2026-07-12_r19_dev_calibration.json

{ "schema_version": "oczy/execution-report/v1", "record_type": "r19_dev_calibration_adjudication", "adjudication_date": "2026-07-12", "research_spec": "research/19-lm-as-language-organ.md", "source_repo": "https://github.com/KinoThe-Kafkaesque/oczy.git", "provider": "kaggle", "profile": "cpu", "module": "oczy.experiments.s19_language_organ", "command": ["/usr/bin/python3", "-m", "oczy.experiments.s19_language_organ", "calibrate-dev", "--model-id", "Qwen/Qwen2.5-0.5B-Instruct", "--manifest-out", "/kaggle/working/calibration_manifest.json", "--coupler-out", "/kaggle/working/s19_cortex.pkl", "--source-commit", "bd1ead9a8358b675af5e929c53a01eb505839639", "--source-archive-sha256", "1afe7573438e18a66ac6b23806978fe7662d3cf9d1662e29200da73729bce3eb", "--lr", "0.01", "--coupler-lr", "0.001", "--label-lr", "0.05"], "status": "complete", "exit_code": 0, "error": null, "started_utc": "2026-07-12T11:19:35Z", "finished_utc": "2026-07-12T11:31:46Z", "timeout_seconds": null,

"immutable_hashes": { "source_commit": "bd1ead9a8358b675af5e929c53a01eb505839639", "source_archive_sha256": "1afe7573438e18a66ac6b23806978fe7662d3cf9d1662e29200da73729bce3eb", "source_archive_size_bytes": 1800978, "source_archive_tracked_file_count": 473, "source_dataset": "abdellahkadem/oczy-source-bd1ead9a8358", "source_dirty_entry_count": 0, "source_worktree_dirty_at_packaging": false, "eval_version": "v2.2", "eval_manifest_sha256": "982116463ff94315315e4c4a3c21abc85411d03f78286db710e2a00a25a3c843", "model_repo_id": "Qwen/Qwen2.5-0.5B-Instruct", "model_revision": "main", "model_config_sha256": "f49f217d6c21431063185bac5c274e224afcb38a15b3607cc9532be3a1d34d61", "model_safetensors_sha256": "c6ac65d183d505da320789d9132a14d6bed47a5c8b25622dab07ad3aad1a8fc6", "model_params_requires_grad": false, "manifest_sha256": "77ef4607ff95c116b5b7b088a7f5cfa811b855d76feed9c329eb551ac586a1e2" },

"hardware": { "platform": "Linux-6.12.90+-x86_64-with-glibc2.35", "python": "3.12.13", "torch": "2.10.0+cpu", "cuda_available": false, "cuda_device_count": 0, "torch_cuda_version": null },

"cpu_only_contract": { "cuda_available": false, "cuda_visible_devices": "", "hf_hub_offline": "1", "mkl_num_threads": "4", "oczy_remote_cpu_only": "1", "omp_num_threads": "4", "profile": "cpu", "transformers_offline": "1" },

"model_artifact": { "hf_model_dir": "/kaggle/input/models/qwen-lm/qwen2.5/transformers/0.5b-instruct/1", "gguf_model_path": null, "model_source": "qwen-lm/qwen2.5/transformers/0.5b-instruct/1" },

"calibration_attempts": [ { "attempt": 1, "job_name": "r19-calibrate-dev", "source_commit": "0d628118083e7df5c073e6255c297884018ed8fe", "source_archive_sha256": "198c6916eb39d23d2963a11f0b7dabc6473b267c397ea5b45612df0132dfafb9", "source_dataset": "abdellahkadem/oczy-source-0d628118083e", "kernel_id": "abdellahkadem/oczy-r19-calibrate-0d628118083e", "remote_id": "abdellahkadem/oczy-r19-calibrate-0d628118083e", "state": "failed", "error": "kernel reported error status", "exit_code": null, "submitted_utc": "2026-07-12T00:48:06Z", "completed_utc": "2026-07-12T00:49:07Z", "collected_at": null, "reports_collected": false, "failure_classification": "infrastructure", "failure_detail": "offline model resolution failure", "manifest_out_arg": "experiments_logs/r19_calibrate_dev/calibration_manifest.json", "coupler_out_arg": "experiments_logs/r19_calibrate_dev/s19_cortex.pkl" }, { "attempt": 2, "job_name": "r19-calibrate-dev-v2", "source_commit": "4b73780935ef45dc44fd5d75709ca18d4638a6b0", "source_archive_sha256": "c7d6f08e214aac3fba7e889d79c525536e4a0fe3fefa1458cffe94a38a1cc87b", "source_dataset": "abdellahkadem/oczy-source-4b73780935ef", "kernel_id": "abdellahkadem/oczy-r19-cal-v2-4b73780935ef", "remote_id": "abdellahkadem/oczy-r19-cal-v2-4b73780935ef", "state": "failed", "error": "kernel reported error status", "exit_code": null, "submitted_utc": "2026-07-12T10:39:41Z", "completed_utc": "2026-07-12T10:50:28Z", "collected_at": null, "reports_collected": false, "failure_classification": "infrastructure", "failure_detail": "source-path/provenance failure plus feature explosion", "manifest_out_arg": "experiments_logs/r19_calibrate_dev/calibration_manifest.json", "coupler_out_arg": "experiments_logs/r19_calibrate_dev/s19_cortex.pkl" }, { "attempt": 3, "job_name": "r19-calibrate-dev-v3", "source_commit": "bd1ead9a8358b675af5e929c53a01eb505839639", "source_archive_sha256": "1afe7573438e18a66ac6b23806978fe7662d3cf9d1662e29200da73729bce3eb", "source_dataset": "abdellahkadem/oczy-source-bd1ead9a8358", "kernel_id": "abdellahkadem/oczy-r19-cal-v3-bd1ead9a8358", "remote_id": "abdellahkadem/oczy-r19-cal-v3-bd1ead9a8358", "state": "succeeded", "error": null, "exit_code": 0, "submitted_utc": "2026-07-12T11:05:03Z", "completed_utc": "2026-07-12T11:15:47Z", "collected_at": "2026-07-12T11:15:48Z", "started_utc": "2026-07-12T11:05:17Z", "finished_utc": "2026-07-12T11:15:24Z", "reports_collected": true, "failure_classification": "infrastructure", "failure_detail": "execution succeeded but artifacts written to relative path experiments_logs/r19_calibrate_dev/ instead of /kaggle/working/ — not rooted in /kaggle/working, superseded by v4", "manifest_out_arg": "experiments_logs/r19_calibrate_dev/calibration_manifest.json", "coupler_out_arg": "experiments_logs/r19_calibrate_dev/s19_cortex.pkl" }, { "attempt": 4, "job_name": "r19-calibrate-dev-v4", "source_commit": "bd1ead9a8358b675af5e929c53a01eb505839639", "source_archive_sha256": "1afe7573438e18a66ac6b23806978fe7662d3cf9d1662e29200da73729bce3eb", "source_dataset": "abdellahkadem/oczy-source-bd1ead9a8358", "kernel_id": "abdellahkadem/oczy-r19-cal-v4-bd1ead9a8358", "remote_id": "abdellahkadem/oczy-r19-cal-v4-bd1ead9a8358", "state": "succeeded", "error": null, "exit_code": 0, "submitted_utc": "2026-07-12T11:19:21Z", "completed_utc": "2026-07-12T11:32:08Z", "collected_at": "2026-07-12T11:32:12Z", "started_utc": "2026-07-12T11:19:35Z", "finished_utc": "2026-07-12T11:31:46Z", "reports_collected": true, "failure_classification": null, "failure_detail": null, "manifest_out_arg": "/kaggle/working/calibration_manifest.json", "coupler_out_arg": "/kaggle/working/s19_cortex.pkl" } ],

"v4_manifest_metrics": { "schema_version": "oczy/r19-calibration-manifest/v1", "created_at": "2026-07-12T11:31:42.568622+00:00", "d_embd": 896, "d_cortex": 16, "latent_tokens": 3, "max_labels": 20, "arm_b_input_mode": "inputs_embeds", "fixed_latent_shape": [3, 896], "parameter_total": 60388, "parameter_budget": 64000, "parameter_breakdown": { "W_perceive": 14336, "W_label": 320, "b_label": 20, "W_coupler": 43008, "b_coupler": 2688, "warm_state": 16 }, "label_phrasing_frozen": true, "labels": [ "captain's journal", "submit officially", "map legend", "biology cell", "music record", "tree branch", "fashion model", "river run", "military draft", "football pitch", "chocolate bar", "musical note", "fish scale", "dental bridge", "port wine", "crane bird", "fruit jam", "harbor seal", "spy mole", "body organ" ], "dev_split": "dev", "dev_repeatability_std": 0.0, "dev_confidence_mean": 0.05254819590066161, "dev_confidence_std": 0.00028932323902295125, "dev_confidence_min": 0.05206942558288574, "dev_confidence_max": 0.05289289355278015, "dev_specificity_acc": 0.13432835820895522, "dev_holdout_ids_discarded": true, "split_salt": "v2.2", "split_fraction": 0.3, "proposed_confidence_threshold": 0.05224296171218157, "proposed_specificity_margin": 0.15, "n_dev_probes": 14, "n_labels": 20, "calibrate_coupler_loss_mean": 4.333677, "calibrate_label_loss_mean": 3.012896, "c7_reference": "S3.M2a nearest-neighbor retrieval baseline", "c7_available": true, "c7_blocked_reason": null, "trace_raw_traces_deleted": true, "trace_raw_trace_count": 0, "trace_raw_traces_deleted_count": 14, "trace_embedding_cache_cleared": true, "trace_optimizer_state_deleted": true },

"v4_artifacts": { "cortex_artifact_path": "s19_cortex.pkl", "cortex_artifact_sha256": "45d518f4f91ba7e28502cdbc42c2b322035642f74bd46e70fb28f51b6772afed", "cortex_artifact_bytes": 243422, "coupler_sha256": "4632f4c03714a6b85d79ad2db0056e1e2e3e154b0b7dc27b2696d43793471d66", "coupler_bytes": 183493, "head_sha256": "d42f5f26896d4c168f5b0174ff41827febcf83d5eaa7d0c80c3ff72c63996c44", "head_bytes": 60044, "collected_report_dir": "/tmp/oczy-live-queue/reports/r19-calibrate-dev-v4" },

"articulation_checks": { "articulation_prompt_text": "Show the log.", "articulation_latent_bank_shape": [3, 896], "articulation_raw_trace_count": 0, "articulation_language_organ_hash": "c6ac65d183d505da320789d9132a14d6bed47a5c8b25622dab07ad3aad1a8fc6", "articulation_persistent_cortex_bytes": 243422, "articulation_banned_label_text_absent": true, "articulation_banned_corrected_response_absent": true, "articulation_banned_correction_utterance_absent": true, "articulation_banned_expected_answer_absent": true },

"gate_outcome": { "signoff_thresholds_signed_off": false, "signoff_human_signoff_id": "", "signoff_oracle_ceiling": 0.35714285714285715, "signoff_dev_articulation_gate": false, "signoff_meta_test_conflation_ok": true, "holdout_accessed": false, "classification": "BLOCKED", "classification_detail": "BLOCKED at pre-registered DEV articulation gate (signoff_dev_articulation_gate=false). The v4 calibration infrastructure succeeded (exit 0, artifacts collected), but the scientific DEV gate is not passed." },

"adjudication": { "classification": "BLOCKED", "classification_detail": "BLOCKED at pre-registered DEV articulation gate. signoff_dev_articulation_gate=false. No scientific verdict beyond BLOCKED is permitted.", "infrastructure_vs_scientific_separation": "Attempts v1 and v2 failed at the infrastructure level (kernel error status, no reports collected). Attempt v3 succeeded at execution (exit 0) but failed at the artifact-collection level (outputs written to a relative path instead of /kaggle/working, superseded by v4). Attempt v4 succeeded at both execution and collection (exit 0, artifacts rooted in /kaggle/working, reports collected). The scientific DEV articulation gate is independent of infrastructure success: v4 infrastructure is fully successful, but signoff_dev_articulation_gate=false means the pre-registered DEV gate is not passed. No signoff request was made and no holdout access was attempted.", "signoff_decision": "No signoff request was made because signoff_dev_articulation_gate=false. The pre-registered DEV articulation gate must pass before any signoff request or holdout access.", "holdout_decision": "holdout_accessed=false. No holdout access was attempted because the DEV articulation gate has not passed. R20 remains separately blocked for lack of explicit human signoff.", "r20_status": "R20 remains separately blocked for lack of explicit human signoff." },

"c7_adapter_status": { "c7_reference": "S3.M2a nearest-neighbor retrieval baseline", "c7_available_in_manifest": true, "c7_adapter_exists": false, "c7_adapter_function": "_try_s3m2a_retrieval_adapter", "c7_adapter_returns": null, "c7_discrepancy": "The calibrate-dev phase hardcodes c7_available=true in the manifest, but _try_s3m2a_retrieval_adapter() returns None (no real adapter exists — src/oczy/experiments/s19_language_organ_core.py:1357-1359). The evaluate phase would block on C7 independently of the articulation gate. This discrepancy requires diagnosis before any new claim run.", "c7_code_location": "src/oczy/experiments/s19_language_organ_core.py:1357-1359" },

"direction_reassessment": { "budget_constraint": "Do not spend signed-eval or R20 budget.", "diagnose_before_new_claim_run": [ "Articulation/interface: why does the learned coupler not improve over the no-update baseline on DEV? Is the bottleneck the coupler learning signal, the latent interface width/shape, or the articulation path (soft embeddings vs KV entries)?", "C7 adapter availability: resolve the discrepancy between the manifest c7_available=true and _try_s3m2a_retrieval_adapter() returning None. A real S3.M2a adapter must exist before the evaluate phase can run C7 as an external bar." ], "r19_r20_signoff_separation": "R19 signed evaluation is BLOCKED at the DEV articulation gate. No signoff was requested and no holdout was accessed. R20 (meta-trained cortex) remains separately blocked for lack of explicit human signoff. R19 signoff and R20 signoff are distinct: neither has been requested or granted. The R19 articulation gate failure does not change R20's blocked status." },

"explicit_non_claims": [ "No positive scientific verdict is claimed. The classification is BLOCKED at the pre-registered DEV articulation gate.", "No signoff request was made because signoff_dev_articulation_gate=false.", "No holdout access was attempted (holdout_accessed=false).", "R20 remains separately blocked for lack of explicit human signoff.", "No threshold, metric, baseline, episode, scoring, eval manifest, or research spec was changed.", "The v4 infrastructure success (exit 0, artifacts collected) is distinct from the scientific DEV gate outcome (BLOCKED).", "Attempts v1 and v2 are infrastructure failures, not scientific failures — no manifest was produced.", "Attempt v3 is an infrastructure collection failure, not a scientific failure — execution succeeded but artifacts were not rooted in /kaggle/working.", "The dev_confidence_mean (0.0525482) and dev_specificity_acc (0.134328) are recorded as observed DEV metrics, not as passed thresholds.", "The oracle_ceiling (0.357143) is an upper bound on DEV performance with the frozen language organ, not a claimed result.", "The proposed_confidence_threshold and proposed_specificity_margin are calibration proposals, not accepted thresholds.", "The C7 adapter discrepancy (manifest c7_available=true vs _try_s3m2a_retrieval_adapter() returns None) is a secondary blocker that must be resolved before any evaluate-phase run, but the primary blocker is the DEV articulation gate.", "R19 signoff and R20 signoff are distinct. R19 is blocked at the DEV articulation gate; R20 is blocked on human signoff. Neither has been requested or granted.", "The oracle ceiling (0.357143 > 0) passing does not imply the articulation gate passes — they are independent gates.", "No signed-eval or R20 budget should be spent until the articulation/interface and C7 adapter availability are diagnosed at DEV level." ],

"scope_limit": "R19 DEV calibration from source commit bd1ead9a8358b675af5e929c53a01eb505839639 on Kaggle CPU (Qwen/Qwen2.5-0.5B-Instruct, frozen). Four calibration attempts: v1 failed (offline model resolution), v2 failed (source-path/provenance failure plus feature explosion), v3 succeeded but artifacts not rooted in /kaggle/working, v4 succeeded and collected. DEV-only: holdout_accessed=false. No threshold, metric, baseline, episode, scoring, eval manifest, or research spec was changed. The pre-registered DEV articulation gate (signoff_dev_articulation_gate) is unchanged and remains false. No signoff request was made. No scientific verdict beyond BLOCKED is permitted. R20 remains separately blocked for lack of explicit human signoff." }