100000,pythia,real_text,auto,1.32261751836488,
 EleutherAI/pythia-70m-deduped@step100000,pythia,random_tokens,auto,1.2204931999100648,,70,10000,2048,False,False
 EleutherAI/pythia-70m-deduped@step143000,pythia,real_text,auto,1.2493746664709637,,70,10000,2048,False,False
 EleutherAI/pythia-70m-deduped@step143000,pythia,random_tokens,auto,1.2023289680609859,,70,10000,2048,False,False
+EleutherAI/pythia-410m,pythia,real_text,auto,1.0218530106365162,0.99999561666838,410,10000,2048,False,False
+EleutherAI/pythia-410m,pythia,real_text,4-bit-NF4,1.0262501725670703,0.99999561666838,410,10000,2048,False,False
+EleutherAI/pythia-1.4b,pythia,real_text,auto,0.83289308127506,0.99999561666838,1400,10000,2048,False,False
+EleutherAI/pythia-1.4b,pythia,real_text,4-bit-NF4,0.8327854153348206,0.99999561666838,1400,10000,2048,False,False
+bigcode/starcoder2-3b,starcoder2,real_text,auto,1.006513487201348,0.6666637444455867,3000,999999.4427,16384,False,False
+bigcode/starcoder2-3b,starcoder2,real_text,4-bit-NF4,0.9941659863431367,0.6666637444455867,3000,999999.4427,16384,False,False
+mistralai/Mistral-7B-v0.1,mistral,real_text,4-bit-NF4,1.060750419523944,0.99999561666838,7000,10000,32768,False,False
+mistralai/Mistral-7B-Instruct-v0.1,mistral,real_text,4-bit-NF4,0.9281400361369939,,7000,10000,32768,True,False
+Qwen/Qwen2.5-7B,qwen,real_text,4-bit-NF4,0.996153147006284,0.6666637444455867,7000,1000000,131072,False,False

taf-attention-decay.jsonl

@@ -77,3 +77,12 @@
 {"model_id": "EleutherAI/pythia-70m-deduped", "revision": "step100000", "arch": {"d_model": 512, "n_heads": 8, "n_layers": 6, "d_head": 64, "n_kv_heads": 8, "n_params_M": 70, "rope_theta": 10000, "T_train": 2048, "family": "pythia", "is_instruct": false, "is_moe": false}, "measurement": {"gamma": 1.2204931999100648, "gamma_ci95_lo": 1.0522202974778878, "gamma_ci95_hi": 1.3232134560235709, "method": "pade_d_alias_T", "fit": {"log_A": -0.8927465051706067, "R2": 0.992892, "n_points": 5, "delta_R2_power_minus_exp": 0.1325}, "T_eval": 2048, "corpus": "random_tokens", "n_prompts_per_distance": 30, "seeds": [42, 123, 7], "distances": [10, 50, 100, 250, 500, 1000], "device": "cuda", "dtype": "auto"}, "framework_prediction": {"gamma_pade": null, "gamma_random_pred": null, "imprint_constant_nu": -0.1592}, "decision": "unknown", "provenance": {"taf_version": "0.5", "paper_doi": "10.5281/zenodo.19826343", "source_file": "EleutherAI--pythia-70m-deduped_random_step100000.json", "tool": "tafagent/cli/diagnose_model.py + e4_extended_gamma.py", "license_data": "CC-BY-4.0", "license_code": "Apache-2.0", "session": "32 (2026-05-01) — paper 2 strengthening"}}
 {"model_id": "EleutherAI/pythia-70m-deduped", "revision": "step143000", "arch": {"d_model": 512, "n_heads": 8, "n_layers": 6, "d_head": 64, "n_kv_heads": 8, "n_params_M": 70, "rope_theta": 10000, "T_train": 2048, "family": "pythia", "is_instruct": false, "is_moe": false}, "measurement": {"gamma": 1.2493746664709637, "gamma_ci95_lo": 0.9036724881431525, "gamma_ci95_hi": 1.5802558154712143, "method": "pade_d_alias_T", "fit": {"log_A": -0.7806191250583789, "R2": 0.971774, "n_points": 5, "delta_R2_power_minus_exp": 0.1713}, "T_eval": 2048, "corpus": "real_text", "n_prompts_per_distance": 30, "seeds": [42, 123, 7], "distances": [10, 50, 100, 250, 500, 1000], "device": "cuda", "dtype": "auto"}, "framework_prediction": {"gamma_pade": null, "gamma_random_pred": null, "imprint_constant_nu": -0.1592}, "decision": "unknown", "provenance": {"taf_version": "0.5", "paper_doi": "10.5281/zenodo.19826343", "source_file": "EleutherAI--pythia-70m-deduped_mongo_step143000.json", "tool": "tafagent/cli/diagnose_model.py + e4_extended_gamma.py", "license_data": "CC-BY-4.0", "license_code": "Apache-2.0", "session": "32 (2026-05-01) — paper 2 strengthening"}}
 {"model_id": "EleutherAI/pythia-70m-deduped", "revision": "step143000", "arch": {"d_model": 512, "n_heads": 8, "n_layers": 6, "d_head": 64, "n_kv_heads": 8, "n_params_M": 70, "rope_theta": 10000, "T_train": 2048, "family": "pythia", "is_instruct": false, "is_moe": false}, "measurement": {"gamma": 1.2023289680609859, "gamma_ci95_lo": 0.9651031952002092, "gamma_ci95_hi": 1.4205245025802689, "method": "pade_d_alias_T", "fit": {"log_A": -0.9227782529370149, "R2": 0.986676, "n_points": 5, "delta_R2_power_minus_exp": 0.0937}, "T_eval": 2048, "corpus": "random_tokens", "n_prompts_per_distance": 30, "seeds": [42, 123, 7], "distances": [10, 50, 100, 250, 500, 1000], "device": "cuda", "dtype": "auto"}, "framework_prediction": {"gamma_pade": null, "gamma_random_pred": null, "imprint_constant_nu": -0.1592}, "decision": "unknown", "provenance": {"taf_version": "0.5", "paper_doi": "10.5281/zenodo.19826343", "source_file": "EleutherAI--pythia-70m-deduped_random_step143000.json", "tool": "tafagent/cli/diagnose_model.py + e4_extended_gamma.py", "license_data": "CC-BY-4.0", "license_code": "Apache-2.0", "session": "32 (2026-05-01) — paper 2 strengthening"}}
+{"model_id": "EleutherAI/pythia-410m", "revision": "main", "arch": {"d_model": 1024, "n_heads": 16, "n_layers": 24, "d_head": 64, "n_kv_heads": 16, "n_params_M": 410, "rope_theta": 10000, "T_train": 2048, "family": "pythia", "is_instruct": false, "is_moe": false}, "measurement": {"gamma": 1.0218530106365162, "gamma_ci95_lo": 0.846278741222938, "gamma_ci95_hi": 1.1560602030295848, "method": "pade_d_alias_T", "fit": {"log_A": -1.7669627940483377, "R2": 0.981594, "n_points": 7, "delta_R2_power_minus_exp": 0.1034}, "T_eval": 2048, "corpus": "real_text", "n_prompts_per_distance": 150, "seeds": [42, 123, 7], "distances": [10, 20, 30, 50, 100, 200, 500, 1000, 2000], "device": "cuda", "dtype": "auto"}, "framework_prediction": {"gamma_pade": 0.99999561666838, "gamma_random_pred": null, "imprint_constant_nu": -0.1592}, "decision": "CONFIRMED: γ law holds (γ×ln(θ) = C)", "provenance": {"taf_version": "0.5", "paper_doi": "10.5281/zenodo.19826343", "source_file": "EleutherAI--pythia-410m_mongo.json", "tool": "tafagent/cli/diagnose_model.py + e4_extended_gamma.py", "license_data": "CC-BY-4.0", "license_code": "Apache-2.0", "session": "32 v2 (2026-05-02) — R2-direction rule extension + F9 Mistral pair"}}
+{"model_id": "EleutherAI/pythia-410m", "revision": "main", "arch": {"d_model": 1024, "n_heads": 16, "n_layers": 24, "d_head": 64, "n_kv_heads": 16, "n_params_M": 410, "rope_theta": 10000, "T_train": 2048, "family": "pythia", "is_instruct": false, "is_moe": false}, "measurement": {"gamma": 1.0262501725670703, "gamma_ci95_lo": 0.8390736556802114, "gamma_ci95_hi": 1.1677383264955195, "method": "pade_d_alias_T", "fit": {"log_A": -1.76649015479223, "R2": 0.979366, "n_points": 7, "delta_R2_power_minus_exp": 0.0957}, "T_eval": 2048, "corpus": "real_text", "n_prompts_per_distance": 150, "seeds": [42, 123, 7], "distances": [10, 20, 30, 50, 100, 200, 500, 1000, 2000], "device": "cuda", "dtype": "4-bit-NF4"}, "framework_prediction": {"gamma_pade": 0.99999561666838, "gamma_random_pred": null, "imprint_constant_nu": -0.1592}, "decision": "CONFIRMED: γ law holds (γ×ln(θ) = C)", "provenance": {"taf_version": "0.5", "paper_doi": "10.5281/zenodo.19826343", "source_file": "EleutherAI--pythia-410m_mongo.json", "tool": "tafagent/cli/diagnose_model.py + e4_extended_gamma.py", "license_data": "CC-BY-4.0", "license_code": "Apache-2.0", "session": "32 v2 (2026-05-02) — R2-direction rule extension + F9 Mistral pair"}}
+{"model_id": "EleutherAI/pythia-1.4b", "revision": "main", "arch": {"d_model": 2048, "n_heads": 16, "n_layers": 24, "d_head": 128, "n_kv_heads": 16, "n_params_M": 1400, "rope_theta": 10000, "T_train": 2048, "family": "pythia", "is_instruct": false, "is_moe": false}, "measurement": {"gamma": 0.83289308127506, "gamma_ci95_lo": 0.7526289848600018, "gamma_ci95_hi": 1.0319364818957766, "method": "pade_d_alias_T", "fit": {"log_A": -2.772503095759906, "R2": 0.961413, "n_points": 7, "delta_R2_power_minus_exp": 0.3162}, "T_eval": 2048, "corpus": "real_text", "n_prompts_per_distance": 150, "seeds": [42, 123, 7], "distances": [10, 20, 30, 50, 100, 200, 500, 1000, 2000], "device": "cuda", "dtype": "auto"}, "framework_prediction": {"gamma_pade": 0.99999561666838, "gamma_random_pred": null, "imprint_constant_nu": -0.1592}, "decision": "UNCLEAR: γ=0.833 outside all expected ranges", "provenance": {"taf_version": "0.5", "paper_doi": "10.5281/zenodo.19826343", "source_file": "EleutherAI--pythia-1.4b_mongo.json", "tool": "tafagent/cli/diagnose_model.py + e4_extended_gamma.py", "license_data": "CC-BY-4.0", "license_code": "Apache-2.0", "session": "32 v2 (2026-05-02) — R2-direction rule extension + F9 Mistral pair"}}
+{"model_id": "EleutherAI/pythia-1.4b", "revision": "main", "arch": {"d_model": 2048, "n_heads": 16, "n_layers": 24, "d_head": 128, "n_kv_heads": 16, "n_params_M": 1400, "rope_theta": 10000, "T_train": 2048, "family": "pythia", "is_instruct": false, "is_moe": false}, "measurement": {"gamma": 0.8327854153348206, "gamma_ci95_lo": 0.749181092015435, "gamma_ci95_hi": 1.0323756108077562, "method": "pade_d_alias_T", "fit": {"log_A": -2.808517690987601, "R2": 0.960511, "n_points": 7, "delta_R2_power_minus_exp": 0.3163}, "T_eval": 2048, "corpus": "real_text", "n_prompts_per_distance": 150, "seeds": [42, 123, 7], "distances": [10, 20, 30, 50, 100, 200, 500, 1000, 2000], "device": "cuda", "dtype": "4-bit-NF4"}, "framework_prediction": {"gamma_pade": 0.99999561666838, "gamma_random_pred": null, "imprint_constant_nu": -0.1592}, "decision": "UNCLEAR: γ=0.833 outside all expected ranges", "provenance": {"taf_version": "0.5", "paper_doi": "10.5281/zenodo.19826343", "source_file": "EleutherAI--pythia-1.4b_mongo.json", "tool": "tafagent/cli/diagnose_model.py + e4_extended_gamma.py", "license_data": "CC-BY-4.0", "license_code": "Apache-2.0", "session": "32 v2 (2026-05-02) — R2-direction rule extension + F9 Mistral pair"}}
+{"model_id": "bigcode/starcoder2-3b", "revision": "main", "arch": {"d_model": 3072, "n_heads": 24, "n_layers": 30, "d_head": 128, "n_kv_heads": 2, "n_params_M": 3000, "rope_theta": 999999.4427, "T_train": 16384, "family": "starcoder2", "is_instruct": false, "is_moe": false}, "measurement": {"gamma": 1.006513487201348, "gamma_ci95_lo": 0.9236017157491303, "gamma_ci95_hi": 1.0791762679565378, "method": "pade_d_alias_T", "fit": {"log_A": -2.218862577729128, "R2": 0.993182, "n_points": 7, "delta_R2_power_minus_exp": 0.1663}, "T_eval": 16384, "corpus": "real_text", "n_prompts_per_distance": 150, "seeds": [42, 123, 7], "distances": [10, 20, 30, 50, 100, 200, 500, 1000, 2000], "device": "cuda", "dtype": "auto"}, "framework_prediction": {"gamma_pade": 0.6666637444455867, "gamma_random_pred": null, "imprint_constant_nu": -0.1592}, "decision": "REFUTED: C not constant across θ", "provenance": {"taf_version": "0.5", "paper_doi": "10.5281/zenodo.19826343", "source_file": "bigcode--starcoder2-3b_mongo.json", "tool": "tafagent/cli/diagnose_model.py + e4_extended_gamma.py", "license_data": "CC-BY-4.0", "license_code": "Apache-2.0", "session": "32 v2 (2026-05-02) — R2-direction rule extension + F9 Mistral pair"}}
+{"model_id": "bigcode/starcoder2-3b", "revision": "main", "arch": {"d_model": 3072, "n_heads": 24, "n_layers": 30, "d_head": 128, "n_kv_heads": 2, "n_params_M": 3000, "rope_theta": 999999.4427, "T_train": 16384, "family": "starcoder2", "is_instruct": false, "is_moe": false}, "measurement": {"gamma": 0.9941659863431367, "gamma_ci95_lo": 0.9138927037900326, "gamma_ci95_hi": 1.0602601079660066, "method": "pade_d_alias_T", "fit": {"log_A": -2.2543693248549825, "R2": 0.993781, "n_points": 7, "delta_R2_power_minus_exp": 0.1738}, "T_eval": 16384, "corpus": "real_text", "n_prompts_per_distance": 150, "seeds": [42, 123, 7], "distances": [10, 20, 30, 50, 100, 200, 500, 1000, 2000], "device": "cuda", "dtype": "4-bit-NF4"}, "framework_prediction": {"gamma_pade": 0.6666637444455867, "gamma_random_pred": null, "imprint_constant_nu": -0.1592}, "decision": "REFUTED: C not constant across θ", "provenance": {"taf_version": "0.5", "paper_doi": "10.5281/zenodo.19826343", "source_file": "bigcode--starcoder2-3b_mongo.json", "tool": "tafagent/cli/diagnose_model.py + e4_extended_gamma.py", "license_data": "CC-BY-4.0", "license_code": "Apache-2.0", "session": "32 v2 (2026-05-02) — R2-direction rule extension + F9 Mistral pair"}}
+{"model_id": "mistralai/Mistral-7B-v0.1", "revision": "main", "arch": {"d_model": 4096, "n_heads": 32, "n_layers": 32, "d_head": 128, "n_kv_heads": 8, "n_params_M": 7000, "rope_theta": 10000, "T_train": 32768, "family": "mistral", "is_instruct": false, "is_moe": false}, "measurement": {"gamma": 1.060750419523944, "gamma_ci95_lo": 1.0287296895523124, "gamma_ci95_hi": 1.0879595935791004, "method": "pade_d_alias_T", "fit": {"log_A": -2.143867119472637, "R2": 0.99869, "n_points": 7, "delta_R2_power_minus_exp": 0.2041}, "T_eval": 32768, "corpus": "real_text", "n_prompts_per_distance": 150, "seeds": [42, 123, 7], "distances": [10, 20, 30, 50, 100, 200, 500, 1000, 2000], "device": "cuda", "dtype": "4-bit-NF4"}, "framework_prediction": {"gamma_pade": 0.99999561666838, "gamma_random_pred": null, "imprint_constant_nu": -0.1592}, "decision": "CONFIRMED: γ law holds (γ×ln(θ) = C)", "provenance": {"taf_version": "0.5", "paper_doi": "10.5281/zenodo.19826343", "source_file": "mistralai--Mistral-7B-v0.1_mongo.json", "tool": "tafagent/cli/diagnose_model.py + e4_extended_gamma.py", "license_data": "CC-BY-4.0", "license_code": "Apache-2.0", "session": "32 v2 (2026-05-02) — R2-direction rule extension + F9 Mistral pair"}}
+{"model_id": "mistralai/Mistral-7B-Instruct-v0.1", "revision": "main", "arch": {"d_model": 4096, "n_heads": 32, "n_layers": 32, "d_head": 128, "n_kv_heads": 8, "n_params_M": 7000, "rope_theta": 10000, "T_train": 32768, "family": "mistral", "is_instruct": true, "is_moe": false}, "measurement": {"gamma": 0.9281400361369939, "gamma_ci95_lo": 0.8979110676787743, "gamma_ci95_hi": 0.948413427838792, "method": "pade_d_alias_T", "fit": {"log_A": -2.6016818600864173, "R2": 0.998813, "n_points": 7, "delta_R2_power_minus_exp": 0.2135}, "T_eval": 32768, "corpus": "real_text", "n_prompts_per_distance": 150, "seeds": [42, 123, 7], "distances": [10, 20, 30, 50, 100, 200, 500, 1000, 2000], "device": "cuda", "dtype": "4-bit-NF4"}, "framework_prediction": {"gamma_pade": null, "gamma_random_pred": null, "imprint_constant_nu": -0.1592}, "decision": "unknown", "provenance": {"taf_version": "0.5", "paper_doi": "10.5281/zenodo.19826343", "source_file": "mistralai--Mistral-7B-Instruct-v0.1_mongo.json", "tool": "tafagent/cli/diagnose_model.py + e4_extended_gamma.py", "license_data": "CC-BY-4.0", "license_code": "Apache-2.0", "session": "32 v2 (2026-05-02) — R2-direction rule extension + F9 Mistral pair"}}
+{"model_id": "Qwen/Qwen2.5-7B", "revision": "main", "arch": {"d_model": 3584, "n_heads": 28, "n_layers": 28, "d_head": 128, "n_kv_heads": 4, "n_params_M": 7000, "rope_theta": 1000000, "T_train": 131072, "family": "qwen", "is_instruct": false, "is_moe": false}, "measurement": {"gamma": 0.996153147006284, "gamma_ci95_lo": 0.9218257753296215, "gamma_ci95_hi": 1.0480334312300015, "method": "pade_d_alias_T", "fit": {"log_A": -2.160172020244544, "R2": 0.994921, "n_points": 7, "delta_R2_power_minus_exp": 0.2114}, "T_eval": 131072, "corpus": "real_text", "n_prompts_per_distance": 150, "seeds": [42, 123, 7], "distances": [10, 20, 30, 50, 100, 200, 500, 1000, 2000], "device": "cuda", "dtype": "4-bit-NF4"}, "framework_prediction": {"gamma_pade": 0.6666637444455867, "gamma_random_pred": null, "imprint_constant_nu": -0.1592}, "decision": "REFUTED: C not constant across θ", "provenance": {"taf_version": "0.5", "paper_doi": "10.5281/zenodo.19826343", "source_file": "Qwen--Qwen2.5-7B_mongo.json", "tool": "tafagent/cli/diagnose_model.py + e4_extended_gamma.py", "license_data": "CC-BY-4.0", "license_code": "Apache-2.0", "session": "32 v2 (2026-05-02) — R2-direction rule extension + F9 Mistral pair"}}